Abstract
Rationale: The molecular mechanisms involved in airway oxidative stress responses reported in healthy smokers and in those with chronic obstructive pulmonary disease (COPD) are poorly understood.
Objectives: To assess the expression of genes involved in oxidative stress responses in the bronchial epithelium of smokers with or without COPD and in relation to disease severity.
Methods: Global gene expression was assessed in bronchial brushings in 38 subjects with COPD, 14 healthy nonsmokers, and 18 healthy smokers.
Results: Gene expression analysis using Affymetrix arrays revealed mRNAs representing 341 out of 642 oxidative stress genes from two predefined gene sets to be differentially expressed in healthy nonsmokers when compared with healthy smokers, and 200 differentially expressed oxidative genes in subjects with COPD when compared with healthy smokers. Gene set enrichment analysis showed that pathways involved in oxidant/antioxidant responses were among the most differentially expressed gene pathways in smoking individuals, with further differences seen in COPD. Distinct, nonlinear gene expression patterns were identified across the severity spectrum of COPD, which correlated with the presence of certain transcription factor binding sites in their promoters. Significant changes in oxidant response genes observed in vivo were reproduced in vitro using primary bronchial epithelial cells from the same donors cultured at an air–liquid interface and exposed to cigarette smoke extract.
Conclusions: Cigarette smoke induces significant changes in oxidant defense responses; some of these are further amplified, but not in a linear fashion, in individuals who develop COPD.
Knowledge on gene expression profiles of the bronchial epithelium of COPD patients has been limited, particularly with respect to oxidative stress responses.
This study identifies a set of oxidative stress response genes expressed in the bronchial epithelium in COPD as well as transcription factors involved in their regulation.
This study sought to improve the understanding of molecular mechanisms that regulate oxidant/antioxidant responses in the airways of smokers by investigating the expression of genes in the bronchial epithelium that might be involved in these responses. The epithelium is the first anatomic site exposed to cigarette smoke. We have hypothesized that epithelial expression of genes involved in oxidant/antioxidant responses in smokers is markedly different from that in nonsmokers and that additional differences are seen in smokers who have developed COPD. We have also hypothesized that the altered gene expression is related to the severity of COPD and might thereby be a determinant of disease progression. Furthermore, we have sought to establish whether there is any association between the expression of oxidant responsive genes and the clinical and pathologic outcomes of COPD, including measures of lung function and epithelial content of mucus. Finally, we have sought evidence of overexpression of transcription factors (TFs) in smokers with COPD, with the knowledge that some have been described as being associated with oxidant stress response genes, with the hope that further novel TFs may be found to be overexpressed, thus offering new targets for treatment.
A number of studies have applied DNA microarray technology to investigate gene expression in the lungs (8–12). In one of these studies, which also focused on gene expression in the airway epithelium rather than the whole lung (8), 16 antioxidant-response–related genes were found to be up-regulated in healthy smokers (HS) when compared with nonsmokers (NS). The extent to which these epithelial genes play a role in COPD is unclear because only 15 to 20% of smokers develop COPD (2). We have, therefore, extended our investigations to the complex clinical syndrome of COPD, by analyzing gene expression in bronchial epithelial brushings from volunteer subjects with COPD ranging from mild to severe, categorized according to detailed clinical and physiologic criteria and high-resolution computed tomography (HRCT). To establish a direct link between gene expression and cigarette smoke exposure, fully differentiated epithelial cells obtained by primary culture at air–liquid interface (ALI) were exposed ex vivo to cigarette smoke extract (CSE) and the effects of smoke extract exposure on global gene expression were evaluated using microarray analyses.
Seventy subjects, characterized as previously described (13), underwent bronchoscopy, endobronchial biopsy, and bronchial brushings, and provided samples of epithelial cells of greater than 95% purity. There were 14 healthy NS, 18 HS, and 38 smokers with COPD (Table 1). All subjects underwent HRCT, as previously described (13), to ensure that the HS did not have emphysema that could not be detected by lung function testing. The subjects with COPD were further classified according to GOLD (Global Initiative for Chronic Obstructive Lung Disease) criteria (14) into the following stages: stage 0 (n = 18), stage 1 (n = 7), and stages 2 to 4 (n = 13) (see online supplement); all had a chronic productive cough. The smoking history was slightly longer in subjects with COPD than in HS, but this was not statistically significant.
Parameter | NS | HS | COPD Stage 0 | COPD Stage 1 | COPD 2–4 |
|---|---|---|---|---|---|
| Number Female/male | 14 9/5 | 18 10/8 | 18 11/7 | 7 2/5 | 13 4/9 |
| Age, yr | 54 (40–64) | 44 (26–63) | 51 (40–64) | 59 (50–65) | 54 (43–64) |
| FEV1, % predicted | 105 (92–117) | 104 (83–128) | 99 (76–132) | 93 (85–101) | 61 (25–79) |
| FEV1/FVC, % predicted | 74 (67–86) | 79 (69–90) | 76 (70–82) | 66 (60–70) | 58 (30–69) |
| Smoking, pack-years | 0 (0–0) | 33 (10–59) | 52 (19–160) | 40 (30–53) | 57 (30–86) |
| TlCO, % predicted | 81 (61–100) | 66 (38–91) | 63 (38–89) | 60 (41–84) | 59 (32–87) |
| Total SGRQ | 6 (0–39) | 7 (0–17) | 19 (0–45) | 26 (3–42) | 34 (12–67) |
The study was approved by the Southampton University and General Hospitals ethics committee and volunteers gave their written, informed consent.
All smokers were instructed to smoke their last cigarette before retiring to bed and refrain from smoking on the morning of bronchoscopy to maintain consistency of time from last exposure to cigarette smoke across all subject groups. Bronchoscopy was always performed between 8:30 and 9:30 a.m. Epithelial cells were obtained by bronchial brushings, taking care not to cause bleeding and contaminate samples with blood cells. The samples were processed in Trizol for microarray analysis or placed into cell culture. Cytospins were made for differential counts and phenotype analysis by immunocytochemistry using monoclonal antibodies specific for MUC5AC for goblet cells, cytokeratin 13 for basal epithelial cells, cytokeratin 18 for columnar epithelial cells, and tubulin for ciliated cells.
During a separate bronchoscopy, four endobronchial biopsies were taken for immunohistochemical analysis for MUC5AC, ErbB1, ErbB2, and ErbB3, as previously reported (15), to seek associations between gene expression, epithelial mucus content, and expression of Erb receptors, which are believed to be involved in mucus production (see also the online supplement). In an attempt to correlate gene expression with translation of its protein, immunohistochemistry was also performed for CYP1B1 (Alpha Diagnostics, San Antonio, TX), a key participant in oxidant stress responses.
Epithelial cells obtained from four NS, one HS, four subjects with stage 0 COPD, and two subjects with stage 2 COPD were cultured in bronchial epithelial growth medium until 60 to 80% confluent, passaged twice, and ALI cultures were established. Stimulation with 20% CSE was performed after 19 to 21 days of culture when cells were processed for microarray analysis. Further details are provided in the online supplement.
Nonlinear normalization (16) and probe set reduction (17) were used to obtain gene expression data. The samroc test, a modified t test method (18), was used to assess the degree of differential expression. Detection p values were calculated by the Affymetrix Microarray Suite 5.0 (Affymetrix, Santa Clara, CA) (19). The proportion of samples where the detection p value falls below 5% is referred to as the present rate for any individual gene; this provides a measure of the certainty of the gene expression. To account for multiple tests, false discovery rates (FDRs) were obtained (these are reported in the online supplement) (20).
Gene set enrichment analysis (GSEA) (21) was used to rank pathways (gene sets) in terms of coupling to a biological condition. Gene sets were assessed as to whether they individually scored high when compared with other possible choices of gene sets. This provided an unbiased means of assessing pathways with respect to enrichment or degree of representation of highly regulated genes. For the GSEA, public gene sets were downloaded from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (22) and Biocarta websites. In addition, two sets of genes, a “manual” set and an “automated” set, were generated to reflect the participation of genes in pathways involving common metabolic or signaling processes. The manual set (80 genes) was selected by mining the literature for oxidant response genes. The automated gene set (608 genes) was created using an unbiased criterion in terms of InterPro sequence domains (i.e., the presence of at least one of 199 InterPro sequence domains characteristic of oxidoreductases, defined as Enzyme Classification number 1.-.-.- [23]), and the following additional domains often involved in oxidant responses: metallothioneins, glutaredoxins, Acyl-CoA dehydrogenases, and thioredoxins. For more details on gene sets, see the online supplement. To test the hypothesis that common regulatory mechanisms underlie common gene expression patterns, the algorithm of Elkon and colleagues (24) was used to find TF binding sites that are overrepresented in selected sets of gene promoter regions. A set of 25,000 putative promoter sequences from AstraZeneca's internal Gene Catalogue (25) was used as the background promoter set. The Elkon method was used on two gene clusters defined by expression patterns. For comparisons of genes by similarity of promoter motifs, a standard two-dimensional clustering function in the Spotfire DecisionSite for Functional Genomics module was used (Spotfire, Inc., Cambridge, MA). A binary matrix was used (1 for presence of a TF site in a promoter, 0 for absence). Presence of a TF site was assessed as described in Reference 24.
Principal components analysis (PCA) provided an unbiased and unsupervised overview of data. Partial least squares (PLS) and the PLS–discriminant analysis (PLS-DA) methods were used to establish relationships between clinical variables and gene expression and to identify readouts of clinical severity category (26). For further details, see the online supplement.
Brushings were analyzed with respect to bronchial epithelial cell subtype composition using molecular markers. The numbers of brushed epithelial cells staining positively for MUC5AC were increased in subjects with COPD when compared with NS (p < 0.0001), and in HS when compared with NS (p < 0.0001), with no difference between HS and subjects with COPD (p = 0.53). There were no differences in numbers of epithelial cells positive for cytokeratins or tubulin.
Samples from 14 healthy volunteers, 18 smokers without signs of COPD, and 38 patients with COPD in different disease stages fulfilled the criterion of 95% pure epithelial cells; the RNA from these cells was run individually on DNA microarrays.
Of the 44,929 probe sets present on the Affymetrix U133A and B arrays, 18,702 probe sets (13,755 different genes) detected corresponding mRNAs as present in more than 20% of all samples. This cutoff was used as recommended by Li and Wong (27) to ensure detection of a very selective expression pattern and omit genes with uncertain expression.
When comparing gene expression on arrays, 7,141 of the probe sets were differentially expressed when comparing HS and NS (p < 0.05); of these, 2,291 probe sets (2,133 genes) were up-regulated and the rest (4,850 probe sets, 3,900 genes) were down-regulated in HS. When comparing smokers with and without COPD, expression levels of 3,981 probe sets were significantly different, with 2,863 probe sets (2,360 genes) being up-regulated and 1,118 probe sets (1,074 genes) being down-regulated in subjects with COPD.
Our primary analysis focused on two predefined gene sets called the manual and automated sets, consisting of genes which are considered to be associated with oxidant/antioxidant responses. A 5% p value threshold applied to these sets corresponds to an FDR of less than 3.5% in the COPD versus HS comparison and an FDR of less than 12% in the HS versus NS comparison (for summary of expression data for the two gene sets, see Tables E1 and E2 in the online supplement). Subanalysis of the manual set of genes showed these to be among the most differentially expressed: 22 genes were up-regulated and 11 down-regulated when HS were compared with NS, and 40 genes were up-regulated and 1 down-regulated when comparing smokers with and without COPD (see the online supplement). In terms of FDR, 81 probe sets of 121 in this gene set were significant at a 10% FDR level when comparing HS and NS, whereas 45 of 121 fell below that level in the COPD versus HS comparison. Using an unbiased method for identifying pathways or processes overrepresented in gene lists, GSEA, the manual oxidant responsive gene set was found to be overrepresented in differentially expressed genes, both when comparing HS with NS and HS with subjects with COPD (Table 2). The automated set of all oxidoreductases was overrepresented only in the comparison between HS and NS (Table 2).
Gene Set | Database Identifier | GSEA p Value | Source of Gene Set | |||
|---|---|---|---|---|---|---|
| Comparison of NS and HS | ||||||
| Ribosome | Map03010 | 0.0009 | KEGG | |||
| Automated set, subset: metallothioneins* | None | 0.0016 | Authors | |||
| Role of eosinophils in the chemokine network of allergy | H_eosinophilsPathway | 0.0023 | Biocarta | |||
| Automated set (full set)* | None | 0.0042 | Authors | |||
| Manual set (oxidant responsive)* | None | 0.0126 | Authors | |||
| Fibrinolysis pathway | H_fibrinolysisPathway | 0.0130 | Biocarta | |||
| Comparison of COPD and HS | ||||||
| Oxidative phosphorylation | map00190 | 0.000489 | KEGG | |||
| Manual set (oxidant responsive)* | None | 0.002027 | Authors | |||
| ATP synthesis | Map00193 | 0.002566 | KEGG | |||
| Proteasome | Map03050 | 0.002604 | KEGG | |||
| Automated set, subset: thioredoxins* | None | 0.005926 | Authors | |||
| Glycolysis pathway | h_glycolysisPathway | 0.012183 | Biocarta | |||
Some of the subsets of the automated oxidoreductase gene sets were also prevalent among the pathways that were significantly differentially expressed: comparison of NS and HS showed enrichment of metallothioneins and aldo-keto reductases (Table 2), whereas comparison between HS and subjects with COPD showed enrichment of thioredoxins (Table 2). In addition, there was differential expression of pathways associated with basal energy metabolism, oxidative phosphorylation, ATP synthesis, and free radical–induced apoptosis when comparing subjects with COPD and HS.
Detailed analysis of individual genes from the automated set (Figure 1) showed many genes to be up-regulated in both COPD compared with HS and HS compared with NS. In this analysis, the degree of up-regulation (measured as ratios of mean expression values) of genes up-regulated in HS compared with NS was higher than in subjects with COPD compared with HS. The most prominent changes in both comparisons were in the group of aldo-keto reductases and cytochromes P450 (Figure 1). In terms of FDR, 195 probe sets of 1,095 in this gene set were significant at a 10% FDR level when comparing HS and NS, whereas 180 fell below that level in the COPD versus HS comparison.
Figure 1. Analysis of expression of oxidoreductases, identified by similarities in sequence motifs (defined in the automated gene set as stated in Methods). The ratios of mean gene expression values subjects with chronic obstructive pulmonary disease (COPD)/healthy smokers (HS) and HS/nonsmokers (NS) are shown on the x and y axes, respectively, with values more than 1 indicating up-regulation. The main categories of oxidoreductases shown are as follows: aldo-keto reductases (red), cytochromes P450 (dark blue), glutathione peroxidases (yellow), metallothioneins (green), short-chain dehydrogenases (light gray), thioredoxins (magenta). Genes for which both differences in expression (HS vs. NS and COPD vs. HS) are statistically significant are shown as large squares; others are shown as small light-gray squares. For some genes, several expression values can be seen, reflecting the fact that Affymetrix chips contain several probe sets per gene for some genes and that, for some of those genes, differential expression was detected by more than one probe set.
[More] [Minimize]Ratios of expression values, HS/NS and COPD/HS, of differentially expressed genes from the manual set are displayed in Figure 2A. There was no correlation between genes differentially expressed in the two comparisons (i.e., any increase in gene expression in HS relative to NS was not followed by a similar up-regulation when comparing subjects with COPD and HS). More detailed examination of expression patterns across the disease severity spectrum for the oxidant-responsive genes revealed two distinct expression patterns as shown in Figures 2B and 2C. The most common expression pattern (Figure 2B) was characterized by higher expression among HS when compared with NS, and even higher in subjects with stage 0 COPD, with peak expression in subjects with stage 1 COPD (Figure 2B). In subjects with more advanced disease, expression was similar to stage 0 COPD. This pattern was seen for 13 of 15 genes with ratios of HS/NS and COPD/HS greater than 1. For genes with ratios of HS/NS less than 1 and COPD/HS greater than 1, the most common expression pattern was one of lower expression in HS when compared with NS, higher expression in patients with stage 0 COPD compared with HS, and even higher in patients with stage 1 COPD. Gene expression for these genes was also lower in subjects with stages 2 through 4 COPD compared with those with stage 1 COPD (Figure 2C). An exception to this pattern was the expression of three probe sets for superoxide dismutase 2 (Figure 2C), with lower expression in COPD stage 1 when compared with COPD stage 0.
Figure 2. Ratios of average expression of oxidant-responsive genes identified in the literature (manual set). (A) Scattergraph of ratios of average gene expression in disease groups as indicated on the axes. Genes for which both differences in expression (HS vs. NS and COPD vs. HS) are statistically significant are shown as large squares; other genes are shown as small light-gray squares. (B) Normalized median gene expression values for five subject categories (HS, NS, and COPD stages 0, 1, and 2–4). Typical gene expression profiles across disease stages for genes that are up-regulated in HS when compared with NS and further up-regulated in COPD versus HS (HS/NS and COPD/HS > 1). Note that the expression of all genes peaks in COPD stage 1 (COPD1), except for MGST1 (peaks in COPD0) and DUOX2 (peaks in COPD2–4). Genes shown in B: AKR1B1, AKR1C3, DUOX2, G6PD, GCLC, GCLM, GPX2, GPX3, GSTM3, KIF1B, MGST1, PRDX1, TALDO1, TXN, TXNRD1. (C) Normalized median gene expression values, as in B, but for genes that are down-regulated in HS relative to NS but up-regulated in smokers with COPD compared with HS (HS/NS < 1 and COPD/HS > 1). Three probe sets for superoxide dismutase 2 (SOD2) have a pattern that is distinct from the other probe sets, peaking in COPD stage 0. Although other genes in this cluster, CP, MGST2, AKR1A1, are peaking in COPD stage 1. Genes shown in C: AKR1A1, CP (2 probe sets), MGST2, P5 (2 probe sets), PRDX3, SOD2 (3 probe sets).
[More] [Minimize]It has recently been shown that common expression patterns can be traced back to common sequence motifs in their promoter regions (24). We used a similar approach to search for TF binding sites overrepresented in promoter regions of oxidant response genes that exhibit similar smoking- or COPD-related changes in expression. Two gene clusters were selected from the automated set depending on whether they were both up-regulated in COPD/HS and in HS/NS comparisons (cluster 1) or up-regulated in the first comparison and down-regulated in the second (cluster 2). Among the significantly overrepresented TF sites in the promoters of genes from the first cluster were sites for activator protein-1 (AP-1) and p53 (see Table 3). In the second cluster, the overrepresented TF sites included E2F, nuclear factor (NF)-κB, and FOXO1. The corresponding TFs are known to be involved in responses to oxidative stress (28–31). Interestingly, of 31 TF sites listed in Table 3, 10 have at least one binding TF significantly regulated when comparing COPD with HS, or HS with NS, or in both comparisons.
TF Site | TF Site Accession (Transfac) | p Value | ||
|---|---|---|---|---|
| Promoters of genes that are down-regulated in smokers (HS/NS) and also up-regulated in COPD (COPD/HS) | ||||
| Oct-1 | M00136 | 0.0012 | ||
| E2F | M00425 | 0.0099 | ||
| NF-κB | M00054 | 0.011 | ||
| FOXO4 | M00472 | 0.013 | ||
| Nrf2 | M00821 | 0.015 | ||
| c-Myc/Max | M00322 | 0.016 | ||
| GR | M00921 | 0.016 | ||
| Promoters of genes that are up-regulated both in smokers (HS/NS) and in COPD (COPD/HS) | ||||
| Pax | M00808 | 0.0003 | ||
| p53 | M00761 | 0.0016 | ||
| AP-2 | M00189 | 0.0023 | ||
| HNF-4 | M00764 | 0.0031 | ||
| p53 | M00272 | 0.0032 | ||
| Nrf2 | M00821 | 0.0051 | ||
| AP-1 | M00172 | 0.0093 | ||
| COUP-TF | M00158 | 0.010 | ||
| Lhx3 | M00510 | 0.011 | ||
| NF-AT | M00935 | 0.017 | ||
| AP-2α | M00469 | 0.018 | ||
The genes from clusters 1 and 2 and the significantly overrepresented TF sites were subjected to two-dimensional clustering (see Methods) by similarity of the promoter motifs. Some combinations of TF binding sites were seen as fingerprints in some groups of genes (Figure 3A) (e.g., p53 and Nrf2). There was a trend for such gene groups, clustered by TF sites, to show similar expression patterns across different disease stages (Figures 3B and 3C). Clear-cut expression patterns in Figures 3B and 3C are striking, because the small gene clusters presented there were defined solely by their TF binding site patterns. Thus, the presence of TF binding sites alone allowed separation of genes into the original expression clusters, although the separation was not perfect, with one gene (AKR1B10) not following the pattern seen for other genes in that cluster and rising already in HS as compared with NS.
Figure 3. Distinct gene expression patterns (taken from the automated gene set) across disease stages correlate with distinct patterns of transcription factor (TF) binding sites. (A) Two-way clustering of gene promoters and TF binding sites, using genes (from the automated gene set) and TF sites shown in Table 3. Red squares denote presence of a particular TF site in the promoter of a particular gene, whereas green squares denote absence of the TF site (see Methods). An expanded version of A, showing gene identities, is available in the online supplement, as Figure E2. (B, C) Median expression levels in disease categories (NS, HS, COPD0, COPD1, COPD2-4) for genes in the upper cluster (B, contained above the upper orange dashed line in A) or lower cluster (C, contained below the bottom dashed line in A). The expression data are normalized by the expression in HS. Note that gene groups in B and C were defined solely by their TF binding site patterns. Also note that one gene (AKR1B10) does not follow the general pattern of the cluster of genes shown in B, where gene expression initially falls in HS as compared with NS but rises in subjects with COPD.
[More] [Minimize]Given the prevalence of oxidant-related genes among those regulated as a consequence of smoking and/or presence of COPD, we hypothesized that these genes could be used to discriminate between clinical categories. Therefore, the manual set of genes was subjected to a PCA that clusters data in an unsupervised manner. Good separation between NS and the two smoker groups was noted (Figure 4), but there was significant overlap between the HS and COPD groups.
Figure 4. Principal components analysis. There is good separation of the NS (triangles), and COPD (squares) groups using 122 oxidative stress probe sets from the manually curated set, but there is significant overlap between COPD and HS (circles). PC = principal component.
[More] [Minimize]We assumed that not all genes provide nonredundant information, and that more refined analyses should be able to improve group separation based on clinical category.
Clear distinction between NS, HS, and COPD is often limited by the relative insensitivity of clinical and physiologic tools used, particularly in mild disease. We sought, therefore, to identify associations between oxidant gene expression and individual clinical variables that are either shared by all smokers (e.g., smoking history) or represent a continuum (e.g., lung function). As shown in Figure 5, the clinical variables of age, weight, smoking history (pack-years), St. George's Respiratory Disease Questionnaire (SGRQ) score, VC, FVC, and carbon dioxide gas transfer coefficient (KCO) correlated wellwith the expression of the manual set of genes. The same was true for the pathologic variables: MUC5AC, epidermal growth factor receptor (EGFR) and ErbB3 expression, all expressed as percentage of epithelial staining. When the manual set was combined with a number of clinical variables, a further, albeit moderate, improvement in differentiating among NS, HS, and COPD was achieved in a PLS-DA model, with the Q2 measure of predictability of whether subjects were HS or had COPD increasing only from 0.37 to 0.55. This suggests that the manual set by itself is a relatively good predictor of clinical category.
Figure 5. Partial least squares analysis correlating the manual set of genes to a series of clinical variables (age, weight, smoking pack-years, the respiratory St. George's Respiratory Disease Questionnaire [SGRQ] symptom score), pathologic variables (i.e., expression of MUC5AC and epidermal growth factor receptor [EGFR]) assessed by immunohistochemistry of bronchial biopsies and expressed as percentage of epithelium staining positively, and physiologic parameters (FVC, carbon monoxide gas transfer coefficient [KCO], and VC shown as percentage of predicted). The bars represent the R2 (solid bars) and Q2 (open bars) measures of model fit and predictability (see explanation of R2 and Q2 in Results).
[More] [Minimize]Image analysis of epithelial cells for CYP1B1 protein failed to show any differences between the various subject categories when the staining was quantified by image analysis and results expressed as a percentage of the epithelium staining positively for the antigen detected by the antibody or scored semiquantitatively, and attributed visual scores of + to +++. Neither approach detected any correlation between CYP1B1 expression and COPD severity.
To assess the responses of epithelial cells to cigarette smoke, differentiated ALI cultures from 11 randomly selected subjects with or without COPD (4 NS, 1 HS, 4 subjects with stage 0 COPD, and 2 subjects with stage 2 COPD) were stimulated with CSE and the gene expression of these cells was related to the expression seen in vivo (as assessed in brushed cells). The ratios of mean gene expression values for genes differentially expressed in CSE-treated ALI cultures relative to vehicle-treated cultures (which reflected the in vitro effects of CSE) correlated strongly (r = 0.77) and highly significantly (p ≪ 10−5) with the ratios of gene expression in HS relative to those in NS (Figure E1) (which reflected the in vivo effects of smoking on gene expression in the studied cohort). In contrast, there was no correlation between the relative expression of oxidant-responsive genes in CSE-treated ALI cultures and the expression in subjects with COPD relative to HS (R = 0.01) (the latter ratio reflected the effect of having COPD), with only a subset of the genes showing similar trends in differential expression (data not shown).
This study has shown extensive changes in gene expression in the bronchial epithelium that are related to cigarette smoking and the presence of COPD. Although we have confirmed many of the previous observations of differential gene expression made by others, we have also identified new genes that are up-regulated in COPD. Taking advantage of an unbiased approach offered by GSEA to rank gene expression pathways, as opposed to individual genes, we have shown that pathways involved in oxidant/antioxidant responses are among the most differentially expressed pathways in smoking individuals, with marked differences being seen both between HS and NS and in HS and smokers with COPD. This clearly points to oxidant stress as a major feature of smoke-induced changes in epithelial function, some of which are further associated with the development of COPD. A close association was found between gene expression and a number of clinically relevant and pathologic variables in the airways but, importantly, gene expression was not linear in relation to disease severity: the expression of some genes peaked in mild COPD and declined in more severe disease. Some of the differential gene expression could be reproduced in vitro using cultures of differentiated epithelial cells obtained from a subset of epithelial cell donors; this strongly suggests that the differential expression between smokers and nonsmokers of some genes seen in vivo may be due to direct oxidant effects of cigarette smoke and may not require the full inflammatory response that involves cells such as neutrophils and macrophages. Finally, using the concept that common gene expression patterns can be associated with common sequence motifs in their promoter regions, we have shown that a number of TF binding sites, such as NF-κB and AP-1 sites, were overrepresented in our analyses, strongly suggesting their involvement in smoke-induced changes in epithelial function and COPD.
The strength of the current study is the careful selection of healthy smokers without any evidence of emphysema on HRCT as the control group, the detailed categorization of subjects with varying degrees of COPD severity, and the study of epithelial cells of high purity. Although studies using microarray analysis for global gene expression in surgically resected lung tissue has provided intriguing observations of differential gene expression in emphysema (9, 11, 12), it is unclear whether some of this is due to changes in cell populations. Furthermore, in some of these studies, the presence of peripheral tumors in HS (11) could have had a profound effect on gene expression. Epithelial brushings have been subjected to microarray analysis in two studies to date (8, 11) but only HS and NS have been compared so far.
We have identified more than 13,000 genes expressed in epithelial cells, of which 5,917 were differentially expressed. Many of the observed changes agree with comparisons between asymptomatic smokers and NS reported by Hackett and colleagues (8). The fold-changes between the two groups for the 44 antioxidant-related genes reported in that study correlated well (correlation coefficient, 0.90) with our data, even though different microarray platforms and expression measures (Affymetrix signal vs. robust microarray analysis [RMA] index) were used. The HS/NS fold-changes in manual gene set expression in our study also correlated well (0.85 correlation coefficient) with the current-smoker/never-smoker expression fold-change reported by Spira and colleagues (10). Our reanalysis of the latter study (10) showed that, of the 208 probe sets (158 genes) from the “automated” oxidant response gene set that was differentially expressed in our HS/NS comparison, as many as 79 probe sets (58 genes) were also differentially expressed in the comparison of never-smokers and current smokers: these included CYP1B1, CYP1A1, AKR1B10, AKR1C2, and SOD2. Thus, the degree of “cross-validation,” or overlap, of the significantly changed gene list is relatively high at 37%. In contrast, reanalysis of another study by Spira and colleagues (11), which compared whole lung samples from patients undergoing lung volume reduction for severe emphysema and whole lung samples free of emphysema that were resected because of pulmonary nodules, showed little overlap with our results. Of 152 probe sets (121 genes) from the automated oxidant response gene set regulated in our COPD/HS comparison, only 31 probe sets (30 genes) were also regulated in the study by Spira and colleagues (11), including AKR1B1 and CYP1B1. The lower degree of cross-validation (23%) is not surprising because our study focused on epithelial brushings, whereas the study by Spira and colleagues (11) analyzed samples that were inevitably less homogeneous in respect of cell types. An even lower level of reproducibility was obtained between our study and that of Ning and colleagues (12), which also analyzed whole tissue samples. Of the 121 genes from the automated gene set, only 5 were found to be also regulated in the study by Ning and colleagues—namely, CSPG2, DLD, GPX3, MAOA, and VCL. In this case, the lack of reproducibility is possibly also due to the fact that our study used a more recent Affymetrix platform (with 44,000 probe sets), thus measuring more genes than the study by Ning and colleagues (with 7,000 probe sets).
To gain further confidence in the microarray results, for a number of genes (CYP1B1, AKR1B1, AKR1B10, SOD2, CP) exhibiting the two most characteristic expression patterns as shown in Figures 1 and 2, we undertook validation by real-time reverse transcriptase–polymerase chain reaction, which showed very good reproducibility (for details, see Figure E3).
Our study differs significantly from previous studies by its application of GSEA (21), an unbiased approach to identify differentially expressed gene pathways as opposed to individual genes. In our analyses, we used a larger set of genes that were only potentially (by sequence-based prediction) related to oxidative stress responses. Thus, although a number of genes noted to be differentially expressed have been identified previously as being regulated in a smoking or COPD context (e.g., AKR1B1, SOD2, CYP1A1, TXN), a substantial number of genes are identified here that were differentially expressed but not associated previously with these conditions (e.g., AKR1B10, S-adenosylhomocysteine hydrolase-like 1, chromosome 5 open reading frame 4, GDP-mannose 4,6-dehydratase, protein disulfide isomerase–associated 6, and retinol dehydrogenase-10). Overexpression of AKR1B10 relative to healthy NS has been shown in smokers without airway obstruction (SiegeDB database, http://pulm.bumc.bu.edu/siegeDB/index_flash.html), but not in COPD. Recently, protein expression of AKR1B10 was associated with lung carcinoma and smoking history (32).
The expression of some genes increased in HS and then increased further in smokers with COPD (all the genes in the right upper quadrant of Figure 2A created by the lines indicating ratios of 1), whereas some genes were actually down-regulated in HS relative to NS (ratio of expression < 1 in the HS vs. NS comparisons), possibly due to adaptation to chronic stimulation by cigarette smoke contents. Among the latter group of genes, several (AKR1A1, CYP2B6, DHRS6, CYP4Z1, P5 [PDIA6], SOD2, CP) were up-regulated in smokers with COPD as compared with HS, suggesting that these genes might be more closely associated with the development of airflow limitation. Closer scrutiny of differentially expressed genes reveals intriguing patterns related to smoking status and COPD severity. The expression of a number of genes involved in oxidant stress responses (alcohol dehydrogenase-7 [ADH7], two members of the aldo-keto reductase family 1 [AKR1C3 and AKR1B1], gastrointestinal glutathione peroxidase 2 [GPX2], glucose-6-phosphate dehydrogenase [G6PD], glutathione reductases [GSR], glutathione–cysteine ligase regulatory protein [GCLM], glutamate cysteine ligase catalytic unit [GCLC], dual oxidase 2 [DUOX2], and microsomal glutathione S-transferase [MGST1]) was increased in HS compared with NS. This further increased in individuals with chronic productive cough but no significant airflow limitation (i.e., COPD stage 0), with further increase in stage 1 COPD. In more severe COPD, there was a fall in mean expression to levels noted in stage 0. It is tempting to speculate that decreases in expression of oxidant response genes may reflect a loss of defensive capabilities against infection as disease becomes more severe. A second pattern was observed whereby HS had down-regulated oxidant response gene expression (see Figure 2C), but this appeared to be restored as disease developed. However, as with the first pattern, expression peaked in mild disease and in more severe disease dropped to levels similar to those seen in NS.
A similar relationship was identified when looking at TFs that might be involved in oxidant response, pointing to a number of TFs that are associated with COPD. The TFs NF-κB, AP-1, and AP-2 have been reported previously to be associated both with oxidant responses and COPD; these associations were confirmed in our study. In addition, whereas E2F, FOXO4, Oct-1, and HNF-4 have been previously associated with oxidant responses, no previous study has suggested their involvement in COPD. In this study, Nrf2 was also among the overrepresented TF sites. This TF is involved in the protection against oxidative stress (33–35), regulating large numbers of oxidative stress–related genes, as shown in microarray analysis of Nrf2-knockout mice (36). Recently, it has been shown to be important in preventing the development of emphysema (37). Its expression pattern was significantly down-regulated in HS versus NS and significantly up-regulated in COPD versus HS, which was similar to the pattern of several genes regulated by Nrf2 (peroxiredoxin-1, [PRDX1]; CYP4F3; sushi-repeat-containing protein, X-linked 2 [SRPX2]; cytosolic malic enzyme 1 [ME1]; glutathione reductase [GSR]; NAD[P] menadione oxidoreductase 1 [HNQO1]). These observations offer the possibility of targeting TFs with novel drugs, with the advantage of influencing a number of relevant genes regulated by the same TF.
This is the first study to investigate the effects of cigarette smoke on gene expression both in vivo (i.e., in human bronchial epithelium samples) and ex vivo in epithelial cell culture. The strong correlation between expression fold-changes for genes differentially expressed in the comparison of HS and NS and in the comparison of ALI cultures exposed to CSE and untreated ALI cultures suggests that in vitro treatment recapitulates a number of changes in gene expression seen in HS. However, the absence of a similar correlation with the fold-changes when comparing the in vivo data from the smokers with and without COPD suggests that, in addition to the acute smoking effect, additional, disease-associated changes in oxidant-related gene expression occur in the epithelium of subjects with COPD, only some of which can be recapitulated in vitro. This observation suggests, but does not confirm, that other factors, such as activation of neutrophils, a source of potent oxidants, might be needed to induce the gene expression that was seen in COPD above and beyond the gene dysregulation seen in HS.
Because a clear distinction between HS and patients with COPD is limited by the relative insensitivity of the clinical and physiologic tools used, we have also sought to identify associations between oxidant-related gene expression and individual clinical variables that might either be shared by all smokers or represent one continuum. The manual set of genes correlated well with a number of clinical and pathologic variables presented as a continuum rather than strict categories imposed by the GOLD criteria—notably, quality of life, carbon monoxide transfer factor, and epithelial expression of ErbB1 (EGFR), ErbB3, and MUC5AC. It has been shown that oxidant stress is able to activate the ErbB family of receptor tyrosine kinases and drive mucus hypersecretion (38). We have recently reported (15) up-regulation of both EGFR and ErbB3 protein expression in the epithelium of smokers, as demonstrated by immunohistochemistry in bronchial biopsies of smokers when compared with NS, with a correlation between ErB3 and MUC5AC, suggesting, but not proving, a cause and effect link. In that study, no association was found with the numbers of neutrophils infiltrating the airways, but the current study suggests a link between oxidant stress and the mechanisms involved in regulating mucus hypersecretion.
This study has a number of limitations. Although this is the largest study of its kind with respect to the numbers of subjects whose in vivo gene expression was studied by epithelial brushings, the sample size (n = 11) for the in vitro experiments, in which cells were stimulated with CSE, was too small to enable a study of how disease severity in the studied subjects impacts on the effects of cigarette smoke on oxidant stress gene responses on ALI cultures from the same subjects. Without the knowledge of the validity of the methodologic approach taken and the results that we now have, the complexity and expense of ALI cultures precluded their use for all the subjects studied. This means that the observations made when correlating the in vitro to in vivo gene expression should be seen as preliminary; larger numbers of in vitro experiments using ALI cultures from subjects from the individual disease severity categories will need to be performed, because these might reveal different effects of CSE in mild and severe disease. A further limitation is the grouping of subjects with stage 2–4 COPD into one group, because further differentiation in gene expression might be seen if sufficient numbers of subjects in individual groups of stage 2, 3, and 4 COPD had been studied. The study also raises the important question as to why similar smoking- and disease-related differences seen in gene expression for such oxidant response elements as CYP1B1 could not be seen when quantifying the level of protein expression by immunohistochemical methods. We speculate that this is because immunohistochemistry is at best semiquantitative and may, therefore, not always be sensitive enough—especially if the degree of staining is low—to be related to a very quantitative read-out such as gene expression. Further studies comparing gene and protein expression are therefore needed.
In conclusion, this study shows that oxidative gene expression is influenced by cigarette smoke exposure. The fact that these genes are correlated with mucus production and several aspects of clinical severity suggests that they may contribute to lung damage that characterizes the chronic nature of COPD. Importantly, gene expression does not increase linearly as disease progresses, raising important questions as to what might be the cause of this phenomenon and what may be the functional consequences.
| 1. | Murray CJ. Evidence-based health policy–lessons from the Global Burden of Disease Study. Science 1996;274:740–743. |
| 2. | Barnes PJ. Chronic obstructive pulmonary disease. N Engl J Med 2000;343:269–280. |
| 3. | Horvath I, Donnelly L, Kiss A, Balint B, Kharitonov S, Barnes P. Exhaled nitric oxide and hydrogen peroxide concentrations in asthmatic smokers. Respiration (Herrlisheim) 2004;71:463–468. |
| 4. | Kayyali U, Budhiraja R, Pennella C, Cooray S, Lanzillo J, Chalkley R, Hassoun P. Up-regulation of xanthine oxidase by tobacco smoke condensate in pulmonary endothelial cells. Toxicol Appl Pharmacol 2003;188:59–68. |
| 5. | Lapenna D, Mezzetti A, de Gioia S, Pierdomenico S, Daniele F, Cuccurullo F. Plasma copper and lipid peroxidation in cigarette smokers. Free Radic Biol Med 1995;19:849–852. |
| 6. | Pignatelli B, Li C, Boffetta P, Chen Q, Ahrens W, Nyberg F, Mukeria A, Bruske-Hohlfeld I, Fortes C, Constantinescu V, et al. Nitrated and oxidized plasma proteins in smokers and lung cancer patients. Cancer Res 2001;61:778–784. |
| 7. | Cuzick J, Routledge M, Jenkins D, Garner R. DNA adducts in different tissues of smokers and non-smokers. Int J Cancer 1990;45:673–678. |
| 8. | Hackett NR, Heguy A, Harvey BG, O'Connor TP, Luettich K, Flieder DB, Kaplan R, Crystal RG. Variability of antioxidant-related gene expression in the airway epithelium of cigarette smokers. Am J Respir Cell Mol Biol 2003;29:331–343. |
| 9. | Golpon HA, Coldren CD, Zamora MR, Cosgrove GP, Moore MD, Tuder RM, Geraci MW, Voelkel NF. Emphysema lung tissue gene expression profiling. Am J Respir Cell Mol Biol 2004;31:595–600. |
| 10. | Spira A, Beane J, Shah V, Liu G, Schembri F, Yang X, Palma J, Brody JS. Effects of cigarette smoke on the human airway epithelial cell transcriptome. Proc Natl Acad Sci USA 2004;101:10143–10148. |
| 11. | Spira AE, Beane J, Pinto-Plata V, Kadar A, Liu G, Shah V, Celli B, Brody JS. Gene expression profiling of human lung tissue from smokers with severe emphysema. Am J Respir Cell Mol Biol 2004;31:601–610. |
| 12. | Ning W, Li CJ, Kaminski N, Feghali-Bostwick CA, Alber SM, Di YP, Otterbein SL, Song R, Hayashi S, Zhou Z, et al. Comprehensive gene expression profiles reveal pathways related to the pathogenesis of chronic obstructive pulmonary disease. Proc Natl Acad Sci USA 2004;101:14895–14900. |
| 13. | O'Donnell RA, Ward JA, Daraker A, Angco G, Broberg P, Pierrou S, Lund J, Holgate ST, Davies DE, Delany DJ, et al. Relationship between peripheral airway dysfunction, airway obstruction, and neutrophilic inflammation in COPD. Thorax 2004;59:837–842. |
| 14. | Pauwels RA, Buist AS, Ma P, Jenkins CR, Hurd SS. Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: National Heart, Lung, and Blood Institute and World Health Organization Global Initiative for Chronic Obstructive Lung Disease (GOLD): executive summary. Respir Care 2001;46:798–825. |
| 15. | O'Donnell RA, Richter A, Ward J, Angco G, Mehta A, Rousseau K, Swallow DM, Holgate ST, Djukanovic R, Davies DE, et al. Expression of ErbB receptors and mucins in the airways of long term current smokers. Thorax 2004;59:1032–1040. |
| 16. | Åstrand M. Contrast normalization of oligonucleotide arrays. J Comput Biol 2003;10:95–102. |
| 17. | Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res 2003;31:e15. |
| 18. | Broberg P. Statistical methods for ranking differentially expressed genes. Genome Biol 2003;4:R41. |
| 19. | Liu WM, Mei R, Di X, Ryder TB, Hubbell E, Dee S, Webster TA, Harrington CA, Ho MH, Baid J, et al. Analysis of high density expression microarrays with signed-rank call algorithms. Bioinformatics 2002;18:1593–1599. |
| 20. | Storey JD, Tibshirani R. Statistical significance for genomewide studies. Proc Natl Acad Sci USA 2003;100:9440–9445. |
| 21. | Mootha VK, Lindgren CM, Eriksson KF, Subramanian A, Sihag S, Lehar J, Puigserver P, Carlsson E, Ridderstrale M, Laurila E, et al. PGC-1a-responsive genes involved in oxidative phoshorylation are coordinately down-regulated in human diabetes. Nat Genet 2003;34:267–273. |
| 22. | Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M. The KEGG resource for deciphering the genome. Nucleic Acids Res 2004;32:D277–D280. |
| 23. | International Union of Biochemistry and Molecular Biology. Enzyme nomenclature: recommendations (1992) of the Nomenclature Committee of the International Union of Biochemistry and Molecular Biology. San Diego, CA: Academic Press; 1992. |
| 24. | Elkon R, Linhart C, Sharan R, Shamir R, Shiloh Y. Genome-wide in silico identification of transcriptional regulators controlling the cell cycle in human cells. Genome Res 2003;13:773–780. |
| 25. | Thomas EEP, Dewhirst B, Todd J, Anim T, Downey-Jones M, Tilley D, Servenius B, Pawlowski K, Westcott A, Woodwark M, et al. AstraZeneca GeneCatalogue: a novel view of the human transcriptome and proteome. Paper presented at: ISMB/ECCB 2004. 2004 Jul 31–Aug 5; Glasgow, U.K. |
| 26. | Johansson DLP, Berglund A. A multivariate approach applied to microarray data for identification of genes with cell cycle-coupled transcription. Bioinformatics 2003;19:467–473. |
| 27. | Li C, Wong HW. Model-based analysis of oligonucleotide arrays: model validation, design issues and standard error application. Genome Biol 2001;2:32.1–32.11. |
| 28. | Birkenkamp KU, Coffer PJ. FOXO transcription factors as regulators of immune homeostasis: molecules to die for? J Immunol 2003;171: 1623–1629. |
| 29. | Karin M, Takahashi T, Kapahi P, Delhase M, Chen Y, Makris C, Rothwarf D, Baud V, Natoli G, Guido F, et al. Oxidative stress and gene expression: the AP-1 and NF-kappaB connections. Biofactors 2001;15:87–89. |
| 30. | Rahman I. Oxidative stress, transcription factors and chromatin remodelling in lung inflammation. Biochem Pharmacol 2002;64:935–942. |
| 31. | Zhou J, Schmid T, Brune B. HIF-1alpha and p53 as targets of NO in affecting cell proliferation, death and adaptation. Curr Mol Med 2004;4:741–751. |
| 32. | Fukumoto S, Yamauchi N, Moriguchi H, Hippo Y, Watanabe A, Shibahara J, Taniguchi H, Ishikawa S, Ito H, Yamamoto S, et al. Overexpression of the aldo-keto reductase family protein AKR1B10 is highly correlated with smokers' non-small cell lung carcinomas. Clin Cancer Res 2005;11:1776–1785. |
| 33. | Cho HY, Jedlicka AE, Reddy SP, Kensler TW, Yamamoto M, Zhang LY, Kleeberger SR. Role of NRF2 in protection against hyperoxic lung injury in mice. Am J Respir Cell Mol Biol 2002;26:175–182. |
| 34. | Chen XL, Kunsch C. Induction of cytoprotective genes through Nrf2/antioxidant response element pathway: a new therapeutic approach for the treatment of inflammatory diseases. Curr Pharm Des 2004;10: 879–891. |
| 35. | Mathers J, Fraser JA, McMahon M, Saunders RD, Hayes JD, McLellan LI. Antioxidant and cytoprotective responses to redox stress. Biochem Soc Symp 2004;71:157–176. |
| 36. | Cho HY, Reddy SP, Debiase A, Yamamoto M, Kleeberger SR. Gene expression profiling of NRF2-mediated protection against oxidative injury. Free Radic Biol Med 2005;38:325–343. |
| 37. | Rangasamy T, Cho CY, Thimmulappa RK, Zhen L, Srisuma SS, Kensler TW, Yamamoto M, Petrache I, Tuder RM, Biswal S. Genetic ablation of Nrf2 enhances susceptibility to cigarette smoke-induced emphysema in mice. J Clin Invest 2004;114:1248–1259. |
| 38. | Takeyama KJ, Shim J, Dao-Pick T, Ueki IF, Nadel JA. Oxidative stress causes mucin synthesis via transactivation of epidermal growth factor receptor: role of neutrophils. J Immunol 2000;164:1546–1552. |
