An In Vitro Alveolar Macrophage Assay for the Assessment of Inflammatory Cytokine Expression Induced by Atmospheric Particulate Matter Zana Sijan,1 Dagmara S. Antkiewicz,2 Jongbae Heo,1 Norman Y. Kado,3,4 James J. Schauer,1,2 Constantinos Sioutas,5 Martin M. Shafer1,6 1
Department of Environmental Chemistry and Technology, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA 2
Department of Environmental Toxicology, Wisconsin State Laboratory of Hygiene, University of Wisconsin-Madison, Madison, Wisconsin 53718, USA
3
Department of Environmental Toxicology, University of California-Davis, Davis, California, USA
4
California Environmental Protection Agency, Air Resources Board, Sacramento, California, USA
5
Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, California 90089, USA
6
Wisconsin State Laboratory of Hygiene, University of Wisconsin-Madison, Madison, Wisconsin 53718, USA
Received 24 September 2013; revised 9 January 2014; accepted 19 January 2014 ABSTRACT: Exposures to air pollution in the form of particulate matter (PM) can result in excess production of reactive oxygen species (ROS) in the respiratory system, potentially causing both localized cellular injury and triggering a systemic inflammatory response. PM-induced inflammation in the lung is modulated in large part by alveolar macrophages and their biochemical signaling, including production of inflammatory cytokines, the primary mechanism via which inflammation is initiated and sustained. We developed a robust, relevant, and flexible method employing a rat alveolar macrophage cell line (NR8383) which can be applied to routine samples of PM from air quality monitoring sites to gain insight into the drivers of PM toxicity that lead to oxidative stress and inflammation. Method performance was characterized using extracts of ambient and vehicular engine exhaust PM samples. Our results indicate that the reproducibility and the sensitivity of the method are satisfactory and comparisons between PM samples can be made with good precision. The average relative percent difference for all genes detected during 10 different exposures was 17.1%. Our analysis demonstrated that 71% of genes had an average signal to noise ratio (SNR) 3. Our time course study sug-
Additional Supporting Information may be found in the online version of this article.
(SCAQMD); the California Air Resources Board (CARB); the USC Viterbi School of Engineering.
Correspondence to: D. S. Antkiewicz; e-mail: [email protected]
Published online 00 Month 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/tox.21961
Contract grant sponsors: Southern California Particle Center (SCPC) funded by USEPA; the South Coast Air Quality Management District C 2014 Wiley Periodicals, Inc. V
1
2
SIJAN ET AL.
gests that 4 h may be an optimal in vitro exposure time for observing short-term effects of PM and capturing the initial steps of inflammatory signaling. The 4 h exposure resulted in the detection of 57 genes (out of 84 total), of which 86% had altered expression. Similarities and conserved gene signaling regulation among the PM samples were demonstrated through hierarchical clustering and other analyses. Overlying the core congruent patterns were differentially regulated genes that resulted in distinct sample-specific gene expression “fingerprints.” Consistent upregulation of Il1f5 and downregulation of Ccr7 was observed across all samples, while TNFa was upregulated in half of the samples and downregulated in the other half. Overall, this PM-induced cytokine expression assay could be effectively integrated into health studies and air quality monitoring programs to better understand relationships between specific PM components, oxidative stress activC 2014 Wiley Periodicals, Inc. Environ Toxicol 00: 000–000, 2014. ity and inflammatory signaling potential. V Keywords: air pollution; particulate matter; inflammation; toxicity; cytokines; gene expression; PCR array; oxidative stress; ROS; macrophages
INTRODUCTION Air pollution in the form of atmospheric particulate matter (PM) is of major concern to human health. According to the World Health Organization, PM contributes to 800,000 premature deaths annually (WHO, 2002). Ambient PM exposures have been linked to asthma, airway irritation, bronchitis, respiratory infections, reduced lung function, cardiovascular and cardiopulmonary diseases (for review see: Anderson et al., 2012; Rohr and Wyzga, 2012). Particulate matter released into the atmosphere comes from natural (e.g., volcanic eruptions, sand storms, and soil resuspension) and anthropogenic (e.g., power plant combustion, automobile exhaust, and industrial emissions) processes and has diverse chemical and physical properties. Although the chemical composition of these different sources is well studied, understanding of the differential toxicity of these and other PM sources is central to the development of effective and cost-efficient control strategies and regulations. The relative toxicity of PM is strongly dependent on the particle size and composition. Fine and ultrafine particles, PM2.5 and PM0.1, have the ability to penetrate deep into the airways, thus potentially having a greater impact (reviewed in Valavanidis et al., 2008). Several mechanisms of aerosol toxicity have been identified, but oxidative stress and resulting inflammation along with DNA and lipid oxidative damage underlie many disease processes (reviewed in Valavanidis et al., 2008). Oxidative stress is induced by reactive oxygen species (ROS), which include highly reactive compounds containing molecular oxygen, superoxide and hydroxyl radicals, as well as non-radical compounds such as hydrogen peroxide and lipid peroxides (reviewed in Lodovici and Bigagli, 2011). Components of PM that have been associated with production of ROS include organic compounds (e.g., polycyclic aromatic hydrocarbons and quinones), vehicle exhaust particles (e.g., black carbon) and transition metals including iron, vanadium and nickel (for review see MazzoliRocha et al., 2010; Lodovici and Bigagli, 2011). In recent years multiple studies have referred to the interplay of inflammation and oxidative stress in PM-induced toxicity (reviewed in Tao et al., 2003; Happo et al., 2007; Jalava
Environmental Toxicology DOI 10.1002/tox
et al., 2007; Michael et al., 2013). One of the key cells involved in PM-induced inflammation are alveolar macrophages (AMs), which are located at and recruited to the interface of air and lung tissue, and play a major role in infection control and pulmonary house-keeping by ingesting foreign particles (reviewed in Twigg, 2004). Phagocytosis of PM by macrophages, as well as stimuli from soluble PM components, and other oxidative stressors, may result in NADPH oxidase activation and superoxide production (reviewed in Iles and Forman, 2002). This oxidative burst in turn can lead to activation of signaling pathways resulting in cell proliferation, gene expression and apoptosis (reviewed in MazzoliRocha et al., 2010). The major signaling pathways activated by ROS are: NFjB, MAPK, and AP-1 (Shukla and Mossman, 2008; reviewed in Corcoran and Cotter, 2013). Inflammatory cytokines, such as TNFa, various interleukins (Il1b, Il6, & Il8), and TGFb are key signaling proteins synthesized and released by macrophages under the oxidative stress conditions (for review see Laskin and Pendino, 1995; Foster, 2001; Hoek and Pastorino, 2002). The type and quantity of cytokines released and their interaction with the receptors, regulators, and inhibitors of the inflammatory processes control the class of immune cells being activated (reviewed in Borroni et al., 2010). This ultimately directs the outcome of the inflammatory cascade. Thus, macrophages and cytokine signaling, play a key regulatory role in innate alveolar defenses as well as in the inflammatory processes in the lungs (reviewed in Stow et al., 2009) and are well fit to use as an in vitro model to explore the connections between PM-induced ROS and inflammation. The primary objective of this study was to develop a flexible and robust in vitro method for advancing our understanding of the drivers of PM toxicity that lead to oxidative stress and inflammation in mammals. We utilized a relevant rat alveolar macrophage model, applying a cell line that exhibits all the hallmarks of PM toxicity, including inflammatory signaling, resembling those seen in human primary macrophages (Hidalgo et al., 1992; Shi et al., 1996; Lane et al., 1998). The NR8383 cells have been used in many previous studies of PM-induced ROS activity (Fach et al., 2002; Riley et al., 2005; Zhang et al., 2008; Shafer et al., 2010) and of
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
cytokine-related inflammation processes (Wong et al., 2005; Haberzettl et al., 2007; Ren et al., 2011; Zeng et al., 2013). ROS was measured using a fluorescent indicator DCFHDA and a suite of inflammatory cytokines and cytokine receptors was assessed using quantitative reverse transcription polymerase chain reaction (RT-PCR) arrays. Method performance was characterized using extracts of ambient PM and engine exhaust PM samples. This novel application of the PCR Array method, as a high throughput screen of varied PM samples for their induction of inflammatory signaling molecules, tied with parallel measurement of ROS induction and up-front chemical speciation of the samples will help us gain insight into the mechanisms underlying PM-induced toxicity and inflammation.
MATERIALS AND METHODS Cell Culture The rat alveolar macrophage cell line, NR8383 (ATCC, Manassas, VA, USA) was cultured in Hams F12 medium (Sigma, Saint Louis, MO, USA) and 15% fetal bovine serum, at 37 C in a humidified 5% CO2 incubator. Cell culture was maintained by transferring non-adherent cells to new flasks bi-weekly at a floating cell concentration of 400 cells lL21 media.
PM Sample Selection The NR8383 cell line was exposed to four contrasting ambient air PM samples and two engine exhaust emission particulate samples (Supporting Information Table SI) to assess method performance and evaluate the expression levels of a large group of inflammatory cytokines in representative realworld samples. The ambient PM from Milan was collected as the fine aerosol fraction, PM2.5 (Dp < 2.5 lm). The Milan PM sample was a composite of four weekly collections from an urban setting of Milan, Italy, one of the most heavily polluted cities in Europe (Daher et al., 2012a). Engine exhaust samples were collected from a 2009 six-cylinder light duty flex fuel vehicle burning E-85 fuel (85% ethanol, 15% regular unleaded California gasoline). The vehicle was run on a chassis dynamometer using a Los Angeles adapted transient test cycle (Unified Driving Schedule) and collected over multiple runs. The two engine PM samples were collected approximately one week apart. The PM samples were analyzed using high resolution (magnetic-sector) inductively coupled plasma mass spectrometry (ICP-MS), and considerably higher levels of redox-active metals (e.g., Fe, Cr, Ni, Pb, and Pt) were measured in Engine 1b compared to Engine 1a (data not shown). The samples were collected as total suspended particulate [TSP; though the particle size of the engine exhaust PM is predominantly ultrafine ( 30, a derivative reporter value 3 C between the two peaks) to identify problematic PCR products. Only three genes (Ccl25, Il10, and Il1f5) from three separate sample exposures did not pass the screening criteria. Note: the PCR Array version we used has subsequently been upgraded and the three genes mentioned above are no longer present.
Cluster Analysis Hierarchical cluster analysis was applied to the gene fold regulation values from each sample to assist in the identification of related and coregulated genes. The clustering method used was that coded in the Cluster 3.0 software available at http://rana.lbl.gov/EisenSoftware.htm. Dendrograms were produced with Tree View software (M. Eisen;
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
5
Fig. 1. The reproducibility of the PCR Array method evaluated from duplicate experiments of ten individual PM exposures, represented as the median relative percent difference (RPD) of fold change for expressed genes (ranked from smallest to largest RPD value).
http://rana.lbl.gov/EisenSoftware.htm), and the similarity indices determined from the clustering analysis were plotted in heat-maps to visualize the outcomes. A detailed description of the clustering method used on our data set can be found in the Supporting Information. It is important to note that the heat-map outcomes do not necessarily represent significantly up or down regulated gene expression as presented in the gene expression color matrix [Fig. 4(b)].
Signal to Noise (SNR) The SNR was calculated for each gene of interest by dividing the average Ct value of duplicate exposure experiments by the standard deviation of the untreated control’s Ct values at the corresponding exposure time. All gene expression data used in the SNR analysis were normalized to the housekeeping genes. In our presentation of the SNR, we have used a ratio of 3.0 (three sigma) as our estimated limit of detection (LOD).
Method Performance Metrics Fold Change Relative Percent Difference (RPD)
RESULTS
As noted above, replicates of individual sample exposures and their respective negative controls were all processed separately (full method duplicates) and analyzed on separate PCR Arrays. The RPD of gene-specific fold change for each detected gene was calculated by taking the absolute value of the difference in fold change between the sample duplicates and dividing by the arithmetic mean of the fold changes (and multiplied by one hundred). The overall study RPD of each gene was determined by calculating the median of the gene-specific RPD values obtained across the ten separate duplicated exposures considered for this analysis (n 5 10).
Method Performance The Rat Inflammatory Cytokines and Receptors PCR Array (SABiosciences) platform was used in conjunction with the rat alveolar macrophage cell line (NR8383) and six contrasting PM samples to characterize the sample-specific inflammatory signaling. To assess the performance of our PCR Array based method, we evaluated the method’s reproducibility and sensitivity. As one measure of reproducibility, the RPD was calculated from individual gene fold changes for experimental duplicates of ten different exposures (Fig. 1 and Supporting Information Table SII). Each exposure was a
Environmental Toxicology DOI 10.1002/tox
6
SIJAN ET AL.
17.1%. Moreover, 68% of the detected genes had a median RPD value below 20%. Only two genes, Il3 and Il1f6, showed a median RPD value above 35%. Overall, the reproducibility of our method was satisfactory, suggesting that comparisons between various PM samples can be made with reasonable precision. To evaluate the sensitivity of our method, the SNR was determined and presented in Supporting Information Figure S1 for all of the detectable genes of the following samples: Riverside, USC and Engine 1a. We found that 77% of genes for Riverside, 50% for USC, and 61% for Engine 1a sample extracts were above the SNR 5 3 level for gene expression. Overall, our PCR Array based method detected the PM-induced expression of inflammatory cytokines with good sensitivity, with 71% of genes having an average SNR of 3.
Time Course Analysis of InflammationRelated Gene Expression
Fig. 2. A heat map of the Euclidean distance hierarchical clustering depicting the time-course (3, 4, 6 h) of gene expression changes (fold regulation) observed in three different samples: ambient PM, engine emission PM, and Zymosan. Each column represents a PM sample and each row represents a gene. The shades of red and green represent high and low expression levels relative to the mean fold regulation (see bar color).
complete method replicate to ensure that all the potential sources of the method’s variability were captured. The median percent RPD for all of the genes detected was
Environmental Toxicology DOI 10.1002/tox
To determine the optimal exposure time for the PCR Array method, a time course study was conducted with endpoint times of 3, 4, and 6 h. Three contrasting samples were chosen for evaluation to ensure that native variability was captured: soluble fractions of one atmospheric PM sample (Riverside) and an engine exhaust PM sample (Engine 1a), as well as a suspension of a particulate control (zymosan). Euclidean distance hierarchical clustering of the resulting fold regulation in gene expression (relative to untreated controls) was performed to visualize the gene expression patterns (Fig. 2). The 4-h exposure time point resulted in the largest number of genes with altered expression and the greatest number of upregulated genes for the ambient and engine PM samples. The Riverside sample exhibited altered expression in 33% of genes at 3 h, 69% of genes at 4 h, and 26% genes at 6 h out of the total detected genes. A substantial portion of the altered genes (36%) showed a pattern of downregulation at 3 h, followed by upregulation at 4- and 6-h exposure. Out of the total detected genes at the 4-h exposure, 57% were upregulated. The outcomes of the time course analysis conducted with Engine 1a revealed a gene expression profile similar to that observed with the Riverside sample, with the largest number of genes altered at the 4-h time point (46% of total number of detectable genes). However, while the pattern of upand downregulation was comparable to Riverside at 3- and 4-h exposure, significantly greater downregulation was observed at 6 h (50% of altered genes). In addition, the 4h exposure resulted in the largest number of upregulated genes (26% of total number of detectable genes). Interestingly, while the number of genes that are induced by the Riverside PM at 4 h was more than double that at 3- or 6h exposure, the total number of altered genes in Engine 1a exposures remained fairly similar throughout the time course.
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
In contrast to the atmospheric and engine exhaust PM samples, the yeast-derived particulate control, zymosan (which was not 0.22-lm-filtered, and exists as a colloidal/ particulate suspension) resulted in a distinct gene expression pattern (Fig. 2). Longer exposure led to progressively stronger and more prevalent upregulation of altered genes, and importantly upregulation was the super-dominant mode of expression at all time points (only eight out of 73 detected genes were downregulation throughout the time course). Overall, the ambient and engine exhaust samples exhibited the strongest upregulation and the largest number of genes altered at the 4-h time point. Based on these observations, the 4-h exposure time was chosen for the remainder of the study as the time point most likely to capture the largest number of differentially expressed genes.
Regulation of Inflammatory Signaling by Contrasting PM Samples Six contrasting PM samples, exposed for 4 h, were used to characterize the sample-specific signature of inflammatory signaling (Fig. 3). The PM mass used 4 h ranged between 64 and 302 lg mL21, with the Engine 1a sample being the lowest in mass, followed by Milan, USC, Long Beach, Riverside, and Engine 1b. Overall, across all samples, out of a total of 84 genes we detected 57 differentially expressed genes (those that met QC criteria and fell below the PCR detection cutoff Ct value of 33). A threshold of 1.7-fold change was used for gene expression to be classified as meaningfully altered. Across the six sample exposures, a total of 49 genes resulted in a meaningfully up- or downregulated expression, as shown in Figure 3. Samples varied considerably in their potential to initiate inflammatory signaling, with the Riverside sample exhibiting the highest proportion of altered genes (70% of total detected), followed by Long Beach (57%), Engine 1b (53%), Engine 1a (46%), Milan (37%), and USC (9%). Furthermore, a majority of the altered genes were found to be upregulated (66% of all altered). As shown in Figure 3, the general patterns of gene regulation in the Riverside, Milan, Engine 1a and Engine 1b PM extracts are similar. In particular, they all share a high fold regulation (30 to 55-fold) of Ccl25 and Il1r2 (refer to Supporting Information Table SII for a complete list of full gene names and their abbreviations). Furthermore, the core nine genes that were meaningfully altered in at least five out of six samples had their expression levels regulated in the same direction in the above-mentioned four samples. Three genes were found to be substantially altered in all six samples: Ccr5, Il1f5, and Il1f6. Il1f5 and Il1f6 showed significant upregulation for all six exposures. In contrast, Ccr5 was downregulated in the USC, but upregulated in all the remaining samples. On the flip-side, many gene regulation differences were observed between samples which may form the basis for a
7
sample-specific fingerprint. For example, the Riverside sample could be distinguished from the rest of the samples with its substantial upregulation of Ccl22, Il6st, Itgam, and Tnfrsf1b. The measured gene expression profiles of the Long Beach and USC samples differed from one another and from the rest of the samples. The USC sample induced an upregulated expression for Ccl2 while the other ambient samples induced a downregulated expression. Furthermore, Il1r2, Il13ra1, and Tollip showed downregulated expression in the USC sample but upregulated expression for the rest of the samples. In addition, the Long Beach sample is the only ambient sample which exhibited a downregulation of Cxcl2. The Engine 1a and Engine 1b emission samples, collected a week apart from the same engine running on E85 fuel, shared very similar gene regulation profiles (Fig. 3). Out of the 49 genes with meaningful up- and downregulation, 71% had the same direction of regulation. Some differences were however apparent (e.g., the pro-inflammatory cytokines: Cxcl1, Cxcl2 Ccl2 were downregulated in Engine 1a, but upregulated in Engine 1b) and these contrasts became more apparent in the cluster analysis (refer to section below). Overall, our method was capable of identifying commonalities and differences in inflammatory signaling regulation across various ambient PM samples.
Hierarchical Clustering of Contrasting PM Samples Another valuable tool for analyzing the similarities and differences among samples, as well as for finding patterns of gene expression is Euclidean distance hierarchical clustering. We applied this analysis to replicates of 4-h exposures of PM samples [Fig. 4(a,b)]. The 56 genes analyzed fell into seven major clusters as shown in Figure 4(a). Cluster 1 comprises of pro-inflammatory cytokines and shows high similarity for the Riverside sample in the positive direction and for the Milan sample in the negative direction. Cluster 2, also consisting of pro-inflammatory cytokines, revealed differences between the two engine exhaust samples with genes significantly upregulated for Engine 1b but downregulated or neutral for Engine 1a [Fig. 4(b)]. Interestingly, the three core genes Ccr5, Il1f5, and Il1f6, which are significantly altered in all six samples, do not cluster together. Ccr5 clusters separately from Il1f5 and Il1f6 likely due to downregulation induced by the USC sample, which was not observed for Il1f5 and Il1f6. Furthermore, a distinct cluster separate from the rest of the genes and consisting only of Il6r and Spp1 was downregulated for the Riverside, Long Beach and Engine 1b, but not for the remaining samples. Considering that Il6r is downregulated in all six samples and Spp1 is upregulated in the Milan, USC and Engine 1b only (Fig. 3), clustering revealed an association not readily observable with other data presentation methods.
Environmental Toxicology DOI 10.1002/tox
8
SIJAN ET AL.
Fig. 3. Comparison of fold regulation of genes following 4-h exposure. The dashed lines represent threshold values for gene expression upregulation (1.7 fold) or downregulation (1.7 fold). Grouping and color coding was used to represent the frequency with which a gene was significantly altered across the six samples. Gene expression significantly altered in only one of the six PM samples is indicated with an asterisk. The first three groups of genes correspond to the left y-axis and the latter three groups of genes correspond to the right y-axis.
Clustering also revealed that the Riverside and Long Beach samples shared similar expression patterns and together with the Milan and Engine 1a samples formed a larger cluster, separate from the USC and Engine 1b. In general, this observation agrees with what is presented in Figure 3, where the Riverside and Long Beach induce the same
Environmental Toxicology DOI 10.1002/tox
direction of regulation in 83% of the altered genes. However, clustering exposed contrasts in the PM-induced gene expression which were not readily apparent in Figure 3—for example, the engine emission samples (Engine 1a and Engine 1b) do not cluster together. This disparate clustering may be driven by the differential regulation of select genes.
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
9
Fig. 4. Fold regulation of inflammation-related genes induced by 4-h exposure to PM samples. A heat map of hierarchical clustering where each column represents a sample and each row represents a gene. Genes included in the analysis were expressed in three or more samples. A color matrix showing the pattern of significantly (61.7) up (red) and down (green) regulated genes, where the ordering of the genes is matched to the heat map [Fig 4(a)]. Gray corresponds to unaltered expression and black represents no signal detected.
As shown in Figure 3, out of the 49 altered genes, 14 are differentially regulated between the two engine samples (e.g., Ccl2 and Cxcl2 are upregulated in Engine 1b and downregulated in Engine 1a). Interestingly, a majority of those genes are found in Clusters 2 and 4 with high similarity in the positive direction for Engine 1b and in the negative direction for Engine 1a. Overall, clustering is a valuable tool, sensitive to direction of regulation and can present the gene data in a
unique manner, not easily observed in other data presentations.
Comparison of ROS Induction and Inflammatory Cytokine Activation To explore relationships between ROS generation and inflammation, the ROS activity of each PM sample was assessed
Environmental Toxicology DOI 10.1002/tox
10
SIJAN ET AL.
A Pearson Correlation matrix was constructed to help identify relationships between a sample’s ROS activity and the potential to regulate individual cytokine expression. Figure 5(b) shows that four genes (Ccl17, Cxcl1, Il1a and Il3) have a high positive correlation with ROS activity. Across the six samples, however, 80% of meaningfully altered inflammation-related genes had a low positive or a low negative correlation with the ROS activity. Seven genes do not have a significantly altered gene expression and correlate with ROS activity. These results, suggest that the relationship between ROS activity and inflammatory signaling is complex. However, these results are based on a relatively small data set and a larger data set may strengthen or reveal new relationships between ROS generation and inflammatory signaling.
DISCUSSION
Fig. 5. ROS activity and the 4-h exposure inflammationrelated expression relationships for the study samples. PMinduced ROS activity, in alveolar macrophages, normalized to the amount of PM (in mg, shown above each bar) used in the PCR Array experiments (sample-specific). Categorization of genes based upon a Pearson Correlation matrix of ROS activity and inflammation related gene fold regulation. Categories 1–3 represent genes which have significantly altered expression in at least one sample and Category 4 represents genes which are not significantly altered in any sample but correlate with ROS activity, the asterisk indicates that the gene has an R > 0.7. The sign and R cutoff criteria are given in the column headings.
following a 2.5-h exposure, as described previously (Landreman et al., 2008). The ROS activity of each PM sample was normalized to the corresponding total PM mass used for a given sample in the PCR Array analysis experiments. Therefore, direct comparisons can be made between ROS activity and the amount of cytokine signaling regulation produced by the same PM sample [Fig. 5(a)]. When normalized in such matter, the highest ROS activity was induced by the Riverside sample, which also had the largest magnitude of cytokine induction, both in terms of individual gene fold-change upregulation and in the proportion of genes altered. The second highest fold upregulation of gene expression was observed in the Milan sample, which was also second in the order of ROS induction. Interestingly, the Long Beach and USC samples induced a very similar ROS activity but the USC sample induced the least number of meaningfully altered genes. ROS activity induced by Engine 1a and Engine 1b is similar; however, Engine 1b generated a larger ROS activity.
Environmental Toxicology DOI 10.1002/tox
The NR8383 rat alveolar macrophage cell line is a relevant and reproducible biological model for measuring inflammatory cytokine expression and ROS production induced by atmospheric PM. When compared to primary cell cultures, NR8383 cells exhibit all the essential characteristics of primary macrophages, with appropriate response to microbial, particulate and soluble stimuli with phagocytosis and killing (Lane et al., 1998). The model cell line also exhibits strong oxidative burst with documented secretion of macrophagespecific cytokines, and replicative response to exogenous growth factors (ATCC; http://www.atcc.org/). In addition, the ROS activity of PM induced in NR8383 cells have been shown by Delfino et al. (2010) to correlate with both airway and systemic inflammation markers in a human cohort exposed to the same environmental PM. Myatt et al. (2011) have shown that upon PM2.5 exposure rat alveolar macrophages release TNFa in strong correlation with TNFa release in human dendritic cells. Our alveolar macrophage-based cytokine expression assay was found to have excellent reproducibility and sensitivity, with performance metrics at least as good as or better than those reported for other gene expression assays. The median RPD of fold changes for all detectable genes was 17.1%, with the RPD values reflecting complete method replicates, from cell exposure through the RT-PCR analysis. This inclusion of most of the potential sources of variability in our analysis likely increased the reported RPD values. Nonetheless, the exhibited RPD values are well within the range of other interassay reproducibility values, ranging from 1 to 24% coefficient of variance (CV), reported in gene expression studies (Reddy and Wilkie, 2000; Hoadley and Hopkins, 2003; Cui et al., 2006; Allen et al., 2007; Pathak et al., 2012). Even the intra-assay reproducibility values reported in literature (typically replicate wells or arrays with the same starting material), can be in similar range of up to 17% CV (Hoadley and Hopkins, 2003; Cui et al., 2006; Allen et al., 2007; Pathak et al., 2012).
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
The manufacturer-reported sensitivity of the PCR Array that we applied is >80% positive detection with total RNA input as low as 25.0 ng and as high as 1.0 lg per array plate (Arikawa et al., 2007). In our experiments, a range of 0.34–1 lg of total RNA was available for input per RT-PCR Array analysis. The usable sensitivity of our method was further quantified by determining SNR metrics of the individual gene’s Ct values (Supporting Information Fig. S1). Using ambient and engine exhaust samples with PM masses that might typically be available from routine air pollution monitoring; we showed that 71% of the detectable genes had SNR 3. For example, we observed a high level of sensitivity for Tnf, Bcl6, Il1a, and Il1b with SNR values of 14.77, 16.51, 5.01, and 3.93, respectively. The SNR3 target has been previously considered as a LOD in other assays and is comparable or higher than SNR cutoffs used in several microarray and RT-PCR gene expression studies (Fanganiello et al., 2007; Colombo et al., 2009; Colak et al., 2012; Dao et al., 2012). Colombo et al. (2009) analyzed a large microarray data set showing that 12% of genes had a SNR1, while our Riverside data processed in a manner consistent with the above mentioned study resulted in 61% of the genes having an SNR 1 (data not shown). Interestingly, our data showed no clear correlation between the Ct value and the SNR of a gene. Overall, this novel macrophage based method for assessing PM-induced toxicity proved to be robust with a high degree of reproducibility and sensitivity. The sensitivity, as well as applicability, of a cytokine detection method is also influenced by the exposure time. PM-induced cytokine expression has been reported for treatments ranging from 2 to 72 h (Carter et al., 1997; reviewed in Becker et al., 2005; Ishii et al., 2005; Imrich et al., 2007; Mazzarella et al., 2012; Michael et al., 2013). In our time course study we were able to detect significantly altered gene expression as early as 3 h (Fig. 2), which pushes the lower-end of the range (5–24 h) suggested by some research groups as optimal to obtain measurable effects on cytokine production (reviewed in Mitschik et al., 2008). Thus, in addition to good reproducibility, the macrophage model appears to exhibit more than sufficient, if not exceptional sensitivity, for gene expression studies in the context of air pollution toxicity. Furthermore, single time point, extended exposures can fail to capture the initial induction of inflammatory signaling, as can be seen in case of Ccl25 and several interleukins in the Riverside sample (Fig. 2). Our choice of the 4-h exposure time was also dictated in part by its proximity to the optimum exposure time (2.5 h) for the measurement of PM-induced ROS (Landreman et al., 2008), thus allowing for more robust insights into the ROS-driven inflammatory gene induction. A review of the existing studies of cytokine expression revealed that the exact time course of expression will vary for each gene, cell type, as well as PM dose, PM size and composition (Pozzi et al., 2003, 2005; Ovrevik et al., 2009; Sun et al., 2012; Michael et al., 2013). There-
11
fore, to obtain meaningful results, a preliminary time course analysis of cytokine regulation should be considered for each cell exposure system and PM type. Other factors to consider in the choice of exposure time for gene expression (and therefore method appropriateness) are the viability of the test organism with extended exposure and the composition of the exposure medium. Even though a large number of studies have been conducted with exposures 24 h, it is important to note that in many of these studies, toxicity or loss of cell viability was evident at this point (Li et al., 2000; Happo et al., 2007; Ovrevik et al., 2009; Michael et al., 2013). In our method, we have purposely chosen a simple assay medium (SGM) that will not substantially alter the speciation of components extracted from the PM (e.g., not chelate metals). This is critically important, since an accurate assessment of the toxicity of PM depends upon maintaining the integrity and speciation of the constituent species. The SGM (and many similar media) is, however, not capable of supporting optimal cell health for prolonged periods of time, thus dictating shorter exposure times. Moreover, our 6-h time course study demonstrates that the 4-h exposure time point works well for detecting primary changes in the PM-induced signaling, since by 6 h we could already detect evidence of secondary feedback loops. Several published studies have shown that fine and ultrafine PM exposure can lead to inflammation (reviewed in Shrey et al., 2011). As shown in Figure 3, seven out of the core nine genes (Ccr5, Il1f6, Ccl25, Ccl5, Cxcl10, Cxcr2, and Il8ra), which were upregulated in all the PM samples evaluated have pro-inflammatory characteristics (Menten et al., 2002; Cheong et al., 2006; Matheson et al., 2006; Dinarello, 2009; Schmutz et al., 2010; Skuljec et al., 2011; Musah et al., 2012). Furthermore, the Riverside, Milan, Engine 1a and Engine 1b samples induced a 30- to 49-fold increase of Ccl25. The Riverside sample also induced significant upregulation of Ccl22, Il6st, Itgam, and Tnfrsf1b, which have been shown to have pro-inflammatory characteristics (reviewed in Heinrich et al., 2003; Solovjov et al., 2005; Semkova et al., 2011; Lilly et al., 2012). In addition, the Engine 1b sample caused considerable upregulation of Ccl2 and Cxcl2, both considered pro-inflammatory chemokines (Wolpe et al., 1989; reviewed in Melgarejo et al., 2009). Conversely, we also observed a substantial upregulation of Il1r2, which acts as a decoy receptor for proinflammatory cytokines Il1a and Il1b, suggesting an anti-inflammatory effect (Dinarello, 2009). Out of the core nine cytokines, Il1f5 and Bcl6 have anti-inflammatory functions (Dinarello, 2009; Gongol et al., 2013). Overall, the method provides a powerful and relevant approach for assessment of a wide array of both pro- and anti-inflammatory cytokines from small masses of aerosol PM. Our analysis of PM-induced cytokine regulation revealed that the water extracts of various PM samples induced a predominantly pro-inflammatory response in alveolar macrophages. Importantly, outside of the conserved
Environmental Toxicology DOI 10.1002/tox
12
SIJAN ET AL.
upregulation of the core inflammatory cytokines, each of the PM samples tested could be differentiated from the others by some unique aspect of the gene expression profile (Fig. 3). However, our study had a relatively limited number of replicate experiments and a larger data set may reveal new information in the gene expression profiles. Some of the expression profile differences observed in our samples may partly be due to differences in the PM sizefraction studied. The ambient air samples from the LA Basin were collected as the quasi-ultrafine (PM0.25), while the Milan sample was collected as the fine fraction (PM2.5) and the engine PM samples nominally
Department of Environmental Chemistry and Technology, University of Wisconsin-Madison, Madison, Wisconsin 53706, USA 2
Department of Environmental Toxicology, Wisconsin State Laboratory of Hygiene, University of Wisconsin-Madison, Madison, Wisconsin 53718, USA
3
Department of Environmental Toxicology, University of California-Davis, Davis, California, USA
4
California Environmental Protection Agency, Air Resources Board, Sacramento, California, USA
5
Department of Civil and Environmental Engineering, University of Southern California, Los Angeles, California 90089, USA
6
Wisconsin State Laboratory of Hygiene, University of Wisconsin-Madison, Madison, Wisconsin 53718, USA
Received 24 September 2013; revised 9 January 2014; accepted 19 January 2014 ABSTRACT: Exposures to air pollution in the form of particulate matter (PM) can result in excess production of reactive oxygen species (ROS) in the respiratory system, potentially causing both localized cellular injury and triggering a systemic inflammatory response. PM-induced inflammation in the lung is modulated in large part by alveolar macrophages and their biochemical signaling, including production of inflammatory cytokines, the primary mechanism via which inflammation is initiated and sustained. We developed a robust, relevant, and flexible method employing a rat alveolar macrophage cell line (NR8383) which can be applied to routine samples of PM from air quality monitoring sites to gain insight into the drivers of PM toxicity that lead to oxidative stress and inflammation. Method performance was characterized using extracts of ambient and vehicular engine exhaust PM samples. Our results indicate that the reproducibility and the sensitivity of the method are satisfactory and comparisons between PM samples can be made with good precision. The average relative percent difference for all genes detected during 10 different exposures was 17.1%. Our analysis demonstrated that 71% of genes had an average signal to noise ratio (SNR) 3. Our time course study sug-
Additional Supporting Information may be found in the online version of this article.
(SCAQMD); the California Air Resources Board (CARB); the USC Viterbi School of Engineering.
Correspondence to: D. S. Antkiewicz; e-mail: [email protected]
Published online 00 Month 2014 in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/tox.21961
Contract grant sponsors: Southern California Particle Center (SCPC) funded by USEPA; the South Coast Air Quality Management District C 2014 Wiley Periodicals, Inc. V
1
2
SIJAN ET AL.
gests that 4 h may be an optimal in vitro exposure time for observing short-term effects of PM and capturing the initial steps of inflammatory signaling. The 4 h exposure resulted in the detection of 57 genes (out of 84 total), of which 86% had altered expression. Similarities and conserved gene signaling regulation among the PM samples were demonstrated through hierarchical clustering and other analyses. Overlying the core congruent patterns were differentially regulated genes that resulted in distinct sample-specific gene expression “fingerprints.” Consistent upregulation of Il1f5 and downregulation of Ccr7 was observed across all samples, while TNFa was upregulated in half of the samples and downregulated in the other half. Overall, this PM-induced cytokine expression assay could be effectively integrated into health studies and air quality monitoring programs to better understand relationships between specific PM components, oxidative stress activC 2014 Wiley Periodicals, Inc. Environ Toxicol 00: 000–000, 2014. ity and inflammatory signaling potential. V Keywords: air pollution; particulate matter; inflammation; toxicity; cytokines; gene expression; PCR array; oxidative stress; ROS; macrophages
INTRODUCTION Air pollution in the form of atmospheric particulate matter (PM) is of major concern to human health. According to the World Health Organization, PM contributes to 800,000 premature deaths annually (WHO, 2002). Ambient PM exposures have been linked to asthma, airway irritation, bronchitis, respiratory infections, reduced lung function, cardiovascular and cardiopulmonary diseases (for review see: Anderson et al., 2012; Rohr and Wyzga, 2012). Particulate matter released into the atmosphere comes from natural (e.g., volcanic eruptions, sand storms, and soil resuspension) and anthropogenic (e.g., power plant combustion, automobile exhaust, and industrial emissions) processes and has diverse chemical and physical properties. Although the chemical composition of these different sources is well studied, understanding of the differential toxicity of these and other PM sources is central to the development of effective and cost-efficient control strategies and regulations. The relative toxicity of PM is strongly dependent on the particle size and composition. Fine and ultrafine particles, PM2.5 and PM0.1, have the ability to penetrate deep into the airways, thus potentially having a greater impact (reviewed in Valavanidis et al., 2008). Several mechanisms of aerosol toxicity have been identified, but oxidative stress and resulting inflammation along with DNA and lipid oxidative damage underlie many disease processes (reviewed in Valavanidis et al., 2008). Oxidative stress is induced by reactive oxygen species (ROS), which include highly reactive compounds containing molecular oxygen, superoxide and hydroxyl radicals, as well as non-radical compounds such as hydrogen peroxide and lipid peroxides (reviewed in Lodovici and Bigagli, 2011). Components of PM that have been associated with production of ROS include organic compounds (e.g., polycyclic aromatic hydrocarbons and quinones), vehicle exhaust particles (e.g., black carbon) and transition metals including iron, vanadium and nickel (for review see MazzoliRocha et al., 2010; Lodovici and Bigagli, 2011). In recent years multiple studies have referred to the interplay of inflammation and oxidative stress in PM-induced toxicity (reviewed in Tao et al., 2003; Happo et al., 2007; Jalava
Environmental Toxicology DOI 10.1002/tox
et al., 2007; Michael et al., 2013). One of the key cells involved in PM-induced inflammation are alveolar macrophages (AMs), which are located at and recruited to the interface of air and lung tissue, and play a major role in infection control and pulmonary house-keeping by ingesting foreign particles (reviewed in Twigg, 2004). Phagocytosis of PM by macrophages, as well as stimuli from soluble PM components, and other oxidative stressors, may result in NADPH oxidase activation and superoxide production (reviewed in Iles and Forman, 2002). This oxidative burst in turn can lead to activation of signaling pathways resulting in cell proliferation, gene expression and apoptosis (reviewed in MazzoliRocha et al., 2010). The major signaling pathways activated by ROS are: NFjB, MAPK, and AP-1 (Shukla and Mossman, 2008; reviewed in Corcoran and Cotter, 2013). Inflammatory cytokines, such as TNFa, various interleukins (Il1b, Il6, & Il8), and TGFb are key signaling proteins synthesized and released by macrophages under the oxidative stress conditions (for review see Laskin and Pendino, 1995; Foster, 2001; Hoek and Pastorino, 2002). The type and quantity of cytokines released and their interaction with the receptors, regulators, and inhibitors of the inflammatory processes control the class of immune cells being activated (reviewed in Borroni et al., 2010). This ultimately directs the outcome of the inflammatory cascade. Thus, macrophages and cytokine signaling, play a key regulatory role in innate alveolar defenses as well as in the inflammatory processes in the lungs (reviewed in Stow et al., 2009) and are well fit to use as an in vitro model to explore the connections between PM-induced ROS and inflammation. The primary objective of this study was to develop a flexible and robust in vitro method for advancing our understanding of the drivers of PM toxicity that lead to oxidative stress and inflammation in mammals. We utilized a relevant rat alveolar macrophage model, applying a cell line that exhibits all the hallmarks of PM toxicity, including inflammatory signaling, resembling those seen in human primary macrophages (Hidalgo et al., 1992; Shi et al., 1996; Lane et al., 1998). The NR8383 cells have been used in many previous studies of PM-induced ROS activity (Fach et al., 2002; Riley et al., 2005; Zhang et al., 2008; Shafer et al., 2010) and of
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
cytokine-related inflammation processes (Wong et al., 2005; Haberzettl et al., 2007; Ren et al., 2011; Zeng et al., 2013). ROS was measured using a fluorescent indicator DCFHDA and a suite of inflammatory cytokines and cytokine receptors was assessed using quantitative reverse transcription polymerase chain reaction (RT-PCR) arrays. Method performance was characterized using extracts of ambient PM and engine exhaust PM samples. This novel application of the PCR Array method, as a high throughput screen of varied PM samples for their induction of inflammatory signaling molecules, tied with parallel measurement of ROS induction and up-front chemical speciation of the samples will help us gain insight into the mechanisms underlying PM-induced toxicity and inflammation.
MATERIALS AND METHODS Cell Culture The rat alveolar macrophage cell line, NR8383 (ATCC, Manassas, VA, USA) was cultured in Hams F12 medium (Sigma, Saint Louis, MO, USA) and 15% fetal bovine serum, at 37 C in a humidified 5% CO2 incubator. Cell culture was maintained by transferring non-adherent cells to new flasks bi-weekly at a floating cell concentration of 400 cells lL21 media.
PM Sample Selection The NR8383 cell line was exposed to four contrasting ambient air PM samples and two engine exhaust emission particulate samples (Supporting Information Table SI) to assess method performance and evaluate the expression levels of a large group of inflammatory cytokines in representative realworld samples. The ambient PM from Milan was collected as the fine aerosol fraction, PM2.5 (Dp < 2.5 lm). The Milan PM sample was a composite of four weekly collections from an urban setting of Milan, Italy, one of the most heavily polluted cities in Europe (Daher et al., 2012a). Engine exhaust samples were collected from a 2009 six-cylinder light duty flex fuel vehicle burning E-85 fuel (85% ethanol, 15% regular unleaded California gasoline). The vehicle was run on a chassis dynamometer using a Los Angeles adapted transient test cycle (Unified Driving Schedule) and collected over multiple runs. The two engine PM samples were collected approximately one week apart. The PM samples were analyzed using high resolution (magnetic-sector) inductively coupled plasma mass spectrometry (ICP-MS), and considerably higher levels of redox-active metals (e.g., Fe, Cr, Ni, Pb, and Pt) were measured in Engine 1b compared to Engine 1a (data not shown). The samples were collected as total suspended particulate [TSP; though the particle size of the engine exhaust PM is predominantly ultrafine ( 30, a derivative reporter value 3 C between the two peaks) to identify problematic PCR products. Only three genes (Ccl25, Il10, and Il1f5) from three separate sample exposures did not pass the screening criteria. Note: the PCR Array version we used has subsequently been upgraded and the three genes mentioned above are no longer present.
Cluster Analysis Hierarchical cluster analysis was applied to the gene fold regulation values from each sample to assist in the identification of related and coregulated genes. The clustering method used was that coded in the Cluster 3.0 software available at http://rana.lbl.gov/EisenSoftware.htm. Dendrograms were produced with Tree View software (M. Eisen;
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
5
Fig. 1. The reproducibility of the PCR Array method evaluated from duplicate experiments of ten individual PM exposures, represented as the median relative percent difference (RPD) of fold change for expressed genes (ranked from smallest to largest RPD value).
http://rana.lbl.gov/EisenSoftware.htm), and the similarity indices determined from the clustering analysis were plotted in heat-maps to visualize the outcomes. A detailed description of the clustering method used on our data set can be found in the Supporting Information. It is important to note that the heat-map outcomes do not necessarily represent significantly up or down regulated gene expression as presented in the gene expression color matrix [Fig. 4(b)].
Signal to Noise (SNR) The SNR was calculated for each gene of interest by dividing the average Ct value of duplicate exposure experiments by the standard deviation of the untreated control’s Ct values at the corresponding exposure time. All gene expression data used in the SNR analysis were normalized to the housekeeping genes. In our presentation of the SNR, we have used a ratio of 3.0 (three sigma) as our estimated limit of detection (LOD).
Method Performance Metrics Fold Change Relative Percent Difference (RPD)
RESULTS
As noted above, replicates of individual sample exposures and their respective negative controls were all processed separately (full method duplicates) and analyzed on separate PCR Arrays. The RPD of gene-specific fold change for each detected gene was calculated by taking the absolute value of the difference in fold change between the sample duplicates and dividing by the arithmetic mean of the fold changes (and multiplied by one hundred). The overall study RPD of each gene was determined by calculating the median of the gene-specific RPD values obtained across the ten separate duplicated exposures considered for this analysis (n 5 10).
Method Performance The Rat Inflammatory Cytokines and Receptors PCR Array (SABiosciences) platform was used in conjunction with the rat alveolar macrophage cell line (NR8383) and six contrasting PM samples to characterize the sample-specific inflammatory signaling. To assess the performance of our PCR Array based method, we evaluated the method’s reproducibility and sensitivity. As one measure of reproducibility, the RPD was calculated from individual gene fold changes for experimental duplicates of ten different exposures (Fig. 1 and Supporting Information Table SII). Each exposure was a
Environmental Toxicology DOI 10.1002/tox
6
SIJAN ET AL.
17.1%. Moreover, 68% of the detected genes had a median RPD value below 20%. Only two genes, Il3 and Il1f6, showed a median RPD value above 35%. Overall, the reproducibility of our method was satisfactory, suggesting that comparisons between various PM samples can be made with reasonable precision. To evaluate the sensitivity of our method, the SNR was determined and presented in Supporting Information Figure S1 for all of the detectable genes of the following samples: Riverside, USC and Engine 1a. We found that 77% of genes for Riverside, 50% for USC, and 61% for Engine 1a sample extracts were above the SNR 5 3 level for gene expression. Overall, our PCR Array based method detected the PM-induced expression of inflammatory cytokines with good sensitivity, with 71% of genes having an average SNR of 3.
Time Course Analysis of InflammationRelated Gene Expression
Fig. 2. A heat map of the Euclidean distance hierarchical clustering depicting the time-course (3, 4, 6 h) of gene expression changes (fold regulation) observed in three different samples: ambient PM, engine emission PM, and Zymosan. Each column represents a PM sample and each row represents a gene. The shades of red and green represent high and low expression levels relative to the mean fold regulation (see bar color).
complete method replicate to ensure that all the potential sources of the method’s variability were captured. The median percent RPD for all of the genes detected was
Environmental Toxicology DOI 10.1002/tox
To determine the optimal exposure time for the PCR Array method, a time course study was conducted with endpoint times of 3, 4, and 6 h. Three contrasting samples were chosen for evaluation to ensure that native variability was captured: soluble fractions of one atmospheric PM sample (Riverside) and an engine exhaust PM sample (Engine 1a), as well as a suspension of a particulate control (zymosan). Euclidean distance hierarchical clustering of the resulting fold regulation in gene expression (relative to untreated controls) was performed to visualize the gene expression patterns (Fig. 2). The 4-h exposure time point resulted in the largest number of genes with altered expression and the greatest number of upregulated genes for the ambient and engine PM samples. The Riverside sample exhibited altered expression in 33% of genes at 3 h, 69% of genes at 4 h, and 26% genes at 6 h out of the total detected genes. A substantial portion of the altered genes (36%) showed a pattern of downregulation at 3 h, followed by upregulation at 4- and 6-h exposure. Out of the total detected genes at the 4-h exposure, 57% were upregulated. The outcomes of the time course analysis conducted with Engine 1a revealed a gene expression profile similar to that observed with the Riverside sample, with the largest number of genes altered at the 4-h time point (46% of total number of detectable genes). However, while the pattern of upand downregulation was comparable to Riverside at 3- and 4-h exposure, significantly greater downregulation was observed at 6 h (50% of altered genes). In addition, the 4h exposure resulted in the largest number of upregulated genes (26% of total number of detectable genes). Interestingly, while the number of genes that are induced by the Riverside PM at 4 h was more than double that at 3- or 6h exposure, the total number of altered genes in Engine 1a exposures remained fairly similar throughout the time course.
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
In contrast to the atmospheric and engine exhaust PM samples, the yeast-derived particulate control, zymosan (which was not 0.22-lm-filtered, and exists as a colloidal/ particulate suspension) resulted in a distinct gene expression pattern (Fig. 2). Longer exposure led to progressively stronger and more prevalent upregulation of altered genes, and importantly upregulation was the super-dominant mode of expression at all time points (only eight out of 73 detected genes were downregulation throughout the time course). Overall, the ambient and engine exhaust samples exhibited the strongest upregulation and the largest number of genes altered at the 4-h time point. Based on these observations, the 4-h exposure time was chosen for the remainder of the study as the time point most likely to capture the largest number of differentially expressed genes.
Regulation of Inflammatory Signaling by Contrasting PM Samples Six contrasting PM samples, exposed for 4 h, were used to characterize the sample-specific signature of inflammatory signaling (Fig. 3). The PM mass used 4 h ranged between 64 and 302 lg mL21, with the Engine 1a sample being the lowest in mass, followed by Milan, USC, Long Beach, Riverside, and Engine 1b. Overall, across all samples, out of a total of 84 genes we detected 57 differentially expressed genes (those that met QC criteria and fell below the PCR detection cutoff Ct value of 33). A threshold of 1.7-fold change was used for gene expression to be classified as meaningfully altered. Across the six sample exposures, a total of 49 genes resulted in a meaningfully up- or downregulated expression, as shown in Figure 3. Samples varied considerably in their potential to initiate inflammatory signaling, with the Riverside sample exhibiting the highest proportion of altered genes (70% of total detected), followed by Long Beach (57%), Engine 1b (53%), Engine 1a (46%), Milan (37%), and USC (9%). Furthermore, a majority of the altered genes were found to be upregulated (66% of all altered). As shown in Figure 3, the general patterns of gene regulation in the Riverside, Milan, Engine 1a and Engine 1b PM extracts are similar. In particular, they all share a high fold regulation (30 to 55-fold) of Ccl25 and Il1r2 (refer to Supporting Information Table SII for a complete list of full gene names and their abbreviations). Furthermore, the core nine genes that were meaningfully altered in at least five out of six samples had their expression levels regulated in the same direction in the above-mentioned four samples. Three genes were found to be substantially altered in all six samples: Ccr5, Il1f5, and Il1f6. Il1f5 and Il1f6 showed significant upregulation for all six exposures. In contrast, Ccr5 was downregulated in the USC, but upregulated in all the remaining samples. On the flip-side, many gene regulation differences were observed between samples which may form the basis for a
7
sample-specific fingerprint. For example, the Riverside sample could be distinguished from the rest of the samples with its substantial upregulation of Ccl22, Il6st, Itgam, and Tnfrsf1b. The measured gene expression profiles of the Long Beach and USC samples differed from one another and from the rest of the samples. The USC sample induced an upregulated expression for Ccl2 while the other ambient samples induced a downregulated expression. Furthermore, Il1r2, Il13ra1, and Tollip showed downregulated expression in the USC sample but upregulated expression for the rest of the samples. In addition, the Long Beach sample is the only ambient sample which exhibited a downregulation of Cxcl2. The Engine 1a and Engine 1b emission samples, collected a week apart from the same engine running on E85 fuel, shared very similar gene regulation profiles (Fig. 3). Out of the 49 genes with meaningful up- and downregulation, 71% had the same direction of regulation. Some differences were however apparent (e.g., the pro-inflammatory cytokines: Cxcl1, Cxcl2 Ccl2 were downregulated in Engine 1a, but upregulated in Engine 1b) and these contrasts became more apparent in the cluster analysis (refer to section below). Overall, our method was capable of identifying commonalities and differences in inflammatory signaling regulation across various ambient PM samples.
Hierarchical Clustering of Contrasting PM Samples Another valuable tool for analyzing the similarities and differences among samples, as well as for finding patterns of gene expression is Euclidean distance hierarchical clustering. We applied this analysis to replicates of 4-h exposures of PM samples [Fig. 4(a,b)]. The 56 genes analyzed fell into seven major clusters as shown in Figure 4(a). Cluster 1 comprises of pro-inflammatory cytokines and shows high similarity for the Riverside sample in the positive direction and for the Milan sample in the negative direction. Cluster 2, also consisting of pro-inflammatory cytokines, revealed differences between the two engine exhaust samples with genes significantly upregulated for Engine 1b but downregulated or neutral for Engine 1a [Fig. 4(b)]. Interestingly, the three core genes Ccr5, Il1f5, and Il1f6, which are significantly altered in all six samples, do not cluster together. Ccr5 clusters separately from Il1f5 and Il1f6 likely due to downregulation induced by the USC sample, which was not observed for Il1f5 and Il1f6. Furthermore, a distinct cluster separate from the rest of the genes and consisting only of Il6r and Spp1 was downregulated for the Riverside, Long Beach and Engine 1b, but not for the remaining samples. Considering that Il6r is downregulated in all six samples and Spp1 is upregulated in the Milan, USC and Engine 1b only (Fig. 3), clustering revealed an association not readily observable with other data presentation methods.
Environmental Toxicology DOI 10.1002/tox
8
SIJAN ET AL.
Fig. 3. Comparison of fold regulation of genes following 4-h exposure. The dashed lines represent threshold values for gene expression upregulation (1.7 fold) or downregulation (1.7 fold). Grouping and color coding was used to represent the frequency with which a gene was significantly altered across the six samples. Gene expression significantly altered in only one of the six PM samples is indicated with an asterisk. The first three groups of genes correspond to the left y-axis and the latter three groups of genes correspond to the right y-axis.
Clustering also revealed that the Riverside and Long Beach samples shared similar expression patterns and together with the Milan and Engine 1a samples formed a larger cluster, separate from the USC and Engine 1b. In general, this observation agrees with what is presented in Figure 3, where the Riverside and Long Beach induce the same
Environmental Toxicology DOI 10.1002/tox
direction of regulation in 83% of the altered genes. However, clustering exposed contrasts in the PM-induced gene expression which were not readily apparent in Figure 3—for example, the engine emission samples (Engine 1a and Engine 1b) do not cluster together. This disparate clustering may be driven by the differential regulation of select genes.
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
9
Fig. 4. Fold regulation of inflammation-related genes induced by 4-h exposure to PM samples. A heat map of hierarchical clustering where each column represents a sample and each row represents a gene. Genes included in the analysis were expressed in three or more samples. A color matrix showing the pattern of significantly (61.7) up (red) and down (green) regulated genes, where the ordering of the genes is matched to the heat map [Fig 4(a)]. Gray corresponds to unaltered expression and black represents no signal detected.
As shown in Figure 3, out of the 49 altered genes, 14 are differentially regulated between the two engine samples (e.g., Ccl2 and Cxcl2 are upregulated in Engine 1b and downregulated in Engine 1a). Interestingly, a majority of those genes are found in Clusters 2 and 4 with high similarity in the positive direction for Engine 1b and in the negative direction for Engine 1a. Overall, clustering is a valuable tool, sensitive to direction of regulation and can present the gene data in a
unique manner, not easily observed in other data presentations.
Comparison of ROS Induction and Inflammatory Cytokine Activation To explore relationships between ROS generation and inflammation, the ROS activity of each PM sample was assessed
Environmental Toxicology DOI 10.1002/tox
10
SIJAN ET AL.
A Pearson Correlation matrix was constructed to help identify relationships between a sample’s ROS activity and the potential to regulate individual cytokine expression. Figure 5(b) shows that four genes (Ccl17, Cxcl1, Il1a and Il3) have a high positive correlation with ROS activity. Across the six samples, however, 80% of meaningfully altered inflammation-related genes had a low positive or a low negative correlation with the ROS activity. Seven genes do not have a significantly altered gene expression and correlate with ROS activity. These results, suggest that the relationship between ROS activity and inflammatory signaling is complex. However, these results are based on a relatively small data set and a larger data set may strengthen or reveal new relationships between ROS generation and inflammatory signaling.
DISCUSSION
Fig. 5. ROS activity and the 4-h exposure inflammationrelated expression relationships for the study samples. PMinduced ROS activity, in alveolar macrophages, normalized to the amount of PM (in mg, shown above each bar) used in the PCR Array experiments (sample-specific). Categorization of genes based upon a Pearson Correlation matrix of ROS activity and inflammation related gene fold regulation. Categories 1–3 represent genes which have significantly altered expression in at least one sample and Category 4 represents genes which are not significantly altered in any sample but correlate with ROS activity, the asterisk indicates that the gene has an R > 0.7. The sign and R cutoff criteria are given in the column headings.
following a 2.5-h exposure, as described previously (Landreman et al., 2008). The ROS activity of each PM sample was normalized to the corresponding total PM mass used for a given sample in the PCR Array analysis experiments. Therefore, direct comparisons can be made between ROS activity and the amount of cytokine signaling regulation produced by the same PM sample [Fig. 5(a)]. When normalized in such matter, the highest ROS activity was induced by the Riverside sample, which also had the largest magnitude of cytokine induction, both in terms of individual gene fold-change upregulation and in the proportion of genes altered. The second highest fold upregulation of gene expression was observed in the Milan sample, which was also second in the order of ROS induction. Interestingly, the Long Beach and USC samples induced a very similar ROS activity but the USC sample induced the least number of meaningfully altered genes. ROS activity induced by Engine 1a and Engine 1b is similar; however, Engine 1b generated a larger ROS activity.
Environmental Toxicology DOI 10.1002/tox
The NR8383 rat alveolar macrophage cell line is a relevant and reproducible biological model for measuring inflammatory cytokine expression and ROS production induced by atmospheric PM. When compared to primary cell cultures, NR8383 cells exhibit all the essential characteristics of primary macrophages, with appropriate response to microbial, particulate and soluble stimuli with phagocytosis and killing (Lane et al., 1998). The model cell line also exhibits strong oxidative burst with documented secretion of macrophagespecific cytokines, and replicative response to exogenous growth factors (ATCC; http://www.atcc.org/). In addition, the ROS activity of PM induced in NR8383 cells have been shown by Delfino et al. (2010) to correlate with both airway and systemic inflammation markers in a human cohort exposed to the same environmental PM. Myatt et al. (2011) have shown that upon PM2.5 exposure rat alveolar macrophages release TNFa in strong correlation with TNFa release in human dendritic cells. Our alveolar macrophage-based cytokine expression assay was found to have excellent reproducibility and sensitivity, with performance metrics at least as good as or better than those reported for other gene expression assays. The median RPD of fold changes for all detectable genes was 17.1%, with the RPD values reflecting complete method replicates, from cell exposure through the RT-PCR analysis. This inclusion of most of the potential sources of variability in our analysis likely increased the reported RPD values. Nonetheless, the exhibited RPD values are well within the range of other interassay reproducibility values, ranging from 1 to 24% coefficient of variance (CV), reported in gene expression studies (Reddy and Wilkie, 2000; Hoadley and Hopkins, 2003; Cui et al., 2006; Allen et al., 2007; Pathak et al., 2012). Even the intra-assay reproducibility values reported in literature (typically replicate wells or arrays with the same starting material), can be in similar range of up to 17% CV (Hoadley and Hopkins, 2003; Cui et al., 2006; Allen et al., 2007; Pathak et al., 2012).
METHOD FOR ASSESSING INFLAMMATORY CYTOKINE EXPRESSION INDUCED BY PARTICULATE MATTER
The manufacturer-reported sensitivity of the PCR Array that we applied is >80% positive detection with total RNA input as low as 25.0 ng and as high as 1.0 lg per array plate (Arikawa et al., 2007). In our experiments, a range of 0.34–1 lg of total RNA was available for input per RT-PCR Array analysis. The usable sensitivity of our method was further quantified by determining SNR metrics of the individual gene’s Ct values (Supporting Information Fig. S1). Using ambient and engine exhaust samples with PM masses that might typically be available from routine air pollution monitoring; we showed that 71% of the detectable genes had SNR 3. For example, we observed a high level of sensitivity for Tnf, Bcl6, Il1a, and Il1b with SNR values of 14.77, 16.51, 5.01, and 3.93, respectively. The SNR3 target has been previously considered as a LOD in other assays and is comparable or higher than SNR cutoffs used in several microarray and RT-PCR gene expression studies (Fanganiello et al., 2007; Colombo et al., 2009; Colak et al., 2012; Dao et al., 2012). Colombo et al. (2009) analyzed a large microarray data set showing that 12% of genes had a SNR1, while our Riverside data processed in a manner consistent with the above mentioned study resulted in 61% of the genes having an SNR 1 (data not shown). Interestingly, our data showed no clear correlation between the Ct value and the SNR of a gene. Overall, this novel macrophage based method for assessing PM-induced toxicity proved to be robust with a high degree of reproducibility and sensitivity. The sensitivity, as well as applicability, of a cytokine detection method is also influenced by the exposure time. PM-induced cytokine expression has been reported for treatments ranging from 2 to 72 h (Carter et al., 1997; reviewed in Becker et al., 2005; Ishii et al., 2005; Imrich et al., 2007; Mazzarella et al., 2012; Michael et al., 2013). In our time course study we were able to detect significantly altered gene expression as early as 3 h (Fig. 2), which pushes the lower-end of the range (5–24 h) suggested by some research groups as optimal to obtain measurable effects on cytokine production (reviewed in Mitschik et al., 2008). Thus, in addition to good reproducibility, the macrophage model appears to exhibit more than sufficient, if not exceptional sensitivity, for gene expression studies in the context of air pollution toxicity. Furthermore, single time point, extended exposures can fail to capture the initial induction of inflammatory signaling, as can be seen in case of Ccl25 and several interleukins in the Riverside sample (Fig. 2). Our choice of the 4-h exposure time was also dictated in part by its proximity to the optimum exposure time (2.5 h) for the measurement of PM-induced ROS (Landreman et al., 2008), thus allowing for more robust insights into the ROS-driven inflammatory gene induction. A review of the existing studies of cytokine expression revealed that the exact time course of expression will vary for each gene, cell type, as well as PM dose, PM size and composition (Pozzi et al., 2003, 2005; Ovrevik et al., 2009; Sun et al., 2012; Michael et al., 2013). There-
11
fore, to obtain meaningful results, a preliminary time course analysis of cytokine regulation should be considered for each cell exposure system and PM type. Other factors to consider in the choice of exposure time for gene expression (and therefore method appropriateness) are the viability of the test organism with extended exposure and the composition of the exposure medium. Even though a large number of studies have been conducted with exposures 24 h, it is important to note that in many of these studies, toxicity or loss of cell viability was evident at this point (Li et al., 2000; Happo et al., 2007; Ovrevik et al., 2009; Michael et al., 2013). In our method, we have purposely chosen a simple assay medium (SGM) that will not substantially alter the speciation of components extracted from the PM (e.g., not chelate metals). This is critically important, since an accurate assessment of the toxicity of PM depends upon maintaining the integrity and speciation of the constituent species. The SGM (and many similar media) is, however, not capable of supporting optimal cell health for prolonged periods of time, thus dictating shorter exposure times. Moreover, our 6-h time course study demonstrates that the 4-h exposure time point works well for detecting primary changes in the PM-induced signaling, since by 6 h we could already detect evidence of secondary feedback loops. Several published studies have shown that fine and ultrafine PM exposure can lead to inflammation (reviewed in Shrey et al., 2011). As shown in Figure 3, seven out of the core nine genes (Ccr5, Il1f6, Ccl25, Ccl5, Cxcl10, Cxcr2, and Il8ra), which were upregulated in all the PM samples evaluated have pro-inflammatory characteristics (Menten et al., 2002; Cheong et al., 2006; Matheson et al., 2006; Dinarello, 2009; Schmutz et al., 2010; Skuljec et al., 2011; Musah et al., 2012). Furthermore, the Riverside, Milan, Engine 1a and Engine 1b samples induced a 30- to 49-fold increase of Ccl25. The Riverside sample also induced significant upregulation of Ccl22, Il6st, Itgam, and Tnfrsf1b, which have been shown to have pro-inflammatory characteristics (reviewed in Heinrich et al., 2003; Solovjov et al., 2005; Semkova et al., 2011; Lilly et al., 2012). In addition, the Engine 1b sample caused considerable upregulation of Ccl2 and Cxcl2, both considered pro-inflammatory chemokines (Wolpe et al., 1989; reviewed in Melgarejo et al., 2009). Conversely, we also observed a substantial upregulation of Il1r2, which acts as a decoy receptor for proinflammatory cytokines Il1a and Il1b, suggesting an anti-inflammatory effect (Dinarello, 2009). Out of the core nine cytokines, Il1f5 and Bcl6 have anti-inflammatory functions (Dinarello, 2009; Gongol et al., 2013). Overall, the method provides a powerful and relevant approach for assessment of a wide array of both pro- and anti-inflammatory cytokines from small masses of aerosol PM. Our analysis of PM-induced cytokine regulation revealed that the water extracts of various PM samples induced a predominantly pro-inflammatory response in alveolar macrophages. Importantly, outside of the conserved
Environmental Toxicology DOI 10.1002/tox
12
SIJAN ET AL.
upregulation of the core inflammatory cytokines, each of the PM samples tested could be differentiated from the others by some unique aspect of the gene expression profile (Fig. 3). However, our study had a relatively limited number of replicate experiments and a larger data set may reveal new information in the gene expression profiles. Some of the expression profile differences observed in our samples may partly be due to differences in the PM sizefraction studied. The ambient air samples from the LA Basin were collected as the quasi-ultrafine (PM0.25), while the Milan sample was collected as the fine fraction (PM2.5) and the engine PM samples nominally
