Twenty-four SAGE libraries were constructed from bronchial epithelial specimens acquired from eight current smokers, twelve former smokers and four never smokers (Table 1). A former smoker was defined as someone who had stopped smoking for one year or longer. The smoking status was verified using exhaled carbon monoxide monitoring. Raw SAGE data for these transcriptomes has been made publicly available at National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO) with series accession number GSE5473. From these 24 libraries, we have collectively sequenced 3,111,471 SAGE tags, yielding 231,866 unique tags, making this the largest human SAGE study reported to date (Figure 1A). Of the unique tags, nearly half were present in more than one library at a tag count of one or greater, and 70% (82,983 tags) of these tags map to a UniGene cluster. As multiple tags frequently map to the same UniGene cluster, 25,653 unique UniGene clusters are represented in our dataset. Significantly, over 27,000 tags did not map to existing annotated genes, reiterating the continuing potential of re-mining this large dataset as tag-to-gene mapping improves with the continuing annotation of human transcripts.

We determined both the number of SAGE tags present in each of the current, former and never smoker transcriptomes, as well as those tags equally represented among the three different datasets. The criteria chosen for preferential expression was a threshold of a raw tag count of ≥ 2 across all samples in a particular set, but not existing in the other sets. Out of 3,033 tags expressed in all current smokers, we found 227 preferentially expressed tags (Additional file 1). In former smokers, 102 tags were found to be preferentially expressed (out of 2,579 tags) (Additional file 2), and in never smokers, 2,013 tags were found to be preferentially expressed (out of 5,192) (Additional file 3). It should be noted that the number of tags preferential to the never smoker set is substantially higher, most likely due to the lower sample size of never smokers relative to the other two groups. However, since we are using never smokers as a reference, a larger transcriptome will lessen the likelihood that we would find transcripts that are preferentially expressed in current and former smokers that were not correct. Looking at those tags which are common to all three groups, it was found that 1,970 tags (mapping to 1,733 unique genes) were expressed in all 24 libraries (Additional file 4). A Venn diagram illustrating the expression patterns of these three groups is given in Figure 1B.

We used a Mann Whitney U test to identify tags differentially expressed in the transcriptomes of current and never smokers. Using cut-off requirements of p ≤ 0.05, and a fold change of the means ≥ 2, we identified 609 SAGE tags (mapping to 487 unique genes) to be differentially expressed between current and never smokers (Additional file 5).

Using the 609 tags found to be differentially expressed between current and never smokers (Additional file 5), single link hierarchical clustering was performed using the program Genesis 23. We hypothesized that these 609 tags would classify current, former and never smokers. Indeed, distinct clusters emerged separating groups of current and former smokers with one exception of Current4 (Figure 2A). Of note, the former smoker who ceased smoking for only one year (Former 2) clustered with other former smokers. Moreover, principal component analysis (PCA) further validates the distinct groups of current, former and never smokers (Figure 2B).

To determine reversibility of smoking-related gene expression changes, we intersected tags differentially expressed between current and never smokers against tags showing significant expression difference between current and former smokers using similar criteria. By comparing these two sets, we can deduce which gene expressions are reversible, i.e., which genes are largely influenced by active smoking. This analysis yielded 161 tags mapping to 121 unique genes, which were deemed statistically significant, and representing 26% of the total number of differentially expressed tags between current and never smokers (Figure 3A, Additional file 6). Further analysis of these 121 differentially expressed genes has identified two main functions: xenobiotic metabolism and nucleotide metabolism (representing 33% of the reversible gene expression changes) (Table 2) and airway mucus secretion (representing 12% of the reversible gene expression changes) (Table 3). Genes related to oxidative stress were considered as part of the xenobiotic metabolism/nucleic acid metabolism category, and those genes previously associated with xenobiotic metabolism and oxidative stress through smoke exposure were among those identified 52425.

For example, ectonucleoside triphosphate diphosphohydrolase 8 (ENTPD8), an extracellular nucleic acid metabolic enzyme, is among 18 novel genes (labeled in bold in Table II) not previously associated with smoking and whose expression is increased in response to active smoking. According to enzyme classification, ENTPD8 is involved in purine and pyrimidine metabolism. Hence, this gene may potentially play a role in the chemical formation of DNA adducts.

Gene expression related to airway muco-ciliary function is also elevated in both current versus former smokers and current versus never smokers (Table 3). For example, trefoil factor 3 (TFF3), a structural component of mucus that is elevated in inflammatory response 2627, and calcium binding tyrosine-(Y) phosphorylation regulated (CABYR), originally shown to be localized in the principal part of the human sperm flagellum 28, are both highly expressed in current smokers relative to former and never smokers. Though TFF3 was recently shown to be expressed in response to chronic exposure of nicotine in intestinal cells 29, this is the first report of this gene being overexpressed within the bronchial epithelium in response to active smoking. Based on its assumed role in sperm motility, CABYR may be involved in ciliary function associated with muco-ciliary clearance response within the lung 28. Interestingly, overexpression of CABYR variants have been reported in a variety of brain tumors 30, suggesting a role in carcinogenesis. Previous observation of increased MUC5AC expression in current relative to never smokers and increased expression of microseminoprotein, beta- (MSMB), a gene shown to be present in mucosal secretions 31, supports the possibility of induction of airway mucosal response in active smokers 5242532.

By intersecting genes which are differentially expressed between current and never smokers with those that are different between former and never smokers, we can identify irreversible gene expression changes upon smoking cessation. This analysis yielded 152 tags (124 unique genes) meeting the criteria of statistical significance (p ≤ 0.05) at a fold change ≥ 2 (Figure 3B, Additional file 7). Although genes identified by this analysis appear to be functionally diverse, a small number of genes related to the cell cycle process and DNA repair have been identified here. For example, expression of P21/Cdc42/Rac1-activated kinase 1 (PAK1), cyclin D1 (CCND1), and cyclin G2 (CCNG2) all appear to be irreversibly lower in ever (former and current) smokers relative to never smokers. This finding is consistent with a previous report of increased inhibition of cell proliferation through genes such as CDKN1A in a higher stage (GOLD-2) of chronic obstructive pulmonary disease (COPD) versus the lowest stage (GOLD-0) 33.

We also found genes associated with DNA repair to be differentially expressed between current and never smokers, but similar between current and former smokers. APEX nuclease (multifunctional DNA repair enzyme) 1 (APEX1), High-mobility group box 1 (HMGB1), REV1-like (REV1L), and Tumor suppressor candidate 4 (TUSC4) are repair genes which we have found to be irreversibly under-expressed in ever smokers. Significantly, APEX1 has been shown to harbor SNPs associated with lung cancer susceptibility 34. Moreover, REV1L is involved with the recruitment of DNA polymerase eta to assist in DNA replication at arrested replication forks in areas of DNA lesions such as those formed by thymine dimmers 3536. TUSC4, also known as NPRL2, has recently been shown to increase sensitivity to cisplatin 37. Finally, HMGB1 has also been suggested to be involved with the recruitment of other repair-related proteins 38.

It should be noted that a significant proportion of former smokers in our sample set exhibited low FEV₁ levels, raising the possibility that airflow obstruction may be a confounding issue in this analysis. To address this, we used the 20 individuals with available FEV₁ data to compare individuals with moderate or severe COPD (FEV₁ < 80%, n = 12) with those individuals that would be classified with at most mild COPD (FEV₁ ≥ 80%, n = 8) according to the GOLD staging classification based on FEV₁ status 3940. Of the 157 tags differentially expressed between these two groups, only 6 tags overlap with our list of irreversible genes (Additional file 8). This minimal overlap suggests that the irreversible genes identified are not significantly associated with airway obstruction based on FEV₁ status. Nonetheless, airway obstruction should be considered in the interpretation of differential gene expression between current and former smokers.

A similar approach to that described here was undertaken by Spira et al. where the expression of 13 genes, including some putative oncogenes and tumor suppressor genes, was deemed irreversible upon cessation of smoking. However, none of these 13 genes overlapped with those identified in our study. This lack of overlap may reflect the differing locations from which the bronchial brushings were obtained as Spira et al 5 sampled from the right main bronchus whereas we have sampled peripheral sub-segmental airways.

It is interesting to note that MUC5AC appears in both the lists of statistically reversible and irreversible gene expression changes suggesting that expression of this gene exhibits distinct states of expression among current, former and never smokers. Moreover, it should also be noted that although 311 of the 609 tags were classified as either reversible or irreversible, the remaining 298 tags did not meet the statistical criteria for either category.

In addition to the SAGE analysis, which identified genes associated with airway mucosal response and xenobiotic/nucleic acid metabolism as distinguishing features between current and former smokers, we have performed quantitative RT-PCR on a secondary cohort of current, former and never smokers to validate selected genes for expression changes (Additional file 9). In total, five genes were selected for validation. From the set of reversible genes, we have chosen CABYR, ENTPD8, and TFF3 because their expression has not been associated with smoking previously. In addition, from the irreversible genes, we have selected MUC5AC. Using the delta-delta-Ct method to derive expression values, we then employed a Mann Whitney U Test to determine significance. The pattern of reversible over-expression in current smokers for CABYR, ENTPD8, and TFF3 (Figure 4A) and the irreversible over-expression of MUC5AC (Figure 4B) observed from the SAGE data, was validated by quantitative RT-PCR (Additional file 10). Raw cycle thresholds for each gene are available in Additional file 9.

Although the role of xenobiotic metabolism in smoking-induced carcinogenesis has been well documented 915, the potential influence mediated by changes in the composition of the airway mucosa in the development of lung cancer, has not been thoroughly investigated. It is possible that constant dysregulation of expression of genes associated with mucus secretion (such as TFF3 and MUC5AC) by smoking could potentially have a direct or indirect role in smoking-induced carcinogenesis.

One of the many genes involved in lung cancer development is cyclooxygenase 2 (COX2), which plays a multi-faceted role in cellular proliferation, migration and invasiveness 41. Notably, secretoglobin, family 1A, member 1 (SCGB1A1) protein has been shown to inhibit COX2 at the mRNA level 4243. We observed that SCGB1A1 expression is drastically reduced in current smokers but is expressed at similar levels in former and never smokers, and a previous study showed decreased serum SCGB1A1 level in smokers 44. It should also be noted that none of the SAGE sequence tags identified in the analysis mapping to SCGB1A1 are the most reliable tag according to SAGE Genie 45. However, even though the most reliable tag to this gene, CTTTGAGTCC did not pass statistically, the trend of reduced expression in current smokers relative to former smokers and similar expression between former and never smokers is consistent with the sequence tags that did appear in the analysis. Moreover, given that multiple tags have appeared from our analysis, although not as reliably mapped, we are confident that we are detecting SCGB1A1 mRNA expression. Interestingly, COX2 mRNA expression was not detected in the bronchial epithelium of current, former and never smokers from our SAGE data. A recent report demonstrated a significant increase in COX2 expression in normal lung fibroblasts when exposed to cigarette smoke extracts 46. It is possible that SCGB1A1 involvement is in the stroma and not in epithelial cells.

Despite lack of knowledge about CABYR, one of its few known interactions occurs with GSK3B 30.CABYR is a substrate of GSK3B 30, and exhibits reversible, increased expression with active smoking (Figure 4A). Though GSK3B was not identified as a smoking-related gene in our primary analysis, investigation of the SAGE data revealed a trend of similar decreased expression in current and former smokers relative to never smokers. Moreover, quantitative RT-PCR using a secondary cohort of samples validated that GSK3B expression is irreversibly reduced in ever smokers (Figure 4B). Recently, a published report using porcine tracheobronchial epithelial cells exposed to cigarette smoke components in vitro, demonstrated an inhibition of GSK3B gene expression 47. GSK3B has been shown to negatively interact with COX2 48. Reduced expression of GSK3B may therefore account for exaggerated inflammatory response despite smoking cessation and may contribute to development of lung cancer.

In this study, we have demonstrated differential expression of various components of respiratory tract mucus (including TFF3 and MUC5AC) according to smoking status (Table 3). However, our data indicates that MUC5AC expression is not completely reversible upon smoking cessation and in fact, exhibits three statistically distinct levels of expression between current, former and never smokers (Figure 4B). TFF2, a related motogen to TFF3, in conjunction with epidermal growth factor (EGF), has been shown to promote airway restitution, (i.e., movement of neighboring airway epithelial cells in response to injury mimicking rapid epithelium regeneration), through the activation of the epidermal growth factor receptor (EGFR) 49, expressed in the normal bronchial mucosa 5051. Other studies have also demonstrated increased expression of MUC5AC, along with EGFR and v-erb-b2 erythroblastic leukemia viral oncogene homol