Reversible and permanent effects of tobacco smoke exposure on airway epithelial gene expression
© Beane et al; licensee BioMed Central Ltd. 2007
Received: 8 January 2007
Accepted: 25 September 2007
Published: 25 September 2007
Tobacco use remains the leading preventable cause of death in the US. The risk of dying from smoking-related diseases remains elevated for former smokers years after quitting. The identification of irreversible effects of tobacco smoke on airway gene expression may provide insights into the causes of this elevated risk.
Using oligonucleotide microarrays, we measured gene expression in large airway epithelial cells obtained via bronchoscopy from never, current, and former smokers (n = 104). Linear models identified 175 genes differentially expressed between current and never smokers, and classified these as irreversible (n = 28), slowly reversible (n = 6), or rapidly reversible (n = 139) based on their expression in former smokers. A greater percentage of irreversible and slowly reversible genes were down-regulated by smoking, suggesting possible mechanisms for persistent changes, such as allelic loss at 16q13. Similarities with airway epithelium gene expression changes caused by other environmental exposures suggest that common mechanisms are involved in the response to tobacco smoke. Finally, using irreversible genes, we built a biomarker of ever exposure to tobacco smoke capable of classifying an independent set of former and current smokers with 81% and 100% accuracy, respectively.
We have categorized smoking-related changes in airway gene expression by their degree of reversibility upon smoking cessation. Our findings provide insights into the mechanisms leading to reversible and persistent effects of tobacco smoke that may explain former smokers increased risk for developing tobacco-induced lung disease and provide novel targets for chemoprophylaxis. Airway gene expression may also serve as a sensitive biomarker to identify individuals with past exposure to tobacco smoke.
Tobacco use remains the leading preventable cause of death in the United States, and cigarette smoking is the primary cause of chronic obstructive pulmonary disease and respiratory-tract cancers. Smoking is responsible for approximately 440,000 deaths per year in the US, resulting in 5.6 million years of potential life lost, $75 billion in direct medical costs, and $82 billion in lost productivity . Exposure to tobacco smoke is widespread - approximately 45 million Americans are current smokers and 46 million are former smokers . The risk of dying from smoking related diseases such as lung cancer and chronic obstructive pulmonary disease remains elevated for former smokers compared to never smokers . In the Dorn Study of US veterans, the Kaiser Permanente Prospective Mortality Study, and American Cancer Society Cancer Prevention Study I (CPS-I) populations, the risk of death from lung cancer among former smokers was elevated above never smokers 20 or more years following cessation . The Iowa Women's Health Study also found that former smokers had an elevated lung cancer risk compared with never smokers and that the risk for adenocarcinoma was elevated up to 30 years after quitting . As an increasing fraction of current smokers become former smokers, more lung cancer cases will occur in former smokers as the absolute risk of lung cancer in the population declines . It would be useful, therefore, to understand why former smokers remain at risk for lung cancer after smoking cessation in order to develop chemoprophylaxis treatments that might reduce risk.
A number of studies have shown that histologically normal large airway epithelial cells of current and former smokers with and without lung cancer display allelic loss [7, 8], genomic instability , p53 mutations , changes in DNA methylation in the promoter regions of several genes (including RARβ, H-cadherin, APC, p16INK4a, and RASFF1 [11, 12]), as well as changes in telomerase activity [13, 14]. Many of the changes persist in smokers for years after cessation [8, 9]. These observations suggest that the entire respiratory tree is affected by cigarette smoke, and that large airway cells might provide insight into the types and degree of epithelial cell injury that have occurred in current or former smokers.
We have previously reported a genome-wide expression profiling study of large bronchial airway epithelial cells obtained via bronchoscopy from never, current, and former smokers . In that study, we defined the baseline airway gene expression profile among healthy never smokers and identified gene expression changes that occur in response to smoke exposure. Of note, we found that a subset of genes modulated by smoking did not return to baseline years after smoking cessation. However, the limited sample size of the former smoker group (n = 18) precluded a detailed study of gene expression reversibility post-smoking cessation.
In this study, we collected airway epithelial cells from a larger sample of never, current, and former smokers and developed statistical models to identify the gene expression changes associated with smoking and categorized the degree to which these are reversible upon smoking cessation. We further explored the relationship between these gene expression changes and a number of publicly available human bronchial epithelial microarray datasets. The comparison of our dataset with the other datasets provides insights into common mechanisms airway epithelial cells use in response to a variety of different toxins. Lastly, development of a biomarker for ever tobacco smoke exposure using genes irreversibly altered by cigarette smoke provided additional validation of the gene expression changes upon smoking cessation and may provide a useful tool for epidemiological studies.
Demographic information for the never, former, and current smokers
Months since quitting
Effect of smoking and smoking cessation
Three-hundred and forty-three probesets show significant differences in intensity between current and never smokers based on the significance of the current smoking status variable in the linear model (q-value < 0.05 corresponding to a P < 7.6 × 10-4; see Materials and methods). Two-hundred and nineteen probesets remained after applying a filter to retain only probesets where the absolute current smoking status coefficient was greater than or equal to 0.584 (corresponds to an age-adjusted fold change between current and never smokers of 1.5). Finally, after filtering out redundant probesets (probesets representing the same gene) from this set of 219 probesets, probesets representing 175 genes remained. There was a high degree of overlap (78%) between genes we previously identified as being perturbed by active cigarette smoke exposure  and the 175 genes identified by the linear model.
Interestingly, 65% of the slowly reversible and irreversible genes were down-regulated by smoking, while only 23% of rapidly reversible genes were down-regulated by smoking (Fisher exact test P = 7.2 × 10-6). Amongst the rapidly reversible genes, those that were down-regulated tended to be the least reversible as determined by percent reversibility (Fisher exact test P = 0.0001 comparing the proportion of down-regulated genes in each tertile). Genes down-regulated by smoking, for example, account for only 6.5% of the most reversible tertile of rapidly reversible genes (n = 46), but account for 43% of the least reversible tertile (Figure 2a).
EASE analysis results
Permutation P value
GO molecular function
Rapidly reversible genes
GO molecular function
Electron transporter activity
Rapidly reversible genes
Homo sapienspentose phosphate pathway
Rapidly reversible genes
GO molecular function
Oxidoreductase activity, acting on the CH-OH group of donors, NAD or NADP as acceptor
Rapidly reversible genes
GO molecular function
Oxidoreductase activity, acting on CH-OH group of donors
Rapidly reversible genes
Carbohydrate metabolism - Homo sapiens
Rapidly reversible genes
Slowly reversible and irreversible genes
Enrichment of irreversible and reversible genes in bronchial epithelial cell datasets
Two group comparisons examined for each of the GEO datasets
No. of samples in condition 1
No. of samples in condition 2
Smoke 15 min, 24 hr recovery
Smoke 15 min, 4 and 24 hr recovery
Smoke 15 min, 4 hr recovery
Untreated, 2 and 4 h S9+CSCA/CSCB
8 and 12 h S9+CSCA/CSCB
S9 2, 4, 8 and 12 h
S9+CSCA 2, 4, 8, 12 h
S9 2, 4, 8 and 12 h
S9+CSCB 2, 4, 8, 12 h
Untreated 8 and 24 h
INF-gamma treated 24 h
Untreated 8 and 24 h
INF-gamma treated 8 and 24 h
Days 0 through 8
Days 10 through 28
4-PBA 12 and 24 h
4-PBA 24 h
RSV 24 h
RSV 4 and 24 h
IL13 4, 12, and 24 h
IL13 24 h
Control + IL13 4 h
IL13 12 and 24 h
In contrast to the above two datasets, the similarity between the gene expression changes in our dataset and those in GSE1276 was not as strong. GSE1276 used bronchial epithelial cells obtained from cadavers to study the effects of the S9 microsomal fraction from 1254-Aroclor treated rats and cigarette smoke condensate from two different brands of cigarettes at 2, 4, 8, and 12 hour time points . Genes down-regulated by smoking in our dataset were also down-regulated in epithelial cells treated with S9 plus cigarette smoke condensate for 8 and 12 hours compared to earlier time points. The uniqueness of GSE1276 is potentially due to the S9 treatment, which had unexpected broad effects on gene expression that may enhance or suppress the effects of the tobacco smoke condensate .
Genes that are perturbed by tobacco smoke exposure in our dataset also show some evidence of differential expression in six out of seven additional bronchial epithelial cell datasets. Genes up-regulated by smoking tended to be genes that are down-regulated by interferon gamma treatment for 24 hours in (GSE1815) , suggesting that smoking may have an immunosuppressive effect. Genes up-regulated in smoking also tended to be genes that are down-regulated at later time points during mucociliary differentiation (GSE5264) , suggesting that the damage caused by tobacco-smoke induces genes that are expressed more highly in undifferentiated epithelial cells. Genes down-regulated by smoking tended to be genes that are up-regulated in response to zinc sulfate (GSE2111) . These included the metallothionein genes (MT1X, MT1F, and MT1G). Taken together, the above results suggest that the bronchial epithelial cell response to tobacco smoke exposure consists of components that are shared with the response to a variety of other exposures.
Identifying common biological themes across datasets
In order to build upon the relationships between the datasets described above, we sought to establish additional relationships at the functional or pathway level. Gene lists composed of the genes in each of the over-represented gene categories (Table 2) were used to determine if these gene categories tended to be differentially expressed in the other bronchial cell datasets using GSEA (Figure 5b). This analysis shows that genes in five of the six functional categories that are induced by smoking and rapidly reversible upon smoking cessation also tended to be differentially expressed in two of the three smoking datasets. This further strengthens the notion that a similar bronchial epithelial response to tobacco smoke exposure is being detected in these datasets. Additionally, genes involved in oxidoreductase activity (which we found to be induced by smoking and rapidly reversible upon smoking cessation) are enriched among genes down-regulated during differentiation (GSE5264) or in response to interferon gamma treatment (GSE1815). These genes are also enriched among genes up-regulated in response to 4-phenylbutyrate (4-PBA) (GSE620) or interleukin-13 (GSE3183).
Biomarker of past exposure
Biomarker of tobacco smoke exposure constructed using the 28 irreversible genes
Number Classified Correctly
Mean of Random sets
Using linear models, we have identified genes differentially expressed in airway epithelium between never and current smokers and have characterized expression levels of these genes in former smokers who quit smoking for different periods of time. The majority (79%) of genes differentially expressed between current and never smokers are rapidly reversible upon smoking cessation while the remainders are either slowly reversible or irreversible. Differences between the rapidly reversible and slowly reversible or irreversible genes further suggest that their expression might be regulated through different mechanisms. The rapidly reversible genes have different biological functions than the slowly reversible or irreversible genes, suggesting that they might distinguish between an acute response to tobacco smoke and a more long-lasting response to tobacco smoke induced epithelial cell damage. The gene expression consequences of tobacco smoke exposure we identified are similar to gene expression changes observed in other human bronchial airway gene expression datasets involving tobacco smoke. Commonalities with human bronchial airway datasets involving other exposures suggest that the response to tobacco smoke exposure involves a number of common bronchial airway pathways. The accuracy of a biomarker of tobacco smoke exposure using irreversible genes in additional samples suggests that the irreversibility of these gene expression changes may provide a useful tool for assessing past exposure to tobacco smoke.
Many of the rapidly reversible genes are up-regulated by smoking and are involved in a protective or adaptive response to tobacco exposure and the detoxification of tobacco smoke components. The cytochrome p450s, CYP1A1 and CYP1B1, for example, are among the rapidly reversible genes and are involved in the oxidation of many compounds, including fatty acids, steroids, and xenobiotics. CYP1A1 and CYP1B1 have been previously described as being up-regulated in response to smoke  and CYP1B1 polymorphisms can influence the risk of developing lung cancer among never smokers . Several aldo-keto reductases, like AKR1B10 and AKR1C1, are also rapidly reversible upon smoking cessation. Aldo-keto reductases are soluble NADPH oxidoreductases that are involved in the activation of polycyclic aromatic hydrocarbons present in tobacco smoke and in the detoxification of highly carcinogenic nicotine-derived nitrosamino-ketone (NNK) compounds . Another class of rapidly reversible genes are the aldehyde dehydrogenases, such as ALDH3A1, which are involved in the oxidation of toxic aldehydes produced from oxidative stress and exposure to tobacco smoke . Both the cytochrome p450s and the aldehyde dehydrogenases have been found to be up-regulated in respiratory tissue from rats exposed to smoke  and the aldo-keto reductases are up-regulated in normal bronchial epithelium and non-small cell lung tumor tissue from smokers compared with non-smokers . All of the genes listed above as well as most of the differentially expressed genes that are members of the GO molecular function category 'oxidoreductase activity' are among the most highly reversible genes, suggesting that the up-regulation of these genes is driven by the acute exposure to smoke-related toxins and returns to baseline soon after the exposure to these compounds ceases. The induction of these genes in airway epithelial cells after 15 minutes of exposure to tobacco smoke (GSE2302) lends further support to this hypothesis.
In contrast to the rapidly reversible genes, the slowly reversible and irreversible genes reflect a more permanent host-response to tobacco smoke. Interestingly, several of these genes have been associated with the development of cancers of epithelial origin. CEACAM5, carcinoembryonic antigen-related cell adhesion molecule 5, is irreversibly up-regulated by smoking and is elevated in the serum of cancer patients with lung adenocarcinoma  and colorectal cancer . SULF1 (sulfatase 1), a gene irreversibly down-regulated by smoking, influences the sulfation state of residues present on heparin sulfate proteoglycans, which are involved in cell adhesion and mediate growth factor signaling. SULF1 was found to be down-regulated in ovarian, breast, pancreatic, renal, and hepatocellular carcinoma cell lines  and head and neck squamous carcinomas . UPK1B, uroplakin 1B, plays a role in strengthening and stabilizing the apical cell surface through interactions with the cytoskeleton . UPK1B is irreversibly down-regulated by smoking and has been shown to be reduced or absent in bladder carcinomas through CpG methylation of the proximal promoter [38, 39].
The enrichment of down-regulated genes among the irreversible, slowly reversible, and the least rapidly reversible genes suggests that genetic or epigenetic mechanisms, such as chromosomal loss [7, 8] or changes to promoter methylation status [11, 12], might account for the relative permanence of these gene expression differences. Given the rather rapid turnover of airway epithelial cells, the persistence of these changes post-smoking cessation may result from a clonal growth advantage to epithelial cells in the airway harboring these changes. Several of the down-regulated slowly reversible genes are present in cytoband 16q13, where a number of metallothioneins are located. Metallothioneins have the ability to bind both essential metals, like copper and iron, as well as toxic metals, such as cadmium and mercury. They also have detoxification and antioxidant properties and may be involved in cell proliferation and differentiation . MT3 has been shown to be down-regulated by hypermethylation in non-small cell lung tumors and cell lines . In addition, metallothioneins are thought to regulate some zinc-dependent transcription factors, such as the tumor suppressor p53, by donating zinc . Potential loss or methylation of the chromosomal locus containing several metallothionein genes may impair the ability of epithelial cells to protect or to repair cellular injury from future environmental exposures that occur after smoking cessation.
In order to confirm the observed effect of smoking and smoking cessation described above, we compared our dataset with other publicly available human bronchial epithelial cell datasets involving a variety of exposures. Reproducibility of findings using different microarray datasets across similar experimental conditions and cell types has not traditionally been common practice because overlap between differentially expressed gene sets is often surprisingly small . New methodologies for comparing datasets make the task more feasible , and provide more powerful methods for determining commonalities between the observed responses of a particular cell type under one or more conditions. The tobacco exposure associated gene expression changes we observed were concordant in three other datasets involving tobacco smoke exposures. The most significant similarity involved the gene expression consequences of tobacco smoke exposure in the small airway epithelium of never and current smokers (GSE3320). This suggests that the field of injury in response to tobacco smoke is similar throughout both the large and small airways. There was also significant similarity between those genes we found to be up-regulated by smoking and the immediate gene expression changes resulting from acute tobacco exposure (GSE2302). This similarity was significant for both rapidly reversible and irreversible/slowly reversible up-regulated genes (data not shown). The lack of similarity among genes down-regulated by smoking in our dataset and GSE2302 may reflect differences between acute and chronic cigarette smoke exposure, and suggests that up- and down-regulated irreversible gene expression may occur through different biological mechanisms. Additional large datasets of acute and chronic tobacco smoke exposure are needed to further explore these hypotheses.
There were also significant similarities between genes up- and down-regulated by smoking and the gene expression differences in additional datasets such as GSE5264 (cells undergoing mucociliary differentiation) and GSE1815 (interferon gamma treated cells). These may provide biological insights about the nature of airway epithelial response to tobacco smoke exposure. The gene expression program that accompanies mucociliary differentiation has led to the hypothesis that cultured 'undifferentiated' epithelial cells may more closely resemble damaged epithelium or neoplastic lesions in vivo because many genes associated with normal squamous epithelia, squamous cell carcinomas, or epidermal growth factor receptor signaling are more highly expressed in undifferentiated cells . The similarity between genes up-regulated by smoking in our dataset and genes that are more highly expressed early in mucociliary differentiation together with the similarity between genes down-regulated by smoking in our dataset and genes that are more highly expressed late in mucociliary differentiation might, therefore, reflect the cellular damage induced by smoke exposure. In addition, there was similarity between genes up-regulated by smoking in our dataset and genes down-regulated by treatment with interferon gamma. As interferon gamma plays a role in lung inflammatory responses, these similarities suggest that tobacco smoke exposure may suppress inflammatory responses in the airway. The relationships described above and presented in the results between our dataset and the other datasets are confirmed at a pathway level and suggest that oxidoreductase activity and electron transporter activity are among the important molecular functions of the bronchial epithelium that are regulated in response to a wide range of carcinogenic, inflammatory, and toxic exposures.
As an additional validation of the gene changes observed in response to smoking and smoking cessation, we developed a biomarker of tobacco smoke exposure. Using genes irreversibly altered by cigarette smoke, we were able to classify an independent sample set of former and current smokers (GSE4115) and a sample set of smokers and non-smokers (GSE5372) with high accuracy. Other datasets examining additional inhaled toxins (for example, ozone or fumes from charcoal stoves) are needed to determine if the persistent genomic changes we have identified are tobacco smoke specific. However, our preliminary biomarker results demonstrate the potential for developing a useful epidemiological tool if the gene expression biomarker could be ultimately extended to less invasive sites, such as the buccal and nasal epithelium, as these are tissues that are also directly exposed to tobacco smoke. Biomarkers of exposure are frequently used to improve upon or validate information about tobacco smoke exposure obtained by questionnaire; however, current biomarkers of tobacco exposure (for example, cotinine  and NNAL, a metabolite of the tobacco-specific nitrosamine NNK [46, 47]) are limited to detecting recent exposure. Development of a biomarker for long-term past exposure using gene expression could have widespread epidemiological utility. We are further interested to determine if there is sufficient similarity in the gene expression differences caused by distant and low-level tobacco smoke exposure such that a biomarker of past exposure could also detect current or past passive smoke exposure.
We have, for the first time, categorized smoking-related changes in airway gene expression by their degree of reversibility upon smoking cessation, which begins to provide insights into the mechanisms leading to persistent gene expression changes in the airway epithelium exposed to tobacco smoke. Further understanding of these mechanisms may aid in understanding why former smokers remain at risk for developing lung cancer years after quitting smoking and perhaps aid in developing treatments to lower this risk. In addition, a biomarker of past tobacco smoke exposure based on the expression of the genes that do not return to baseline levels after smoking cessation has the potential to provide a useful tool for epidemiological studies.
Materials and methods
We obtained airway epithelial brushings from never, current, and former smokers undergoing fiberoptic bronchoscopy between April 2003 and January 2006 (n = 281 samples, including replicates (n = 12)). Subjects with lung cancer or unknown lung cancer status were excluded from the analyses (n = 119). Demographics, including age, pack years, and months since quitting smoking, were obtained from each subject. The subjects were recruited from four institutions: Boston University Medical Center, Boston, MA; Boston Veterans Administration, West Roxbury, MA; Lahey Clinic, Burlington, MA; and St James's Hospital, Dublin, Ireland. The Institutional Review Boards of all of the medical centers approved the study and all subjects provide written informed consent. With the exception of nine samples, all samples used in the analyses were included in studies previously published by our group [15, 26, 48] (Additional data file 4).
Airway epithelial cell collection
Bronchial airway epithelial cells were obtained from the right mainstem bronchus with an endoscopic cytobrush (Cellebrity Endoscopic Cytobrush, Boston Scientific, Boston, MA, USA). RNA was isolated and its integrity and epithelial cell content was confirmed as described previously .
Microarray data acquisition
We processed, labeled and hybridized 6-8 μg of total RNA to Affymetrix HG-U133A GeneChips containing 22,283 probesets as described previously . We obtained log2-normalized probe-level data using the GCRMA algorithm  because it maximized the correlation between technical replicates compared to the Microarray Suite 5.0 algorithm and performed equivalently to a similar method, RMA (robust multichip average)  (Additional data file 3). All 281 samples (including replicates) collected during the study period were used for sample filtering. A z-score filter was applied to filter out arrays of poor quality. The filter involves computing an average z-score statistic across all probesets for each sample using z-score normalized data so that the mean gene expression value across all samples for each probeset is 0 and the standard deviation is 1 . Samples with high average z-scores were eliminated in addition to the 119 samples with lung cancer or unknown lung cancer status, leaving 104 samples - 21 never smokers without cancer (N), 31 former smokers without cancer (F), and 52 current smokers without cancer (C). The data can be accessed through GEO accession GSE7895.
Modeling the effect of smoking and smoking cessation
Linear regression models were used to identify genes differentially expressed as a function of tobacco smoke exposure. These genes were further analyzed to describe gene expression changes upon smoking cessation. For each probeset, the relationship between gene expression in log2 scale (ge), age, current smoking status (xcurr = 1 for current smokers and 0 otherwise), former smoking status (xform = 1 for former smokers and 0 otherwise), and the interaction between former smoking status and months elapsed since quitting smoke (xtq) was examined with the linear regression model:
ge i = β0 + β age * x age + β curr * x curr + β form * x form + β form.tg * x form * x tq + ε i
where εi represents the error that we assumed was normally distributed. The equation describes the expression of a probe i for never and current smokers as:
Never Smoker: ge i = β0 + β age * x age + ε i
Current Smoker: ge i = β0 + β age * x age + β curr * 1 + ε i
Age was included in the model to control for the potentially confounding effects of age and smoking status (Table 1). By difference, the age-adjusted fold change between current and never smokers is 2^βcurr. The standard least-square method was used to estimate the regression coefficients, and the significance of the regression coefficients was tested using the t-test. Goodness of fit of the models was assessed by analysis of residuals.
Probesets differentially expressed between current and never smokers were defined by two requirements. First, a q-value  for the regression coefficient βcurr < 0.05 (which corresponded to P < 7.6 × 10-4). The q-value is the expected proportion of false positives incurred when calling probesets with this q-value or smaller significant and was used to correct for multiple comparisons. Second, an absolute value of the βcurr coefficient >0.584, which corresponds to an age-adjusted fold change of expression >1.5. A fold change cutoff was chosen because of the little power provided by our sample size to detect smaller changes using multivariate linear regression models . After the q-value and fold change criteria were applied, probesets with the same gene symbol (according to the June 2006 HG-U133A Affymetrix annotation files), were filtered such that only the probeset with the lowest q-value was retained. All probesets without gene symbol annotation, however, were included.
The behavior of the probesets selected in the first comparison was further analyzed in former smokers. The linear model shown in equation 1 describes the expression of a probe i in former smokers as:
Former Smoker: ge i = β0 + β age * x age + β form * 1 + β form.tq * 1 * x tq + ε i
and allows us to further classify probes based on the pattern of expression in former smokers as a function of time since quitting smoking with respect to never smokers (Figure 1). From equation 4, we see that the expression of a probeset in a former smoker differs from that of a never smoker if the regression coefficient βform is significantly different from 0. The difference can be unrelated to time elapsed since quitting if the regression coefficient βform.tq is not significantly different from 0, or it can change over time if βform.tq is significantly different from 0. In the latter case, when the changes over time are monotone, we can identify the time point at which the fold change was equal to 1.5 (|β form + β form.tq * x form * x tq | = 0.584). This led us to the following definitions. First, a gene was defined as 'rapidly reversible' if the regression coefficient βform was not significantly different from 0 (P = 0.001). Second, a gene was defined as 'irreversible' if the regression coefficient βform.tq was not significantly different from 0 (P = 0.01), but the βform coefficient was significantly different from 0 (P < 0.001) and the absolute βform coefficient was >0.584 (corresponding to an age-adjusted fold change between formers and never smokers >1.5). Third, a gene was defined as 'indeterminate' if the regression coefficient βform.tq was not significantly different from 0 (P = 0.01), but the βform coefficient was significantly different from 0 (P < 0.001) and the absolute βform coefficient ≤0.584. Fourth, a gene was defined as 'slowly reversible' if the regression coefficients βform and βform.tq were significantly different from 0 (P < 0.001, and P < 0.01, respectively) and the absolute βform coefficient >0.584. The genes were characterized by the time point (tq) where |β form + β form.tq * x form * x tq | = 0.584. This corresponds to the time point where the age-adjusted fold change of never versus former smokers was equal to 1.5 (since all genes classified as slowly reversible were down-regulated by smoking).
In addition, to characterize the range of reversibility among genes designated as rapidly reversible, the percent reversibility for each gene was calculated according to the formula: . In rare cases where the former smoker versus never smoker fold change was slightly higher than the current versus never smoker fold change, the percentage was set to 100%; and in cases where the former smokers expression levels returned to a slightly lower level than never smokers, the percentage was set at 0%. The reversible genes were divided into tertiles based on this reversibility percentage.
Relationship of irreversible and reversible genes to other bronchial epithelial cell datasets
NCBI's microarray data repository, GEO , was queried for human bronchial epithelial cell samples in August 2006. Processed data were downloaded from GEO for each dataset (ten datasets total) that contained more than three total samples, contained more than two total samples per condition, and that was processed using whole genome arrays (Additional data file 2). The 175 genes differentially expressed between current and never smokers were mapped to the various datasets. PCAs were performed for each dataset across the mapped probesets using z-score normalized data. Graphs of the first versus second principal component were used as guides to decide what groups of samples show differential expression of the genes we identified as being differentially expressed between current and never smokers (data not shown).
The relationship was subsequently defined quantitatively using GSEA  (available through the GenePattern software ). The samples in each dataset from above were divided into two groups based on the experimental design - control versus the treated samples. If the samples were treated at two different time points, however, the time points were either combined into one treated group or kept separate for different comparisons between the control and the treated group at a particular time point (the PCAs from above were used to guide these decisions; Table 3). For each comparison, the probesets were mapped to gene symbols using GSEA's Affymetrix annotation files; or, in the case of the two non-Affymetrix arrays (datasets GSE2302 and GSE1276), the annotation file human-library.txt  was used. The redundant gene symbols were collapsed using a script written in the R Language for Statistical Computing  that retained the probesets with the highest absolute signal to noise ratio. This strategy was chosen so that all potentially differentially expressed genes were included in the analyses. The collapsed datasets were evaluated using GSEA to determine if the gene sets listed below were also differentially expressed in the datasets by the signal to noise statistic comparing treatment versus control. The following gene sets were tested: slowly reversible and irreversible genes up-regulated by smoking; slowly reversible and irreversible genes down-regulated by smoking; rapidly reversible genes up-regulated by smoking; rapidly reversible genes down-regulated by smoking; all genes up-regulated by smoking; all genes down-regulated by smoking. Significant enrichment was defined as a p value < 0.05 and a FDR < 0.25 derived using 10,000 gene-label permutations.
Identifying common biological themes across datasets
EASE  was used to identify GO molecular function categories, KEGG pathways, GenMAPP pathways, and chromosomal cytobands over-represented among genes designated as slowly reversible and irreversible or reversible compared to all annotated genes on the Affymetrix U133A microarray (Permutation P ≤ 0.01). GSEA was subsequently performed using gene lists derived from each significant EASE category to identify which of these over-represented categories were enriched in genes up- or down-regulated in each GEO dataset (Table 3). The enrichment of EASE categories observed in our dataset was confirmed using GSEA in which the βcurr smoking status coefficient (representing the magnitude of the difference between current and never smokers) was used to order the probesets.
Biomarker for past smoke exposure
A biomarker of past exposure using the irreversible genes (n = 28) was trained on the never and former smokers using a SVM classification system with a linear kernel via the R package e1071 . The SVM model was tested on the training set and three different test sets - the current smokers in the present study, current and former smokers that were not included in the present study from dataset GSE4511 previously published by our group, and GSE5372, which included gene expression measurement from large airway epithelial cells in 4 non-smokers and 5 current smokers at different time points (22 samples total) . The biomarker was used to predict the class of the GSE5372 samples taken at the initial time point (n = 9). P values for the performance of the biomarker were established by randomizing the class labels of the training set, re-running the algorithm 1,000 times, and calculating the proportion of the random runs that produced biomarkers that had the same or better accuracy in the test set samples.
Quantitative real time PCR
Quantitative RT-PCR analysis was used to confirm the differential expression of two irreversible and two rapidly reversible genes known to play roles in the detoxification of tobacco smoke and pathogenesis of lung cancer. Primer sequences for the four genes (ALDH3A1, CEACAM5, CYP1B1, and NQO1) were designed with PRIMER EXPRESS software (Applied Biosystems, Foster City, CA)) (Additional data file 5). Primer sequences of the housekeeping gene GAPDH were adopted from Vandesompele et al. . RNA samples (1 μg of residual RNA from the samples used in the microarray analysis) were treated with DNAfree (Ambion, Foster City, CA), according to the manufacturer's protocol, to remove contaminating genomic DNA. Total RNA was reverse-transcribed by using random hexamers (Applied Biosystems) and SuperScript II reverse transcriptase (Invitrogen, Carlsbad, CA). The resulting first-strand cDNA was diluted with nuclease-free water (Ambion) to 4 ng/μl. PCR amplification mixtures (25 μl) contained 20 ng template cDNA, 12.5 μl of 2× SYBR Green PCR master mix (Applied Biosystems) and 300 nM forward and reverse primers. Forty cycles of amplification and data acquisition were carried out in an ABI Prism 7700 Sequence Detector (Applied Biosystems). Threshold determinations were automatically performed by Sequence Detection Software (version 1.9.1; Applied Biosystems) for each reaction. All real-time PCR experiments were carried out in triplicate on each sample (mean of the triplicate shown). Four never, 3 former, and 2 current smokers were chosen for each gene based on the amount of RNA available (17 samples total: 6 current, 7 former, and 1 never smoker from this study and 3 additional never smokers collected prospectively).
All statistical analyses and hierarchical clustering were conducted using R statistical software v 2.2.1 and Bioconductor packages .
Additional data files
The following additional data are available with the online version of this paper. Additional data file 1 lists classifications of genes differentially expressed between current and never smokers according to their behavior in former smokers. For each gene the following information is given: the Affymetrix identification, the HUGO gene symbol, the direction of the change (up- or down-regulated in current smokers with respect to never smokers), the gene classification based on behavior of former smokers, and the percent reversibility. Additional data file 2 provides a Summary of human bronchial epithelial datasets downloaded from GEO. For each dataset the following information is included: GEO series identification, microarray platform, cell type, where the cells were obtained, cell donor information (if applicable), number of samples, experiment type, exposure, experiment description, data preprocessing, and PUBMED identification (if applicable). Additional data file 3 shows that GCRMA and RMA maximize the correlation between replicate samples. Average Pearson correlations between seven pairs of replicate samples where probeset gene expression values were determined using Microarray Suite 5.0 (MAS 5.0), log-transformed data from Microarray Suite 5.0 (Log2 MAS 5.0), and RMA. The average, standard deviation, and median of the correlation coefficients are shown. Additional data file 4 gives GEO identifications for never, former, and current smokers. This file explains how the samples used in the present study overlap with previous publications. GEO identifications are provided for each sample for the present study and for the previously published studies (each study used different data preprocessing). GEO identification 1 refers to the study published in  (15210990), GEO identification 2 refers to the study published in  (17334370), and GEO identification 3 refers to the present study. The study published in  (15608264) did not have an accompanying GEO submission. Additional data file 5 lists the quantitative real time PCR primer sequences. Primer sequences for the four candidate genes (ALDH3A1, CEACAM5, CYP1B1, and NQO1) designed with PRIMER EXPRESS software (Applied Biosystems), and the primer sequences of the housekeeping gene GAPDH adopted from Vandesompele et al. .
false discovery rate
Gene Expression Omnibus
gene set enrichment analysis
principal component analysis
robust multichip average
support vector machine
We thank Xuemei Yang, Sherry Zhang, Katrina Steiling, Frank Schembri, Martine Dumas and Norman Gerry for support with collection of samples and performing the microarray experiments. This work was supported by the Doris Duke Charitable Foundation (AS), NIH/NCI R21CA10650 (AS), NIH/NCI R01CA124640 (AS, MEL, and JB), and the National Institute of Environmental Health Sciences (NIEHS)/NIH U01 ES016035.
- Annual smoking-attributable mortality, years of potential life lost, and economic costs - United States, 1995-1999. MMWR Morb Mortal Wkly Rep. 2002, 51: 300-303.
- Cigarette smoking among adults - United States, 2003. MMWR Morb Mortal Wkly Rep. 2005, 54: 509-513.
- Halpern MT, Gillespie BW, Warner KE: Patterns of absolute risk of lung cancer mortality in former smokers. J Natl Cancer Inst. 1993, 85: 457-464. 10.1093/jnci/85.6.457.PubMedView ArticleGoogle Scholar
- Changes in Cigarette-Related Disease Risks andTheir Implications for Prevention and Control. Monograph No. 8[NIH Publ No 97-4213], 9-10. Edited by: Shopland DR, Burns DM, Garfinkel L, Samet JM. 2007, USDHHS, National Institutes of Health, National Cancer Institute, Ref Type: Serial (Book, Monograph)
- Ebbert JO, Yang P, Vachon CM, Vierkant RA, Cerhan JR, Folsom AR, Sellers TA: Lung cancer risk reduction after smoking cessation: observations from a prospective cohort of women. J Clin Oncol. 2003, 21: 921-926. 10.1200/JCO.2003.05.085.PubMedView ArticleGoogle Scholar
- Burns DM: Primary prevention, smoking, and smoking cessation: implications for future trends in lung cancer prevention. Cancer. 2000, 89: 2506-2509. 10.1002/1097-0142(20001201)89:11+<2506::AID-CNCR33>3.0.CO;2-8.PubMedView ArticleGoogle Scholar
- Powell CA, Klares S, O'Connor G, Brody JS: Loss of heterozygosity in epithelial cells obtained by bronchial brushing: clinical utility in lung cancer. Clin Cancer Res. 1999, 5: 2025-2034.PubMedGoogle Scholar
- Wistuba II, Lam S, Behrens C, Virmani AK, Fong KM, LeRiche J, Samet JM, Srivastava S, Minna JD, Gazdar AF: Molecular damage in the bronchial epithelium of current and former smokers. J Natl Cancer Inst. 1997, 89: 1366-1373. 10.1093/jnci/89.18.1366.PubMedView ArticleGoogle Scholar
- Mao L, Lee JS, Kurie JM, Fan YH, Lippman SM, Lee JJ, Ro JY, Broxson A, Yu R, Morice RC, et al: Clonal genetic alterations in the lungs of current and former smokers. J Natl Cancer Inst. 1997, 89: 857-862. 10.1093/jnci/89.12.857.PubMedView ArticleGoogle Scholar
- Franklin WA, Gazdar AF, Haney J, Wistuba II, La Rosa FG, Kennedy T, Ritchey DM, Miller YE: Widely dispersed p53 mutation in respiratory epithelium. A novel mechanism for field carcinogenesis. J Clin Invest. 1997, 100: 2133-2137.PubMedPubMed CentralView ArticleGoogle Scholar
- Wistuba II, Mao L, Gazdar AF: Smoking molecular damage in bronchial epithelium. Oncogene. 2002, 21: 7298-7306. 10.1038/sj.onc.1205806.PubMedView ArticleGoogle Scholar
- Guo M, House MG, Hooker C, Han Y, Heath E, Gabrielson E, Yang SC, Baylin SB, Herman JG, Brock MV: Promoter hypermethylation of resected bronchial margins: a field defect of changes?. Clin Cancer Res. 2004, 10: 5131-5136. 10.1158/1078-0432.CCR-03-0763.PubMedView ArticleGoogle Scholar
- Miyazu YM, Miyazawa T, Hiyama K, Kurimoto N, Iwamoto Y, Matsuura H, Kanoh K, Kohno N, Nishiyama M, Hiyama E: Telomerase expression in noncancerous bronchial epithelia is a possible marker of early development of lung cancer. Cancer Res. 2005, 65: 9623-9627. 10.1158/0008-5472.CAN-05-0976.PubMedView ArticleGoogle Scholar
- Yashima K, Litzky LA, Kaiser L, Rogers T, Lam S, Wistuba II, Milchgrub S, Srivastava S, Piatyszek MA, Shay JW, et al: Telomerase expression in respiratory epithelium during the multistage pathogenesis of lung carcinomas. Cancer Res. 1997, 57: 2373-2377.PubMedGoogle Scholar
- 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. 10.1073/pnas.0401422101.PubMedPubMed CentralView ArticleGoogle Scholar
- Hosack DA, Dennis G, Sherman BT, Lane HC, Lempicki RA: Identifying biological themes within lists of genes with EASE. Genome Biol. 2003, 4: R70-10.1186/gb-2003-4-10-r70.PubMedPubMed CentralView ArticleGoogle Scholar
- Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, Foulger R, Eilbeck K, Lewis S, Marshall B, Mungall C, et al: The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res. 2004, 32: D258-D261. 10.1093/nar/gkh066.PubMedView ArticleGoogle Scholar
- Kanehisa M, Goto S, Kawashima S, Nakaya A: The KEGG databases at GenomeNet. Nucleic Acids Res. 2002, 30: 42-46. 10.1093/nar/30.1.42.PubMedPubMed CentralView ArticleGoogle Scholar
- Dahlquist KD, Salomonis N, Vranizan K, Lawlor SC, Conklin BR: GenMAPP, a new tool for viewing and analyzing microarray data on biological pathways. Nat Genet. 2002, 31: 19-20. 10.1038/ng0502-19.PubMedView ArticleGoogle Scholar
- Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, et al: Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science. 1999, 286: 531-537. 10.1126/science.286.5439.531.PubMedView ArticleGoogle Scholar
- 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. 10.1165/rcmb.2002-0321OC.PubMedView ArticleGoogle Scholar
- Jorgensen ED, Dozmorov I, Frank MB, Centola M, Albino AP: Global gene expression analysis of human bronchial epithelial cells treated with tobacco condensates. Cell Cycle. 2004, 3: 1154-1168.PubMedView ArticleGoogle Scholar
- Pawliczak R, Logun C, Madara P, Barb J, Suffredini AF, Munson PJ, Danner RL, Shelhamer JH: Influence of IFN-gamma on gene expression in normal human bronchial epithelial cells: modulation of IFN-gamma effects by dexamethasone. Physiol Genomics. 2005, 23: 28-45. 10.1152/physiolgenomics.00011.2005.PubMedView ArticleGoogle Scholar
- Ross AJ, Dailey LA, Brighton LE, Devlin RB: Transcriptional Profiling of Mucociliary Differentiation in Human Airway Epithelial Cells. Am J Respir Cell Mol Biol. 2007Google Scholar
- Li Z, Stonehuerner J, Devlin RB, Huang YC: Discrimination of vanadium from zinc using gene profiling in human bronchial epithelial cells. Environ Health Perspect. 2005, 113: 1747-1754.PubMedPubMed CentralView ArticleGoogle Scholar
- Spira A, Beane JE, Shah V, Steiling K, Liu G, Schembri F, Gilman S, Dumas YM, Calner P, Sebastiani P, et al: Airway epithelial gene expression in the diagnostic evaluation of smokers with suspect lung cancer. Nat Med. 2007Google Scholar
- Nagaraj NS, Beckers S, Mensah JK, Waigel S, Vigneswaran N, Zacharias W: Cigarette smoke condensate induces cytochromes P450 and aldo-keto reductases in oral cancer cells. Toxicol Lett. 2006, 165: 182-194. 10.1016/j.toxlet.2006.03.008.PubMedView ArticleGoogle Scholar
- Wenzlaff AS, Cote ML, Bock CH, Land SJ, Santer SK, Schwartz DR, Schwartz AG: CYP1A1 and CYP1B1 polymorphisms and risk of lung cancer among never smokers: a population-based study. Carcinogenesis. 2005, 26: 2207-2212. 10.1093/carcin/bgi191.PubMedView ArticleGoogle Scholar
- Jin Y, Penning TM: Aldo-keto reductases and bioactivation/detoxication. Annu Rev Pharmacol Toxicol. 2007, 47: 263-292. 10.1146/annurev.pharmtox.47.120505.105337.PubMedView ArticleGoogle Scholar
- Vasiliou V, Nebert DW: Analysis and update of the human aldehyde dehydrogenase (ALDH) gene family. Hum Genomics. 2005, 2: 138-143.PubMedPubMed CentralGoogle Scholar
- Gebel S, Gerstmayer B, Bosio A, Haussmann HJ, Van Miert E, Muller T: Gene expression profiling in respiratory tissues from rats exposed to mainstream cigarette smoke. Carcinogenesis. 2004, 25: 169-178. 10.1093/carcin/bgg193.PubMedView ArticleGoogle Scholar
- Woenckhaus M, Klein-Hitpass L, Grepmeier U, Merk J, Pfeifer M, Wild P, Bettstetter M, Wuensch P, Blaszyk H, Hartmann A, et al: Smoking and cancer-related gene expression in bronchial epithelium and non-small-cell lung cancers. J Pathol. 2006, 210: 192-204. 10.1002/path.2039.PubMedView ArticleGoogle Scholar
- Hotta K, Segawa Y, Takigawa N, Kishino D, Saeki H, Nakata M, Mandai K, Eguchi K: Evaluation of the relationship between serum carcinoembryonic antigen level and treatment outcome in surgically resected clinical-stage I patients with non-small-cell lung cancer. Anticancer Res. 2000, 20: 2177-2180.PubMedGoogle Scholar
- Goldstein MJ, Mitchell EP: Carcinoembryonic antigen in the staging and follow-up of patients with colorectal cancer. Cancer Invest. 2005, 23: 338-351.PubMedView ArticleGoogle Scholar
- Lai J, Chien J, Staub J, Avula R, Greene EL, Matthews TA, Smith DI, Kaufmann SH, Roberts LR, Shridhar V: Loss of HSulf-1 up-regulates heparin-binding growth factor signaling in cancer. J Biol Chem. 2003, 278: 23107-23117. 10.1074/jbc.M302203200.PubMedView ArticleGoogle Scholar
- Lai JP, Chien J, Strome SE, Staub J, Montoya DP, Greene EL, Smith DI, Roberts LR, Shridhar V: HSulf-1 modulates HGF-mediated tumor cell invasion and signaling in head and neck squamous carcinoma. Oncogene. 2004, 23: 1439-1447. 10.1038/sj.onc.1207258.PubMedView ArticleGoogle Scholar
- Yu J, Lin JH, Wu XR, Sun TT: Uroplakins Ia and Ib, two major differentiation products of bladder epithelium, belong to a family of four transmembrane domain (4TM) proteins. J Cell Biol. 1994, 125: 171-182. 10.1083/jcb.125.1.171.PubMedView ArticleGoogle Scholar
- Varga AE, Leonardos L, Jackson P, Marreiros A, Cowled PA: Methylation of a CpG island within the uroplakin Ib promoter: a possible mechanism for loss of uroplakin Ib expression in bladder carcinoma. Neoplasia. 2004, 6: 128-135. 10.1593/neo.03337.PubMedPubMed CentralView ArticleGoogle Scholar
- Cowled P, Kanter I, Leonardos L, Jackson P: Uroplakin Ib gene transcription in urothelial tumor cells is regulated by CpG methylation. Neoplasia. 2005, 7: 1091-1103. 10.1593/neo.05364.PubMedPubMed CentralView ArticleGoogle Scholar
- Cherian MG, Jayasurya A, Bay BH: Metallothioneins in human tumors and potential roles in carcinogenesis. Mutat Res. 2003, 533: 201-209.PubMedView ArticleGoogle Scholar
- Zhong S, Fields CR, Su N, Pan YX, Robertson KD: Pharmacologic inhibition of epigenetic modifications, coupled with gene expression profiling, reveals novel targets of aberrant DNA methylation and histone deacetylation in lung cancer. Oncogene. 2006Google Scholar
- Meplan C, Richard MJ, Hainaut P: Metalloregulation of the tumor suppressor protein p53: zinc mediates the renaturation of p53 after exposure to metal chelators in vitro and in intact cells. Oncogene. 2000, 19: 5227-5236. 10.1038/sj.onc.1203907.PubMedView ArticleGoogle Scholar
- Evsikov AV, Solter D: Comment on " 'Stemness': transcriptional profiling of embryonic and adult stem cells" and "a stem cell molecular signature". Science. 2003, 302: 393-10.1126/science.1082380.PubMedView ArticleGoogle Scholar
- Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al: Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA. 2005, 102: 15545-15550. 10.1073/pnas.0506580102.PubMedPubMed CentralView ArticleGoogle Scholar
- SRNT Subcommittee on Biochemical Verification: Biochemical verification of tobacco use and cessation. Nicotine Tob Res. 2002, 4: 149-159. 10.1080/14622200210123581.View ArticleGoogle Scholar
- Hecht SS, Carmella SG, Chen M, Dor Koch JF, Miller AT, Murphy SE, Jensen JA, Zimmerman CL, Hatsukami DK: Quantitation of urinary metabolites of a tobacco-specific lung carcinogen after smoking cessation. Cancer Res. 1999, 59: 590-596.PubMedGoogle Scholar
- Hecht SS, Murphy SE, Carmella SG, Zimmerman CL, Losey L, Kramarczuk I, Roe MR, Puumala SS, Li YS, Le C, et al: Effects of reduced cigarette smoking on the uptake of a tobacco-specific lung carcinogen. J Natl Cancer Inst. 2004, 96: 107-115.PubMedView ArticleGoogle Scholar
- Shah V, Sridhar S, Beane J, Brody JS, Spira A: SIEGE: Smoking Induced Epithelial Gene Expression Database. Nucleic Acids Res. 2005, 33: D573-D579. 10.1093/nar/gki035.PubMedPubMed CentralView ArticleGoogle Scholar
- Wu Z, IRAGRM-MFSF: A Model Based Background Adjustment for Oligonucleotide Expression Arrays. The Journal of the American Statistical Association. 2004, 99: 909-917. 10.1198/016214504000000683.View ArticleGoogle Scholar
- Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, Speed TP: Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics. 2003, 4: 249-264. 10.1093/biostatistics/4.2.249.PubMedView ArticleGoogle Scholar
- Storey JD, Tibshirani R: Statistical significance for genomewide studies. Proc Natl Acad Sci USA. 2003, 100: 9440-9445. 10.1073/pnas.1530509100.PubMedPubMed CentralView ArticleGoogle Scholar
- Sebastiani PXHRM: Bayesian Analysis of Comparative Microarray Experiments by Model Averaging. Bayesian Analysis. 2006, 1: 707-732.View ArticleGoogle Scholar
- Edgar R, Domrachev M, Lash AE: Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Res. 2002, 30: 207-210. 10.1093/nar/30.1.207.PubMedPubMed CentralView ArticleGoogle Scholar
- Reich M, Liefeld T, Gould J, Lerner J, Tamayo P, Mesirov JP: GenePattern 2.0. Nat Genet. 2006, 38: 500-501. 10.1038/ng0506-500.PubMedView ArticleGoogle Scholar
- human-library.txt. 2007, Ref Type: Data File
- R Development Core Team: R: A language and environment for statistical computing. 2005, Vienna, Austria: R Foundation for Statistical ComputingGoogle Scholar
- The e1071 Package. 2006, Ref Type: Generic, [http://cran.r-project.org/src/contrib/Descriptions/e1071.html]
- Heguy A, Harvey BG, Leopold PL, Dolgalev I, Raman T, Crystal RG: Responses of the human airway epithelium transcriptome to in vivo injury. Physiol Genomics. 2007, 29: 139-148.PubMedView ArticleGoogle Scholar
- Vandesompele J, De Preter K, Pattyn F, Poppe B, Van Roy N, De Paepe A, Speleman F: Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes. Genome Biol. 2002, 3: RESEARCH0034-10.1186/gb-2002-3-7-research0034.PubMedPubMed CentralView ArticleGoogle Scholar
- Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry J, et al: Bioconductor: open software development for computational biology and bioinformatics. Genome Biol. 2004, 5: R80-10.1186/gb-2004-5-10-r80.PubMedPubMed CentralView ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.