Using a t-test, we identified 89 genes differentially expressed between HBEC exposed to PM_2.5 and vehicle control (Table 1). These genes could clearly distinguish primary HBEC exposed to PM_2.5 from the HBEC in control. Of the 89 genes, 38 genes were significantly up-regulated and 51 genes were remarkably down-regulated.

57 human target proteins of curcumin (CID: 969516) were obtained from the PubChem database by PubChem Promiscuity online and identified by UniProt protein IDs (Table 2).

Based on the differentially expressed genes in Table 1 and human target proteins in Table 2, two biological networks showing protein-protein interactions were constructed. The two protein-protein interaction (PPI) networks were visualized using Cytoscape. The nodes represented proteins in the PPI network and the edges represented the biological relationship between two nodes. There were 1,962 nodes and 15,455 edges in the PPI network of HBEC exposed to PM_2.5 (Supplementary Material Figure S1), and 1,284 nodes and 11,541 edges in the PPI network of human target proteins of curcumin (Supplementary Material Figure S2). Appling the function “Intersection” of the Advanced Network Merge plugin in Cytoscape, we found the common proteins and relationships (common network) in the two PPI networks. The common network had 1,197 nodes and 9,521 edges (Figure 1).

The top five functions of the common network and the number of proteins associated with each function were found using Ingenuity Pathways Analysis (IPA). The most significant biological functions were grouped into three categories: (1) Diseases and Disorders, (2) Molecular and Cellular Functions, and (3) Physiological System Development and Function (Table 3).

Table 4 lists the top five canonical pathways associated with the common network as calculated by IPA (Figure 2, Figure 3, Figure 4 and Figure 5, Supplementary Material 3). Calculation was either according to ratio (the number of genes from the data set that map to the canonical pathway in question divided by the total number of proteins that map to the same canonical pathway) or significance (Fischer’s exact test was used to calculate a P-value determining the probability that the association between the proteins in the dataset and the canonical pathway was explained by chance alone).

To partially validate the pathways listed in Table 4, we measured the expression of NF-kappaB p65 and IL-6 in human bronchial epithelial cells (16HBE) exposed to PM_2.5. 16HBE were pre−treated with 10, 20, 40 μM curcumin for 30 min followed by exposure to PM_2.5 (250 μg/mL) for 24 h in the presence or absence of curcumin. After 24 h, cells were collected and measured for NF-kappaB p65 and IL-6 expression by Western blot. Notably, NF-kappaB p65 or IL-6 expression level was markedly increased in 16HBE exposed to PM2.5 as compared with the control cells. However, curcumin treatment could attenuate the high expression of NF-kappaB p65 or IL-6 in cells induced by PM_2.5 (Figure 6).

Predictive analysis was a general method for predicting the accuracy of quantitative experiments. The use of predictive analysis allowed the designer of an experiment to estimate the accuracy that should be obtained from the experiment before the experimental setup was finalized [6]. Until now, there had been no in vivo or in vitro reports about the effects of curcumin on organisms exposed to PM_2.5; therefore, we collected limited data associated with PM_2.5 or curcumin available from online databases such as GEO and PubChem. Because the aim of our study was to outline the potential biofunctions and pathways associated with the effect of curcumin on HBEC exposed to PM_2.5 predictively, we did not restrict all data reanalyzed at identical molecular level.

PubChem [7] is a public repository for biological properties of small molecules hosted by the US National Institutes of Health (NIH). The PubChem BioAssay database contained biological test results for more than 700,000 compounds. From the PubChem BioAssay database, we could retrieve the target proteins of compounds [8]. In our study, 57 human target proteins of curcumin (CID: 969516) were obtained.

PPI were extremely important cellular events that affected many of the most important molecular processes in the cell, such as DNA replication. They formed the basis for many signal transduction pathways and transcriptional regulatory networks. The availability of complete and annotated genome sequences of several organisms had led to a paradigm shift from the study of individual proteins in a cell to proteome−wide analysis in an organism. The whole proteome analysis had illustrated that PPI affected cellular biological functions through many orchestrating networks such as metabolic, signaling and regulatory pathways in an organism [9].

Within the airway, the epithelium forms the mucosal immune barrier, the first structural cell defense against common environmental insults such as microorganisms and particulate matter. Hence, respiratory infectious diseases share similar pathologic processes such as the inflammatory response or oxidative stress with bronchial diseases induced by PM_2.5 [10,11,12]. The inflammatory response was the main acute effect induced by PM_2.5 in the respiratory tract, a target organ of PM_2.5. In vitro studies had shown that airway epithelial cells responded to PM_2.5 exposure by the release of inflammatory cytokines such as IL-1beta, TNF-alpha, and IL-6 [13], chemokines such as IL-8 [14], and erythropoietic cytokines such as G-CSF and GM-CSF [15,16]. Because curcumin was observed to inhibit secretion of the pro-inflammatory cytokines NF-kappaB mediating in HBEC exposed to pollutants [17,18], we predicted that curcumin might also have an anti-inflammatory effect on HBEC exposed to PM_2.5.

Some researchers conducting large epidemiological cohort studies in the United States and Europe had comfirmed the relationship between long-term exposure to particulate air pollution (PM₁₀ and PM_2.5) and increased mortality from lung cancer, especially in combination with other known risk factors, such as smoking, passive smoking, and occupational exposure [19,20]. By contrast, curcumin, a natural antitumor compound, had been shown to have the effect of inhibiting lung cancer cell invasion and metastasis in several studies [21,22,23] and have promising potential as a diet-derived cancer chemopreventive agent [24]. Thus, we inferred that curcumin could inhibit the carcinogenesis of airway epithelial cells resulting from PM_2.5 exposure.

Generally, PM_2.5 led to the proliferation inhibition and apoptosis of HBEC [25,26,27]. PM_2.5 could induce cell cycle arrest in G1 phase, inhibit DNA synthesis, and block airway epithelial cell proliferation [28]. The P53 pathway, tumor necrosis factor-alpha (TNF-alpha) pathway, and mitochondrial pathway played critical roles in the apoptosis processes induced by PM_2.5 [29,30]. However, as a dietary antioxidant, curcumin had been proven to have preventive potential against apoptosis induced by peroxide or cigarette smoke extract in HBEC though inhibition of NF-kappaB [17,31]. Moreover, curcumin was a selective apoptosis modulator. For most noncancerous cells, curcumin was a protector and prevents cells from apoptosis induced by various adverse factors, but for cancer cells, curcumin was a killer and arrested cell cycle, inhibited cell proliferation, and/or caused apoptosis. For example, when mammary epithelial cells and breast cancer cells accumulated a similar amount of curcumin, a significantly higher percentage of apoptotic cells was induced in cancer cells compared to epithelial cells [32]. Similarly, we speculated that curcumin might have a two-way regulating effect on HBEC when exposed to PM_2.5.

Glucocorticoids (GCs) could control airway inflammation in respiratory diseases such as chronic obstructive pulmonary disease (COPD) and asthma, and the airway epithelium was a primary target of GC anti-inflammatory actions [33]. GC effects were mediated through the GC receptor (GR). Previous studies had indicated that cultured HBEC from smokers possess GRs with a lower binding affinity, and this might result from the inflammation found in the airways in smokers [34]. Although there had been no studies involving the effect of PM_2.5 on GRs in HBEC, cigarette smoke and PM_2.5 shared a similar inflammatory effect on HBEC, and we could speculate that PM_2.5 might decrease GR signaling. In addition, GR action was shown to be tightly regulated by histone deacetylase 2 (HDAC2), which suppressed inflammatory gene expression in inflammatory airway disease [35]. Acting as an HDAC activator, curcumin was found to restore HDAC2 activity, thereby restoring the function of the GR [36]. In summary, regulation of the GR pathway was a possible mechanism by which curcumin inhibits the hazardous effects of PM_2.5.

Recent studies have suggested that numerous components of phosphoinositide 3-kinase (PI3K)-dependent signaling, mediated by Akt kinase, played a crucial role in the expression and activation of inflammatory mediators, inflammatory cell recruitment, immune cell function, airway remodeling, and corticosteroid insensitivity in chronic inflammatory respiratory disease [37], especially in COPD and asthma [38]. PM_2.5 or cigarette smoke could induce activation of the PI3K/Akt pathway in HBEC and promote transcription of downstream inflammatory mediators [39,40,41]. However, studies had proved that curcumin could inhibit PI3K/Akt/NF-kappaB signals in human lung epithelial cells [42], block Akt translocation to the nucleus and further decrease inflammation in human tracheal smooth muscle cells [43]. Therefore, we predicted that curcumin also might have potential to prevent HBEC from the toxicity effects of PM_2.5 by modulating PI3K/Akt signaling.

Recent research indicated that the NF-kappaB/IkappaB pathway played an important role in the inflammatory response induced by PM_2.5 in the lung [44]. The activation of the NF-kappaB/IkappaB complex preceded cytotoxicity or inflammation in PM_2.5-exposed human bronchial or lung epithelial cells through the reactive oxygen species (ROS)-dependent NF-kappaB pathway [45,46]. As an inhibitor of NF-kappaB, curcumin exhibited a potent anti-inflammatory effect, and could decrease the airway epithelial cell inflammatory cytokine response to the pollutant cadmium or cigarette smoke extract [17,18]. Like cadmium and cigarette smoke, PM_2.5 was also a pollutant in the environment, so we hypothesized that curcumin might perform its anti-inflammatory effect on PM_2.5 by inhibiting the NF-kappaB pathway.

14-3-3 family members tightly regulated cell fate through interaction with a wide spectrum of proteins that were targeted by various classes of protein kinases [47]. 14-3-3 proteins played particularly important roles in coordinating the progression of cells through the cell cycle, regulating their response to DNA damage and influencing life−death decisions [48]. Studies reported that 14-3-3 might contribute to lung tumorigenesis. In H322 cells, over-expression of 14-3-3 protein resulted in abnormal DNA replication and polyploidization [49], and in A549 cells, 14-3-3 promoted cellular proliferation [50]. Other studies found that curcumin could induce the typical features of apoptosis and inhibited the expression of 14-3-3 in HT-29 cells [51]. Based on the evidence mentioned above, we predicted that curcumin might prevent HBEC exposed to PM_2.5 from carcinogenesis by inhibiting the 14-3-3 pathway.

In the NF-kappaB signaling pathway, NF-kappaB played a pivotal role as inflammatory response regulator, and IL-6 was an important inflammatory factor regulated by NF-kappaB and caused the damage response of PM_2.5 [52]. Therefore, in a validating experiment, we selected NF-kappaB p65 and IL-6 as validated molecules and found that curcumin treatment could attenuate the high expression of NF-kappaB p65 or IL-6 in cells induced by PM_2.5. These results supported our part prediction.

A microarray dataset (accession number GSE7010) [53] was downloaded from the GEO [54], and analyzed it based on the Affymetrix Human Genome U133A Array. This dataset was derived from a study observing global gene expression in HBEC and identifying cellular pathways associated with coarse, fine and ultrafine particulate matter exposure. Ambient PM was collected in three different size fractions from Chapel Hill air; particles were extracted from foam or filter matrices and lyophilized. Primary HBEC were exposed to PM_2.5 at 250 μg/mL or vehicle control for 6 h in culture [55]. In this study, we used three samples from the control group (GSM161787, GSM161793, GSM161798) and three samples from the fine particulate matter (PM_2.5) group (GSM161790, GSM161796, GSM161801). Probes showing differential expression were extracted by volcano plot analysis with the filtering criteria of a 1.5-fold change using GeneSpring GX version 11.0 after per chip and per gene normalization.

The human target proteins of curcumin (CID: 969516) in PubChem [56] were retrieved using PubChem Promiscuity [57] online [58] with the filtering criteria of “not less than one Active Bioassay”.

PPI represented a basic blueprint for the analysis of self-organization and homeostasis in living organisms [59]. In this study, a Cytoscape [60] plugin, BisoGenet [61], was applied for assembling the PPI network. Information on human PPI networks involving relevant genes was obtained from various databases, including HPRD (Human Protein Reference Database), BIND (Biomolecular Interaction Network Database), BioGRID (The General Repository for Interaction Datasets), DIP (Database of Interacting Proteins), IntAct (Database system and analysis tools for protein interaction data), and MINT (Molecular Interactions Database). Two PPI networks were constructed based on the differential expression of genes from microarray data analysis and the target proteins of curcumin from PubChem. Another Cytoscape plugin, Advanced Network Merge, was used to find the common proteins and relations (common network) in the two PPI networks.

For further analysis, a data file was uploaded into IPA (Ingenuity® Systems, www.ingenuity.com, Redwood City, CA, USA). This file contained the proteins in the common network. Each protein identifier was mapped to its corresponding protein object in the Ingenuity Pathways Knowledge Base (IPKB). The functional analysis identified the biological functions and/or diseases that were most significant to the data set. Proteins from the data set that met the P-value threshold of 0.05 (Fisher’s exact test) and were associated with biological functions and/or diseases in the IPKB were kept for analysis. Canonical pathway analysis identified the pathways most significant to the data set, based on two parameters: (1) a ratio of the number of proteins from the data set that map to the pathway divided by the total number of proteins that map to the canonical pathway and (2) a p-value calculated with Fisher’s exact test determining the probability that the association between the proteins in the dataset and the canonical pathway is explained by chance alone.