Most native and deuterated standards for analysis of oxylipins, endocannabinoids, and endocannabinoid-related lipids, including 12-[[(cyclohexylamino)carbonyl]amino]-dodecanoic acid (CUDA) used for recovery calculation purposes, were obtained from Cayman Chemical (Ann Arbor, MI, USA). Only 9,10,13-trihydroxyoctadecenoic acid (TriHOME) and 9,12,13-TriHOME were obtained from Larodan (Malmö, Sweden). Acetonitrile (ACN) and methanol (MeOH) were from Merck (Darmstadt, Germany). Isopropanol was from VWR PROLABO (Fontenay-sous-Bois, France). Acetic acid was purchased from Aldrich Chemical Company, Inc. (Milwaukee, WI, USA). Butylhydroxytoluene (BHT) was from Cayman Chemical (Ann Arbor, MI, USA) and ethylenediaminetetraacetic acid (EDTA) from Fluka Analytical, Sigma-Aldrich (Buchs, Switzerland). Glycerol was from Fischer Scientific (Loughborough, UK). All solvents and chemicals were of HPLC grade or higher and ultrapure water was used (Milli-Q Gradient system, Millipore, Milford, MA, USA).

The study design was based on a series of preceding studies, as previously outlined [18]. In short, the subjects were healthy never smokers and underwent routine health examination including respiratory function. Fifteen individuals with the following characteristics were included in the crossover study design: 8 males, 7 females (mean age 26 years, range 19–34 years); mean body mass index 22 kg/m² (range 20–26 kg/m²). Each subject was instructed to have a light, ordinary breakfast, to avoid ham, and to eat as similarly as possible before the two exposures. Subjects were also instructed to refrain from alcohol and caffeine for 24 h pre-exposure, and to abstain from use of anti-inflammatory drugs or dietary supplements during the week preceding the exposure and bronchoscopy.

Each subject underwent two exposures, with a minimum 3-week interval (Fig. 2). Each 1-h exposure session started around 8 a.m., during which the subjects were exposed to either filtered air or biodiesel exhaust with an average PM concentration of 159 μg/m³ in a human exposure chamber in a controlled randomized, double-blinded, crossover fashion. Biodiesel exhaust emissions were generated from RME in its pure form (B100) by a Volvo engine (Volvo TD40 GJE, 4.0 L, four cylinders) using the urban part of the European Transient Cycle, in order to mimic urban driving conditions [14]. The controlled environment in the chamber was continuously monitored for particulate number and concentration as well as various pollutant gases. During the exposure, subjects alternated 15-min intervals of exercise, at an average minute ventilation of 20 L/m² body surface area, with rest. The bronchoscopy procedure was performed around 6 h after completion of the exposure session, according to a well-established, previously described method [18]. The time point for the bronchoscopy procedure was chosen based on previous findings on the elevated inflammatory response in the airways at 6 h after exposure [18]. A detailed description of the method is provided in the Electronic supplementary material (ESM). Briefly, two 20-mL volumes followed by three 60-mL volumes of sterile saline solution were infused into either the right middle or left lingual lobe of the lungs, and gently suctioned back to sequentially collect bronchial wash (BW) and bronchoalveolar lavage (BAL), respectively. All samples were stored at −80 °C until analysis. The subjects were allowed to have a light breakfast in the morning, prior to the exposure, but were fasting thereafter until 2 h after the bronchoscopy procedure.Fig. 2

The study design, including sampling time points. At the first visit, the subjects were randomized to either exposure A (biodiesel exhaust) or B (sham exposure using filtered air). Bronchial wash (BW) and bronchoalveolar lavage (BAL) were collected by bronchoscopy around 6 h after completion of the 1-h exposure session. At the second visit (at least 3 weeks later), the subjects were crossed over to the other type of exposure and underwent the same sampling protocol

A previously reported solid phase extraction (SPE) protocol [31] was modified to isolate oxylipins, endocannabinoids, N-acylethanolamines, and other related lipids from BW and BAL samples. In summary, extraction was performed on Waters Oasis HLB cartridges (60 mg of sorbent, 30 μm particle size). These were first washed with ethyl acetate (1 mL) and MeOH (2 × 2 mL), then conditioned with 5% MeOH in water (containing 0.1% acetic acid) before loading the sample. Quantitative volumes from each sample (resulting in ∼6 mL BAL and 1–3 mL BW) were spiked with 10 μL internal standard (IS) mixture in methanol containing 50 ng/mL 12,13-DiHOME-d₄ and 12(13)-EpOME-d₄, 25 ng/mL 9(S)-HODE-d₄, PGE₂-d₄ and TXB₂-d₄, 800 ng/mL 2-AG-d₈, 40 ng/mL PGF_2α-EA-d₄ and PGE₂-EA-d₄, 20 ng/mL AEA-d₈, OEA-d₄, SEA-d₃ and PEA-d₄, as well as 10 μL antioxidant solution (0.2 mg/mL BHT/EDTA in methanol/water (1:1)). After applying the sample, IS, and antioxidant solution, the SPE cartridge was washed, dried under high vacuum and eluted with 3 mL acetonitrile, 2 mL methanol, and 1 mL ethyl acetate. All three fractions were pooled into polypropylene tubes containing 6 μL of a glycerol solution (30% in methanol) to trap the analytes. Eluates were evaporated using a MiniVac system (Farmingdale, NY, USA), reconstituted in 100 μL of MeOH, spiked with 10 μL recovery standard (50 ng/mL CUDA), transferred to vials, and analyzed by LC-MS/MS.

The Agilent Ultra-Performance LC system (Infinity 1290) coupled to an Agilent 6490 Triple Quadrupole mass spectrometer with an electrospray ionization source (ESI) equipped with the iFunnel Technology (Agilent Technologies, Santa Clara, CA, USA) was used. Samples for ionization in positive (endocannabinoid-related lipids) and negative (oxylipins) mode were injected separately. Chromatographic separation was performed on a Waters BEH C₁₈ column (2.1 mm × 150 mm, 130 Å, 1.7 μm particle size), with an injection volume of 10 μL for each run. The eluents in the mobile phase consisted of (A) deionized water (0.1% acetic acid) and (B) acetonitrile/isopropanol (90:10). For oxylipin separation, the following gradient was employed: 0.0–3.5 min 10–35% B, 3.5–5.5 min 35–40% B, 5.5–7.0 min 40–42% B, 7.0–9.0 min 42–50% B, 9.0–15.0 min 50–65% B, 15.0–17.0 min 65–75% B, 17.0–18.5 min 75–85% B, 18.5–19.5 min 85–95% B, 19.5–21.0 min 95–10% B, 21.0–25.0 min 10% B [31]. The separation gradient for endocannabinoids and related lipids (including prostamides) was as follows: 0.0–2.0 min 30–45% B, 2.0–2.5 min 45–79% B, 2.5–11.5 min 79% B, 11.5–12.0 min 79–90% B, 12.0–14.0 min 90% B, 14.0–14.5 min 90–79% B, 14.5–15.5 min 79% B, 15.6–19.0 min 30% B [31]. The last 3 min of the gradient was directed to the waste to reduce MS contamination.

The ESI-MS/MS conditions were as follows: capillary and nozzle voltage at 4000 and 1500 V, drying gas temperature 230 °C with a gas flow of 15 L/min, sheath gas temperature 400 °C with a gas flow of 11 L/min; the nebulizer gas flow was 35 psi, and iFunnel high and low pressure RF were set at 90 and 60 V (negative mode) and 150 and 60 V (positive mode). Dynamic multiple reaction monitoring (MRM) mode was used with fixed time windows (retention time ±2 min) to profile two transitions per compound (one quantitative and one qualitative), see Tables S1 and S2 in the ESM. The dynamic MRM option was performed for all compounds with optimized transitions and collision energies. The MassHunter Workstation software was used for instrument control and for manual integration of all peaks.

The stable isotope dilution method was used to quantify the analytes. Two types of internal standard were used: (i) for quantification purposes, deuterated IS was added before extraction, and (ii) for monitoring the loss of IS, the recovery standard CUDA was added after extraction [39]. The addition of a known concentration of CUDA after sample preparation facilitated the assessment of IS recovery by providing a standard not affected by losses during extraction to normalize the IS peak area against. Five IS were used for quantification of endocannabinoids, N-acylethanolamines, and related lipids (AEA-d₈, SEA-d₃, PEA-d₄, OEA-d₄ and 2-AG-d₈), and eight for oxylipin quantification (12,13-DiHOME-d₄, 12(13)-EpOME-d₄, 9-HODE-d₄, PGE₂-d₄, PGD₂-d₄, 5-HETE-d₈, 20-HETE-d₆ and TXB₂-d₄). For each native compound, a suitable IS was selected based on structural similarities (Table S3, ESM). Standard solutions were prepared at 10 different levels (Tables S4 and S5, ESM) to determine calibration curves by the least-squares linear regression model with equal weighting factor, using the equation y = mx + b, where y corresponds to the response ratio (native standard peak area/internal standard peak area), m is the slope of the curve, x corresponds to the on-column concentration of the analyte, and b is the y-interception of the calibration curve.

Basal statistical measures (D’Agostino and Pearson omnibus normality tests and Wilcoxon matched-pairs signed rank test) and receiver-operating characteristic (ROC) curves were performed using the statistical package built into the Graphpad Prism software (version 6, GraphPad Software Inc., San Diego, CA, USA). For analytes below limit of detection (LOD), i.e., non-detects, half of the LOD value was used in the univariate statistical calculations [40]. Bonferroni correction for multiple testing was performed to avoid false positive results. Multivariate analysis using principal component analysis (PCA) and orthogonal projections to latent structures (OPLS) with discriminant analysis (DA) was done using the SIMCA software (Version 14, Umetrics, Umeå, Sweden) with raw data as input to investigate the relationship of lipid profiles with (i) exposure type (filtered air vs biodiesel exhaust) and (ii) lung lavage type (BW vs BAL). All data were mean-centered and scaled to unit variance before modeling.

ROC curves were generated to further investigate these findings by evaluating the oxylipins and N-acylethanolamines as diagnostic markers for exhaust exposure. ROC curves are non-parametric and consider true negatives (specificity) and true positives (sensitivity) identified at a given cutoff value of the feature under analysis, commonly used in metabolomics studies to avoid misinterpretation of the data [41]. The area under the curve (AUC) of a plot of the sensitivity vs 1 minus the specificity (1-specificity) indicates the diagnostic value of each feature analyzed. The maximum value for AUC is 1, and a value of 0.5 indicates no diagnostic value. The closer the AUC is to 1, the better the diagnostic usefulness of a specific feature. The optimal threshold value, called the Youden score, can be determined to assess the cutoff point of the diagnostic marker.

The oxylipin profiling method screened 41 compounds in the samples analyzed (Table S1, ESM). Of these, 21 were present at levels above the limit of quantification (LOQ) in at least 75% of either BW or BAL samples, and therefore included in univariate statistical analysis (Table 1).Table 1

Oxylipin and N-acylethanolamine concentrations (pM) in bronchoalveolar lavage (BAL) and bronchial wash (BW) after filtered air or biodiesel exhaust (BioDE) exposure. Significant differences (Wilcoxon test with α = 0.05) are marked in bold and significant ones after Bonferroni correction (P value <0.002) are italicized. A comparison was made with literature data on lipid concentrations when present [42, 44]

Panels a and b of Fig. 3 present the total oxylipin concentrations (based on raw data) after each exposure in BW and BAL, respectively, according to their enzymatic pathway (Table S1, ESM). LOX products dominated the oxylipin profile, together with CYP metabolites (around 50% of the content for each pathway) in BAL, while LOX metabolites alone were present at the highest relative concentrations in BW (around 75%). Furthermore, BW contained higher levels of oxylipins than BAL, in agreement with Larsson et al. [42]. COX products constituted only 5 of 41 oxylipins in the analytical panel and thereby comprised the smallest fraction of the compounds detected.Fig. 3

Total oxylipin concentrations (pM) in human a bronchial wash (BW) and b bronchoalveolar lavage (BAL) after exposure to filtered air and biodiesel exhaust (BioDE). Black bars represent cyclooxygenase (COX) derivatives, white bars represent lipoxygenase (LOX) pathway derivatives, dark gray bars represent cytochrome P450 (CYP) derivatives, and light gray bars represent auto oxidation products

Of the 14 compounds screened in the endocannabinoid metabolome (Table S2, ESM) using our validated LC-MS/MS protocol, four (OEA, SEA, PEA, and POEA) were present in all BW and BAL samples and LEA was detected in 90% of the BW samples, therefore these five N-acylethanolamines were included in univariate statistical analysis (Table 1). AEA and EPEA were present in only a few samples and were therefore not subjected to further statistical analysis. Hence, in total seven compounds were detected in BW and/or BAL, including AEA and six other N-acylethanolamines (no glycerol fatty acid derivatives were detected). Endocannabinoids and other related lipids (N-acylethanolamines and glycerol fatty acid derivatives) have not previously been reported in human BW or BAL, except in a study by Zoerner et al. [43], in which AEA was elevated in BAL from asthmatics following allergen challenge. We found concentrations in the range of 1.1–803 pM in the human lung lavage fluid, with PEA and SEA being the most abundant (Table 1 and ESM Table S6).

No compound passed the D’Agostino and Pearson omnibus normality test, even with square root transformation. Shifts in oxylipin and N-acylethanolamine levels due to biodiesel exhaust exposure were assessed using the Wilcoxon matched-pairs signed rank test and the results are shown in Table 1.

In BW, decreased levels of PGF_2α (AA derivative, COX pathway) and 15-HETE (AA derivative, LOX pathway), as well as elevated levels of 9-HODE (LA derivative, LOX pathway) were associated with biodiesel exhaust exposure (Fig. 4a). In BAL, another three oxylipins displayed significant shifts following biodiesel exhaust exposure 12,13-DiHOME and 13-HODE (LA derivatives, CYP and LOX pathway, respectively) and PGE₂ (AA derivative, COX pathway), see Fig. 4b. All of the analytes in Fig. 4a and Fig. 4b displayed a sufficient number of quantified values to support the validity of replacing non-detected values with LOD/2 in the statistical analysis [40]. All were present in 100% of BW or BAL samples, except PGF_2α (10% non-detected values) and PGE₂ (13% non-detected values). Among the oxylipins and N-acylethanolamines, there were up to 26 possible comparisons, which corresponded to a threshold significance value of 0.002 (Bonferroni correction for multiple testing). All significant compounds in BAL (PGE₂, 12,13-DiHOME and 13-HODE), but none in BW, reached this level (Table 1).Fig. 4

Concentrations of oxylipins (pM) in human a bronchial wash (BW) samples and b bronchoalveolar lavage (BAL) samples that were significantly altered due to biodiesel exhaust exposure (BioDE). Data are presented as median and interquartile range. Significance determined using the Wilcoxon test, and P values are summarized in Table 1 (*P < 0.05, **P < 0.01, ***P < 0.001)

We determined the AUC for ROC curves of all metabolites detected in BW and BAL and found that metabolites in BAL that had shifted significantly were far from reaching significance in BW and vice versa. The significant ROCs for metabolites in BW, PGF_2α (AUC = 0.72, 95% CI 0.53–0.91, P = 0.038) and 9-HODE (AUC = 0.72, 95% CI 0.53–0.91, P = 0.040), are shown in Fig. 5. The Wilcoxon test also determined that these compounds had shifted significantly (Fig. 4a). In BAL, four oxylipins displayed significant AUC values (Fig. 6), largely corroborating the outcome of the Wilcoxon test (Fig. 4b): PGE₂ (AUC = 0.75, 95% CI 0.57–0.93, P = 0.001), 13-HODE (AUC = 0.76, 95% CI 0.58–0.94, P = 0.015), 12,13-DiHOME (AUC = 0.76, 95% CI 0.58–0.93, P = 0.010), and 9,10-DiHOME (AUC = 0.71, 95% CI 0.52–0.89, P = 0.044). However, 9,10-DiHOME displayed a P value close to the upper limit of significance (α = 0.05) and was not significant according to the Wilcoxon test. In Fig. 5 and Fig. 6, the Youden score for each compound is marked in red; for example, the COX metabolite PGE₂ had a Youden score of >3.22 pM. When analyzing the concentrations of each sample, we found that BAL samples collected after biodiesel exhaust exposure produced PGE₂ levels above 3.22 pM in 9 out of 15 BAL samples, while only 1 out of 15 BAL samples contained PGE₂ levels above 3.22 pM after filtered air exposure. On the contrary, PGF_2α (also a COX metabolite) had a Youden score of <16.2 pM in BW, which suggested that levels below 16.2 pM in BW were indicative of biodiesel exhaust exposure. Indeed, BW samples from biodiesel exhaust exposure contained levels of PGF_2α below 16.2 pM in 11 out of 15 samples, while only four BW samples contained PGF_2α levels below 16.2 pM after filtered air exposure.Fig. 5

Receiver-operating characteristic (ROC) curves for oxylipins in bronchial wash (BW) samples. a PGF_2α (AUC = 0.72, 95% CI 0.53–0.91, P = 0.038; Youden score <16.2 pM); b 9-HODE (AUC = 0.72, 95% CI 0.53–0.91, P = 0.040; Youden score >129 pM)

Similar to what was described by Larsson et al. [42], it was possible to group samples by lung compartment in separate clusters of BW and BAL samples obtained by a significant OPLS-DA model based on analytes found in 75% of BW or BAL samples (CV-ANOVA = 0.0002, Fig. 7). This separation occurred for samples collected after both biodiesel exhaust and filtered air exposure. Furthermore, there was a relationship between the OPLS-DA models generated for biodiesel exhaust and filtered air exposure, presented in a shared and unique structure (SUS) plot (Fig. S1, ESM) that provided evidence that the overall, compartment-specific profiles did not change considerably as a consequence of exposure to biodiesel exhaust. SUS plots of filtered air exposure profiles (y-axis) against biodiesel exposure profiles (x-axis) produced similar characteristic variables for BAL (1st quadrant) and for BW (3rd quadrant) for both types of exposure (Fig. S1, ESM)Fig. 7

Orthogonal projections to latent structures with discriminant analysis (OPLS-DA) score plot showing the separation between bronchoalveolar lavage (BAL) and bronchial wash (BW) samples independent of exposure and inclusive of all variables. Model parameters were: 1 predictive and 1 orthogonal component, R ² X (cum) = 0.425; R ² Y (cum) = 0.302; Q ² (cum) = 0.243; CV-ANOVA = 0.0002

Univariate statistics confirmed the findings that there were significant differences between the BAL and BW concentrations of 15 metabolites, independent of exposure, and 4 metabolites after only one type of exposure (Table 1). The differences between the BW and BAL profiles are of biological importance because they represent different compartments of the lung, an issue that requires further exploration. To that end, the BW and BAL samples from filtered air exposure were inspected by OPLS-DA for all analytes found in >75% of samples included (Fig. 8a), as well as with oxylipins and N-acylethanolamines as separate variables (Fig. 8b, c, respectively). The models generated, comprising all variables (CV-ANOVA = 2.1 × 10⁻⁵), only oxylipins (CV-ANOVA = 1.1 × 10⁻⁶), or only N-acylethanolamines (CV-ANOVA = 0.012), were all significant. Analytes that were responsible for the separation in Fig. 8a were mainly 9(10)-EpOME and 12(13)-EpOME (BAL), as well as 13-HODE and 15-HETE (BW); this finding matches the results from previous studies of healthy individuals [42]. The major N-acylethanolamines responsible for the separation were OEA and SEA which were present at higher concentrations in BW than in BAL (Fig. 8c).Fig. 8

Orthogonal projections to latent structures with discriminant analysis (OPLS-DA) models showing the separation between bronchoalveolar lavage (BAL) and bronchial wash (BW) samples collected after filtered air exposure comprising: a all variables (model parameters: 1 predictive and 1 orthogonal component, R ² X (cum) = 0578; R ² Y (cum) = 1; Q ² (cum) = 0.69; CV-ANOVA = 2.1 × 10⁻⁵); b oxylipin variables (model parameters: 1 predictive and 1 orthogonal component, R ² X (cum) = 0.582; R ² Y (cum) = 1; Q ² (cum) = 0.767; CV-ANOVA = 1.1 × 10⁻⁶); and c N-acylethanolamine variables (model parameters: 2 predictive and 1 orthogonal components, R ² X (cum) = 0.578; R ² Y (cum) = 0.486; Q ² (cum) = 0.556; CV-ANOVA = 0.012)

Multivariate statistical analysis was also conducted to investigate the overall trends between the two exposures (filtered air and biodiesel exhaust). The ESM provides a description of this analysis. In accordance with the results of the univariate analysis and ROC curve calculations, PGE₂ was the major contributor to the exposure-dependent separation in an OPLS-DA model of COX metabolites in BAL samples (CV-ANOVA = 0.02, Fig S2, ESM). LOX and CYP products, separately, as well as all detected lipid metabolites investigated in this study, did not produce valid models.

Below is the link to the electronic supplementary material.ESM 1

(PDF 679 kb)