Global Metabolomic Profiling Reveals an Association of Metal Fume Exposure and Plasma Unsaturated Fatty Acids

Background Welding-associated air pollutants negatively affect the health of exposed workers; however, their molecular mechanisms in causing disease remain largely unclear. Few studies have systematically investigated the systemic toxic effects of welding fumes on humans. Objectives To explore the effects of welding fumes on the plasma metabolome, and to identify biomarkers for risk assessment of welding fume exposure. Methods The two-stage, self-controlled exploratory study included 11 boilermakers from a 2011 discovery panel and 8 boilermakers from a 2012 validation panel. Plasma samples were collected pre- and post-welding fume exposure and analyzed by chromatography/mass spectrometry. Results Eicosapentaenoic or docosapentaenoic acid metabolic changes post-welding were significantly associated with particulate (PM2.5) exposure (p<0.05). The combined analysis by linear mixed-effects model showed that exposure was associated with a statistically significant decline in metabolite change of eicosapentaenoic acid [(95% CI) = −0.013(−0.022∼−0.004); p = 0.005], docosapentaenoic acid n3 [(95% CI) = −0.010(−0.018∼−0.002); p = 0.017], and docosapentaenoic acid n6 [(95% CI) = −0.007(−0.013∼−0.001); p = 0.021]. Pathway analysis identified an association of the unsaturated fatty acid pathway with exposure (p Study−2011 = 0.025; p Study−2012 = 0.021; p Combined = 0.009). The functional network built by these fatty acids and their interactive genes contained significant enrichment of genes associated with various diseases, including neoplasms, cardiovascular diseases, and lipid metabolism disorders. Conclusions High-dose exposure of metal welding fumes decreases unsaturated fatty acids with an exposure-response relationship. This alteration in fatty acids is a potential biological mediator and biomarker for exposure-related health disorders.


Introduction
The impact of welding-associated air pollutants on the health of exposed workers remains a major concern in occupational medicine and public health [1]. Exposure to fine particles (PM 2.5 ) in ambient air increases the risk for acute disorders [2][3][4][5] and chronic diseases [6][7][8][9][10]. Ultra-fine and fine welding particles often contain various metals as a result of combustion; these include iron, nickel, sulfur, copper, vanadium, cadmium, and their oxides [11][12][13]. Metal particles may induce diverse biological effects, including activation of mitogen-activated protein kinase [14], DNA damage and cell apoptosis [5,15,16], lipid peroxidation [16,17], and alteration of gene expression [18]. Further, fine metal particles inhaled into the respiratory tract translocate to blood circulation, which ultimately deposits particles in other organs and produces systemic toxic effects.
Recently, metabolomics has attracted increasing interest in the field of toxicology because it is a fast and reproducible method that directly reflects biological events [19,20]. Metabolomics uses a non-targeted approach to obtain an accurate representation of the metabolome, including small molecules involved in highly complex biochemical networks; global changes in the metabolome may relate to particular diseases or phenotypes [21,22]. The use of peripheral fluids, such as plasma or urine, for metabolomic analysis makes it an attractive method for studying the toxic effects of various exposures such as welding fumes. To date, however, few studies have investigated the systemic toxic effects of welding fumes on humans.
Taking advantage of our well-established occupational cohort of male boilermaker (welder) construction workers, we performed an exploratory metabolomics study using a two-stage, self-controlled design to explore the systemic toxic effects of welding fumes measured as metabolic changes. This study identified biological biomarkers for health risk assessment of welding fume exposure.

Ethics Statement
The Institutional Review Board at the Harvard School of Public Health approved the study protocol, and informed written consent was obtained from each adult prior to participation.

Study Design and Population
Eleven non-smoking, non-diabetic boilermakers at an apprentice welding school (Union Local 29, Quincy, MA) were recruited in 2011 to participate in a discovery study (Study-2011); five of the same subjects from Study-2011 and three new subjects were recruited in 2012 (Study-2012) for validation. All participants were selected from the well-characterized occupational cohort of male boilermaker construction workers in Eastern Massachusetts as previously described [23]. The welding school consists of a large, temperature-controlled room outfitted with ten workstations where the welders receive instruction and perform welding, cutting, and grinding techniques. Boilermakers primarily per-formed shielded metal arc (stick) and gas metal arc welding (TIG), using base metals of mild steel (manganese alloys) and stainless steel (manganese, chromium, and nickel alloys) with electrodes composed mainly of iron with variable amounts of manganese (1,5%). The same breakfast and lunch including turkey, chicken, ham sandwiches and vegetables was provided for each participant during workshop to control the inter-subject variability.

Data Collection
Peripheral blood samples were collected before (pre) and immediately after the ,5 h welding workshop (post) from all subjects. Samples were inventoried and immediately stored at 280uC. Samples of personal, integrated, gravimetric particulate matters with an aerodynamic diameter of #2.5 mm (PM 2.5 ) were collected over the duration of the welding workshop. A selfadministered questionnaire collected information on medical history, medication use, demographics, and smoking history as previously described [13].

Metabolite Profiling
Sample preparation. Frozen samples were sent to Metabolon, Inc. (Durham, NC) and accessioned into the Metabolon LIMS system by a unique identifier associated with the original source only. Samples were prepared using the automated MicroLab STARH system (Hamilton Company, Reno, NV). Recovery standards were added prior to the first step in the extraction process for quality control purposes. Sample preparation used a proprietary series of organic and aqueous extractions to remove proteins while allowing maximum recovery of small molecules. The resulting extract was divided into two fractions: one for analysis by liquid chromatography (LC), and one for analysis by gas chromatography (GC). Samples were placed briefly on a TurboVapH (Zymark, Westborough, MA) to remove organic solvents, frozen, and dried under vacuum. Extracted samples were split into equal parts for analysis by gas chromatography/mass spectrometry (GC/MS) and liquid chromatography/mass spectrometry (LC/MS) platforms. Several technical replicate samples were created from a homogeneous pool containing a small amount of each sample. Instrument variability control. Instrument variability was determined by calculating the median relative standard deviation (RSD) for internal standards added to each sample prior to mass spectrometer injection (RSD = 5%). Overall process variability was determined by calculating the median RSD for all endogenous Data extraction and quality assurance. Raw MS data files were loaded into a relational database. Peaks were identified using Metabolon's proprietary peak integration software, and component parts were stored in a separate and specifically designed complex data structure.
Compound identification. Compounds were identified by comparison to library entries of purified standards or recurrent unknown entities. Identification of known chemical entities was based on comparison to the over 1,000 commercially available, purified standard compounds registered in LIMS for distribution to both LC and GC platforms.
Normalization. For studies spanning multiple days, data normalization was performed to correct variation resulting from instrument inter-day tuning differences. Each compound was Figure 2. Scatter plots of metabolic changes by exposure of EPA and DPA. Scatter plots illustrate the biochemical compounds that had a metabolic change significantly associated with welding fume exposure: A1) eicosapentaenoic acid (EPA); A2) EPA after removal of a potential outlier (subject ID: 358); B) docosapentaenoic acid n 3 (DPA n3 ); and C) docosapentaenoic acid n 6 (DPA n6 ). The x-axis represents total PM 2.5 exposure during the welding workshop, while the y-axis represents biochemical metabolic change (post-welding workshop -pre-welding workshop). Black circles represent data from Study-2011; red triangles represent data from Study-2012. Each mark is labeled with the subject ID. doi:10.1371/journal.pone.0077413.g002 corrected in run-day blocks by equalizing the medians to 1.00 and normalizing each data point proportionately. Normalized metabolites were used for analysis.

Statistical Analysis
Demographics were presented by frequencies for categorical variables and means 6 standard deviation (mean 6 SD) for continuous variables. The change of each metabolite during welding day (D = post-pre) was calculated for each subject. The permutation test with 10,000 perturbations was used to test global metabolic changes between post-and pre-welding workshop samples.
To investigate the association between metabolic change and PM 2.5 exposure, univariate linear regression was used in Study-2011 (Eq. 1). Compounds with p#0.05 were further validated in Study-2012 using the same statistical model, followed by combined analysis of both studies using linear mixed-effects model (LMM) given a random slope of exposure for each study and autoregressive correlation structure (Eq. 2). The multiple comparison was adjusted by false discovery rate (FDR) [24]  For pathway-level association analysis, principal component analysis (PCA) was performed to generate the first principal component (PC 1 ) from all metabolites of each pathway. PC 1 was used as the response to perform linear regression with exposure in Study-2011 and Study-2012 separately, and perform LMM for combined analysis with adjustment for age and medication use. LMM analysis was performed by R ''nlme'' package. All statistical analyses were performed using R Version 15.0 (http://cran.rproject.org).
The functional network was built using MetaCore TM online software (GeneGo Inc., Carlsbad, CA) to illustrate potential biological connections of significant compounds and interactive genes. Enrichment analysis for disease-associated genes was conducted for each network in MetaCore TM . The p value for enrichment analysis was calculated using Eq. 3: where N denotes the total number of genes that were causally associated with all diseases and disorders in MetaCore TM ; n denotes the total number of genes that were causally associated with the intersecting disease of interest (e.g., lipid metabolism disorders); R denotes the total number of genes/proteins visible in the network; and r denotes the total number of genes/proteins visible in the network intersecting with the disease of interest.

Single Biochemical Analysis
Plasma samples were collected from each participant prior to and after a welding workshop for metabolomic profiling. We identified 333 compounds of known biochemical identity that comprised the analysis dataset. In Study-2011, 30 of 333 detected compounds were significantly associated with a metabolic change during welding workshop (D) with total PM 2.5 exposure (p#0.05);  however, only three passed validation in Study-2012 by the criteria of consistent effect direction of exposure and p#0.05 (Figure 1, Table 2). The combined LMM analysis showed that total PM 2.5 exposure associated with a statistically significant (p,0.05) decline in metabolic change of eicosapentaenoic acid (EPA) (b b = 20.013/ mg/m 3 ; 95% CI = 20.022,20.004; p = 0.005; q = 0.249), docosapentaenoic acid n 3 (DPA n3 ) (b b = 20.010/ mg/m 3 ; 95% CI = 20.018,20.002; p = 0.017; q = 0.313), and docosapentaenoic acid n 6 (DPA n6 ) (b b = 20.007/ mg/m 3 ; 95% CI = 20.013,20.001; p = 0.021; q = 0.313) (Figure 2). Adjustment for age and medication use did not affect significance (Table 3). LMM application with random intercepts for each study provided consistent results; however, the model did not reach convergence when random intercept and random slope were simultaneously considered.
In Study-2012, one subject (ID: 358) exhibited EPA metabolic changes much larger than others. These data were therefore removed for subsequent sensitivity analysis, which displayed a consistent trend while losing statistical significance (p = 0.447); the combined analysis remained statistically significant in univariate analysis (p = 0.030) and multivariate analysis (p = 0.041) (Figure 2-A2).
The five subjects who participated in both studies were also separately assessed by sensitivity analysis using an LMM with random slopes for each subject. Borderline statistical significance remained for EPA (p Study-2011 = 0.070; p Study-2012 = 0.088; p Combined = 0.042; p Combined = 0.012 with adjustment for age and medication use) and DPA n3 (p Study-2011 = 0.041; p Study-2012 = 0.031; p Combined = 0.003; p Combined = 0.066 with adjustment for age and medication use). LMM with random intercepts for each subject yielded similar results.
Conversely, sensitivity analysis was performed excluding the five subjects from Study-2011 who participated in both studies; this analysis included six subjects for Study-2011 and eight subjects for Study-2012. Combined analysis remained significant or borderline significant for EPA (p = 0.046), DPA n3 (p = 0.064), and DPA n6 (p = 0.020), even after adjustment for age and medication use.

Pathway Analysis
EPA, DPA n3 , and DPA n6 participate in the biosynthesis of unsaturated fatty acids (KEGG entry: map01040). To explore whether this pathway connects metabolites associated with PM 2.5 metal fume exposure, we performed PCA to integrate multiple biochemicals within each pathway. While two of 63 pathways significantly associated with PM exposure (p#0.05) in Study-2011, only the unsaturated fatty acids pathway was replicated in Study-2012 (p Study-2011 = 0.025; p Study-2012 = 0.021) and appeared in the combined analysis (p = 0.009, q = 0.145; p adj = 0.013, q = 0.204) ( Table 4). Seven biochemical molecules were detected in the unsaturated fatty acids pathway for which metabolic change negatively associated with exposure dosage; six of the seven molecules were statistically significant by combined analysis of both studies (Table S1).

Functional Network Analysis
To explore interactions of EPA and DPA with specific genes, we performed a functional network analysis. EPA extracellular and intracellular metabolites interacted with 19 genes; DPA interacted with 7 genes. All metabolite compounds and 24 interactive genes were then used to build a network (Figure 3). Using a priori knowledge, the top five diseases that associated with the most interactive genes within the network (21/24 genes) were endocrine system diseases (p = 1.6610 260 ), neoplasms (p = 2.0610 259 ), lung disease (p = 2.0610 259 ), respiratory tract diseases (p = 2.0610 259 ), and digestive system diseases (p = 2.0610 259 ). Additionally, 16 of 24 genes associated with cardiovascular disease (p = 1.3610 241 ), and 9 of 24 genes associated with lipid metabolism disorders/ hyperlipidemias (p = 2.2610 227 ). These disease enrichment profiles remained significant when all seven compounds were used to build a network by the same approach.

Discussion
A challenge to understand the adverse effects of welding fume exposure is the complex nature of environmental and behavioral influences. Metabolic profiles are greatly affected by diet, smoking behavior, medical history, or even time [25][26][27][28][29]. Kuo et al. recruited non-smoking, non-alcohol drinking welders for a metabolomic study, but intra-subject variation was not considered [30]. Here, we excluded smoking or diabetic boilermakers to reduce such confounding effects. A self-controlled design eliminated intra-subject variation, and a two-stage design avoided chance findings in our limited sample size.
The main finding of this study is that unsaturated fatty acids are consistently reduced by respirable welding fume exposure, with a significant negative exposure-response relationship. To date, many studies have investigated the beneficial effects of EPA and DHA in neural function [31], tumor suppression [32], cardiovascular event risk reduction [33], diabetes mellitus [34,35], anti-inflammatory activity [36], and blood lipid reduction [37]. Although the roles of DPA have not yet been systematically examined because of the lack of available pure DPA, there is some evidence that DPA has stronger beneficial health effects than EPA and DHA in vitro [38]. We also detected a global metabolite change, which may be explained by diet during work shift, circadian variation [30], or PM 2.5 exposure.
Surprisingly, the functional network built from unsaturated fatty acids includes 24 genes that showed very intense biological functions related to various diseases, including those involving endocrine system diseases, respiratory diseases, neoplasms, cardiovascular diseases, and lipid metabolism disorders. Intracellular EPA activates peroxisome proliferator-activated receptors (PPARs) [39], which are implicated in the pathology of numerous diseases including obesity, diabetes, atherosclerosis, and cancer (RefSeq, Jul 2008). EPA also inhibits COX-1/COX-2 [40], which encode proteins that may promote cell proliferation during tumor progression (RefSeq, Sep 2012). DPA activates RXRG [41], a member of the retinoid X receptor family that is involved in mediating the anti-proliferative effects of retinoic acid and is down-regulated in several types of human cancers, including lung cancer [42]; DPA also activates cysteine-aspartic acid protease (Caspase) family members, which are involved in cancer cell apoptosis [43]. The functional network analysis strengthens the hypothesis that the decline of unsaturated fatty acids is a potential mediator of multiple health disorders in boilermakers.
Previous studies demonstrate the association of welding fume exposure on decline of heart rate variability (HRV), an important indicator of cardiovascular disease morbidity [4,44,45]. Interestingly, several recent studies show that supplementation of EPA and DPA benefit HRV in healthy adults, cardiovascular disease risk populations, or in cardiovascular disease patients [46][47][48][49][50]. These results strongly indicate that the metabolic change of unsaturated fatty acids is an important biological mediator of exposure-related decreases in HRV.
Despite these findings, we recognize limitations to the study: we lack metabolite information from a non-welding day, and p values were not small enough to pass multiple comparison correction due to small sample size. However, pathway-based analysis showed an acceptable FDR (q,0.2), and the two-stage, self-controlled study design helped reduce chance findings.
In summary, this exploratory study shows evidence that high dose exposure of metal welding fumes decreases unsaturated fatty acids with an exposure-response relationship. The metabolic change in unsaturated fatty acids is a potential biomarker for exposure-related health disorders for which further studies are encouraged.

Supporting Information
Table S1 Association analysis for metal welding fume exposure and metabolic change of biochemical compounds of unsaturated fatty acid pathway. (DOCX)