Department of Genetic Medicine, Weill Cornell Medical College, New York, New York
Correspondence: Ronald G. Crystal, M.D., Department of Genetic Medicine, Weill Cornell Medical College, 1300 York Avenue, Box 164, New York, NY 10065, USA. Telephone: 646-962-4363; Fax: 646-962-0220; e-mail: firstname.lastname@example.org
Author contributions: R.S.: conception and design, collection/assembly of data, data analysis and interpretation, manuscript writing; R.W., R.K.Z., and I.W.C.: collection/assembly of data; N.R.H. and C.S.S.: data analysis and interpretation; R.L., Y. S., and J.S.: collection/assembly of data, data analysis and interpretation; M.A.S.M. and R.J.D.: provision of study material or patients, data analysis and interpretation; R.G.C.: conception and design, data analysis and interpretation, manuscript writing, final approval of manuscript.
Activation of the human embryonic stem cell (hESC) signature genes has been observed in various epithelial cancers. In this study, we found that the hESC signature is selectively induced in the airway basal stem/progenitor cell population of healthy smokers (BC-S), with a pattern similar to that activated in all major types of human lung cancer. We further identified a subset of 6 BC-S hESC genes, whose coherent overexpression in lung adenocarcinoma (AdCa) was associated with reduced lung function, poorer differentiation grade, more advanced tumor stage, remarkably shorter survival, and higher frequency of TP53 mutations. BC-S shared with hESC and a considerable subset of lung carcinomas a common TP53 inactivation molecular pattern which strongly correlated with the BC-S hESC gene expression. These data provide transcriptome-based evidence that smoking-induced reprogramming of airway BC toward the hESC-like phenotype might represent a common early molecular event in the development of aggressive lung carcinomas in humans. Stem Cells2013;31:1992-2002
Embryonic stem cells (ESCs) express a unique transcriptional program that determines their continuous self-renewal and pluripotency [[1, 2]]. A comprehensive meta-analysis of the hESC transcriptome [] identified 40 genes that are specifically expressed in hESC but rarely, if ever, detectable in the normal adult tissues. The exception, however, is cancer, where a number of these “hESC-signature” genes are overexpressed [[4-6]].
Although current data indicate that expression of a hESC-like transcriptional program is a molecular feature of advanced cancers, it is possible that elements of this program are acquired by adult healthy tissues chronically exposed to carcinogens prior to clinical manifestations of cancer. Specifically, we hypothesized that the earliest features of transition toward the hESC-like molecular phenotype are already present in the stem/progenitor cells of healthy individuals that, under the influence of chronic oncogenic stress, acquire this program to evolve into cancer-propagating cells.
Cigarette smoking is the dominant environmental carcinogenic stressor for airway epithelial cells, capable of evoking dramatic changes in the epithelial gene expression program [[7, 8]] and inducing oncogenic mutations and epigenetic modifications relevant to lung cancer [[9, 10]]. In susceptible individuals, smoking is responsible for inducing airway epithelial cells to change their normal differentiation pattern, undergo increased proliferation and eventually become malignant [[10, 11]]. The normal human airway epithelium is comprised of four major cell types: ciliated, secretory, intermediate/undifferentiated, and basal cells (BC) []. The BC population constitutes the stem/progenitor cell pool, capable of self-renewing and differentiating into the specialized cellular elements of the mucociliary airway epithelium [[13-16]]. BC hyperplasia and squamous metaplasia are the earliest airway epithelial lesions associated with smoking-induced carcinogenesis [[10, 11]]. However, the role of airway BC as a potential cellular origin of early molecular changes in the airway epithelium relevant to the development of lung cancer in smokers remains unknown. We hypothesized that cigarette smoking reprograms airway BC of healthy individuals toward a hESC-like molecular phenotype relevant to lung cancer.
Materials and Methods
Study Population and Datasets
Large airway epithelium (LAE) was obtained from 21 healthy nonsmokers and 31 healthy smokers (Supporting Information). All individuals were evaluated at the Weill Cornell NIH Clinical and Translational Science Center and Department of Genetic Medicine Clinical Research Facility, under protocols approved by the Weill Cornell Medical College Institutional Review Board. Before enrollment, written informed consent was obtained from each individual. Previously published gene expression data from 193 of 199 primary lung adenocarcinoma (AdCa) of individuals undergoing surgery at Memorial Sloan-Kettering Cancer Center (MSKCC) was used for analysis []. Independent publically available lung cancer datasets included Landi et al. (AdCa, n = 58) [], Kuner et al. (AdCa, n = 42; squamous cell carcinoma [SCC], n = 18) [], Garber et al. (AdCa, n = 40; SCC, n = 13; small cell lung cancer (SCLC), n = 4; large cell lung cancer (LCLC); n = 4), and Bild et al. (AdCa, n = 58; SCC, n = 53) []. The hESC datasets included Avery et al. (n = 3) [] and Denis et al. (GSE8590; n = 2).
Human LAE and Airway BC
LAE was collected via flexible bronchoscopy as described previously []. For purification of airway BC, LAE cells were cultured on type IV collagen using described previously methodology [] (Supporting Information). At days 7–8 of culture, when the cells were 70% confluent, cytospin preparations were made for immunohistochemical characterization, and RNA was extracted (Supporting Information). The resulting cells were >95% positive for cytokeratin five (K5), a BC marker [], and negative for the mesenchymal cell marker N-cadherin, secretory cell marker mucin 5AC, ciliated cell marker β-tubulin IV (Fig. 1A), and neuroendocrine cell markers chromogranin A and calcitonin gene-related peptide (data not shown). The capacity of obtained BC to generate differentiated progenies was confirmed by culturing them using the air–liquid interface (ALI) model of airway epithelial differentiation [] (Supporting Information).
Xenograft-Based Propagation of Human Lung AdCas
Tumor cells isolated from four patients with human lung AdCa were passaged at least twice in nonobese diabetic severe combined immunodeficiency (NOD.CB17-Prkdcscid/J; NOD/SCID) interleukin 2 receptor (IL2R) gamma null immunocompromised mice (Jackson Laboratory; Bar Harbor, ME (http://www.jax.org/); Supporting Information). After the final passage, tumor cells were processed for RNA isolation.
Preparation, Microarray Processing and Data Analysis
Transcriptome analysis of LAE, BC, and mouse-propagated AdCa samples was performed using HG-U133 Plus 2.0 array (Affymetrix, Santa Clara, CA http://www.affymetrix.com/estore), and MAS5-processed data were normalized and analyzed using GeneSpring version 7.3.1 (Agilent Technologies, Palo Alto, CA; http://www.home.agilent.com/agilent/home.jspx?cc=US&lc=eng;) (Supporting Information). To provide a cumulative measure of an individual signature expression in AdCa samples, signature-specific indices were calculated for each individual AdCa sample as a number of signature genes with the expression level above the median in AdCa subjects. The raw data are all publically available at the Gene Expression Omnibus (GEO) site (http://www.ncbi.nlm.nih.gov/geo/), accession number is GSE19722. The expression data for 193 primary human AdCa samples have been published previously []. Independent lung cancer datasets were analyzed using ONCOMINE database [] or using GeneSpring software (for databases imported from the GEO).
Massively Parallel mRNA Sequencing
The RNA sequencing (RNA-Seq) methodology is provided in Supporting Information.
Gene Expression Analysis of Cell Lines
NCI-H522, NCI-HI299, NCI-H338, and A549 lung carcinoma cell lines were purchased from American Type Culture Collection (ATCC) (Rockville, MD; http://www.atcc.org) and cultured according to the ATCC protocols. Expression of selected hESC genes was analyzed using specific TaqMan assays (Applied Biosystems, Foster City, CA; http://www.invitrogen.com/site/us/en/home/brands/taqman.html) as described []. Selection of cell lines for the analysis was based on the UMD p53 mutation database (http://p53.free.fr).
Survival Analysis and Comparison of Clinical Characteristics
Survival analysis was performed by Kaplan–Meier method and multivariate Cox proportional-hazard regression model using MedCalc version 11.3.3 (http://www.medcalc.be/). To analyze the effect of the 6-gene BC-S hESC-signature, AdCa patients were arbitrarily divided into “high expressors” (all six genes expressed above the median level in AdCa cohort) and “low expressors” (none of these genes expressed above the median level in AdCa cohort). To analyze the effect of an antibody set of 25-gene non-BC-S hESC-signature, AdCa patients were arbitrarily divided into “high expressors” (≥10 genes expressed above the median level in AdCa cohort) and “low expressors” (≤4 genes expressed above the median level in AdCa cohort). In Kaplan–Meyer analysis, difference in survival between the groups was analyzed with the log-rank test. In the multivariate Cox analysis, covariates included age, gender, pathologic tumor stage, chronic obstructive pulmonary disease (COPD), and BC-S hESC-signature expression (low and high). Clinical characteristics were compared using χ2 test (for categorical variables) and Kolmogorov-Smirnov test (for continuous variables).
hESC-Signature Genes Are Expressed in Adult Human Airway Epithelium
Based on the knowledge that BC, due to their unique pattern of integrin expression [], exhibit superior capabilities of adhesion and migration and, as stem/progenitor cells, can self-renew and proliferate [[16, 27]], and previous observations of the BC-like phenotype of airway epithelial cells grown in vitro [], we developed a method to obtain pure populations of airway BC by culturing freshly isolated LAE cells on type IV collagen (Methods; []). The BC phenotype of derived cells was confirmed by staining for BC-specific marker cytokeratin five [[16, 27]] (>95% positive cells), but negative staining for mesenchymal (N-cadherin), secretory [mucin 5AC (MUC5AC)], and ciliated cell (β-tubulin IV) markers (Fig. 1A). The isolated BCs were capable of generating differentiated ciliated airway epithelium in an ALI culture (Fig. 1B).
We first analyzed expression of the 40-gene hESC-signature expression in the LAE and LAE-derived BC of healthy nonsmokers (LAE-NS and BC-NS, respectively). Remarkably, 25% of hESC-signature genes were detected in at least 50% of samples in both groups (Supporting Information Table 1). Some of the hESC-signature genes were expressed in the LAE, but not in BC (e.g., ABHD9, CYP26A1, HESX1, and NANOG), that is, were cell differentiation-associated. Others (e.g., CDC25A, DTYMK, EPHA1, ISG20L1, and ORC2L) were expressed more abundantly in the BC population (Supporting Information Fig. 1).
Among 27 hESC-signature gene probes detected (see Supplemental Methods for criteria) in either LAE-NS or BC-NS, 15 were differentially expressed, with the majority (12 of 15) significantly upregulated in BC (Fig. 1C). Microarray analysis of BC differentiation in vitro in ALI revealed that while expression of a minor subset of hESC-signature genes increased with cell differentiation, including ABHD9 and CYP26A1, the majority of hESC genes downregulated during airway epithelial differentiation (Fig. 1D). The major changes occurred within the first 2 weeks of differentiation (Fig. 1D), a time necessary for the appearance of ciliated cells in the ALI differentiation model (Fig. 1B). Consistently, principal component analysis (PCA) revealed a significant difference between complete LAE and BC based on the expression of hESC-signature genes, with BC clustered closer to hESCs but shifted toward completely differentiated in vivo LAE during the first 2wk of differentiation in ALI (Fig. 1E).
Smoking Induces hESC-Signature in Airway BC
Next, we asked whether smoking induces expression of hESC-signature elements in the airway epithelium. Whereas hESC-signature expression by the complete LAE of healthy smokers (LAE-S) did not differ significantly from LAE-NS (Fig. 2A, left panel), BC of healthy smokers (BC-S) exhibited a broad upregulation of hESC-signature genes compared with BC-NS (Fig. 2A; right panel). Of the 35 hESC-signature gene probes expressed in BC-NS and/or BC-S, 18 (51%) probes corresponding to 13 (33%) genes were differentially expressed between these two groups, with all significantly upregulated in BC-S (Supporting Information Table 1). Notably, of these 13 genes, 10 were not detected in BC-NS, indicative of their de novo expression in BC-S (Supporting Information Table 1).
These differences were not due to the nonspecific BC transcriptome activation by smoking, as expression of housekeeping genes was unchanged (Supporting Information Fig. 2). Moreover, PCA revealed that, whereas smoking-induced transcriptome-wide changes had only limited contribution to variability between different groups for both LAE (Supporting Information Fig. 3A; left panel) and BC (Fig. 2B, left panel), healthy smokers and nonsmokers were completely segregated from each other based on the hESC-signature expression in BC (Fig. 2B, right panel), but not in the complete LAE (Supporting Information Fig. 3B, left panel). Consistently, unsupervised hierarchical cluster analysis completely separated BC-S from BC-NS based on the hESC-signature expression (Fig. 2C).
We used RNA-Sequencing (RNA-Seq) to validate differential expression of hESC-signature genes in BC-S versus BC-NS. This analysis revealed overlap between differentially expressed hESC-signature genes identified by RNA-Seq and microarray (Supporting Information Fig. 4A). Consistently, all 13 hESC-signature genes identified by microarray as upregulated in BC-S displayed similar direction of expression differences in the RNA-Seq analysis (Figs. 2D, Supporting Information Fig. 4B). RNA-Seq revealed two additional hESC-signature genes upregulated in BC-S (Supporting Information Table 2). Thus, using both methods, a total of 15 hESC-signature genes were found upregulated in BC-S compared with BC-NS. This set of genes was referred to as smoking-induced BC hESC-signature (“BC-S hESC-signature”).
To determine whether upregulation of the hESC-signature genes in BC-S was a result of the direct effect of cigarette smoke on BC, BC-NS were stimulated in vitro with cigarette smoke extract (CSE) as described previously []. Indeed, CSE significantly upregulated expression of the hESC-signature genes found induced in BC-S in vivo, but not those whose expression was unchanged in BC-S in vivo nor those associated with airway epithelial differentiation (Fig. 2E). As additional evidence that upregulation of the hESC-signature genes in BC after stimulation with CSE was smoke-dependent and not due to the nonspecific activation of the BC transcriptome, BC exposed to 2% CSE in vitro showed upregulation of CYP1A1, CYP1B1 and NQO1, well-known smoking-responsive genes in the airway epithelium [[7, 22]], whereas expression of the BC-signature genes KRT5, KRT6B, and ITGA6 [] remained unchanged (Supporting Information Fig. 5).
BC-S hESC-Signature Contributes to the hESC-like Phenotype of Human Lung AdCa
Based on observations that lung AdCas exhibit a hESC-like molecular profile [], we asked whether there is a commonality between hESC signatures induced in AdCa and BC-S. We first assessed the hESC-signature expression in primary human lung AdCa cells that had been passaged serially in NOD/SCID/IL2Rgamma-null immunodeficient mice, a strategy that permits evaluation of a pure epithelial compartment of carcinoma cells without the complicating contamination of noncancer cellular elements contributing to tumor microenvironment that might exhibit hESC-like features []. Of the 40 hESC-signature genes, 20 were significantly differentially expressed in AdCa xenografts as compared with both LAE-NS and LAE-S with 19 (95%) being up-regulated in AdCa xenografts (Fig. 3A, upper panels). While AdCa-xenografts displayed a considerable number of upregulated hESC-signature genes compared with BC-NS (Fig. 3A, left lower panel), the hESC-signature induced in BC-S was similar to that of AdCa-xenografts (Fig. 3A, right lower panel). Both unsupervised hierarchical clustering (Fig. 3B) and PCA (Fig. 3C) demonstrated that, based on the hESC-signature expression, BC-S were completely segregated from the LAE and BC-NS and distributed close to AdCa-xenografts. Consistently, comparative analysis of the hESC index, a cumulative measure of overexpression of hESC-signature genes (see Supporting Information Methods), revealed significantly increased average expression of hESC-signature genes in AdCa versus BC-NS, whereas there was no significant difference between AdCa and BC-S (Fig. 3D). Of the 15 BC-S hESC-signature genes, 12 (80%) were among those overexpressed in AdCa-xenografts (Supporting Information Table 1).
Next, the hESC-signature gene expression was assessed in primary tumors obtained from 193 lung AdCa patients []. Consistent with the xenograft data, 68% of hESC-signature genes were upregulated in primary lung AdCa (Supporting Information Table 1), showing an 89% overlap with the hESC-signature overexpressed in lung AdCa-xenografts. Twelve of the 15 (80%) BC-S hESC-signature genes, but only six of 25 (24%) remaining hESC-signature genes were upregulated in primary human lung AdCa (Supporting Information Table 1), indicating that it is the BC-S hESC-signature genes that predominantly contribute to the hESC-like phenotype in lung AdCa.
We next determined the overall BC-S hESC-signature gene expression in 192 AdCa patients with known clinical information using the BC-S hESC index, a cumulative measure of overexpression of 15 BC-S hESC-signature genes (a number of these genes whose expression was above the median in AdCa subjects). Six hESC-signature genes were identified (BRRN/NCAPH, DCC1/DSCC1, DTYMK, FLJ20105/ERCC6L, MCM10, and MYBL2), whose upregulation in BC-S versus BC-NS was detected by both microarray and RNA-Seq and whose expression in AdCa correlated with the BC-S hESC index (rho >0.6, p < .0001), representing, therefore, a cluster of coexpressed BC-S hESC-signature genes.
Based on the expression of these 6 BC-S hESC-signature genes, AdCa patients were categorized into “high expressors” (all six genes expressed above the median; n = 44), and “low expressors” (all six genes expressed below the median; n = 42). These 2 AdCa groups displayed strikingly opposite clinical and pathologic features (Table 1). Consistent with the smoking-dependent nature of the BC-S hESC-signature genes, 91% of high expressors were smokers versus 71% in the low expressor group. BC-S hESC-signature expression was significantly lower in AdCa patients who quit smoking compared with actively smoking AdCa patients (Fig. 3E). The high expressors exhibited higher comorbidity with chronic obstructive pulmonary disease (p < .03), lower lung function parameters such as forced expiratory volume in 1 second (FEV1; p < .05) and diffusing capacity of the lungs for carbon monoxide (DLCO; p < .05). High expressors had more advanced tumors (p < .04) with larger tumor size (p < .04), markedly poorer differentiation grade (p < .0001) and lower frequency of the prognostically favorable bronchoalveolar carcinoma features (p < .0001) than low expressors. Furthermore, AdCa recurrence was observed in 50% of high expressors compared with 19% of low expressors. Strikingly, high expressors had markedly shorter overall median survival than the low expressors (1,579 days versus 3,956 days; p < .0005 by log-rank test; Fig. 3E). Only 34% of high expressors versus 74% of low expressors were alive at the time of analysis (Table 1). In contrast to the BC-S hESC-signature genes, high expression of the non-BC-S hESC-signature genes was not associated with shorter survival of AdCa patients (Supporting Information Fig. 6). A multivariate survival analysis including various clinical covariates, which may also affect lung cancer survival, such as age, gender, pathologic tumor stage, smoking, and COPD, revealed that high BC-S hESC-signature expression is an independent prognostic factor negatively correlating with AdCa patient survival (p < .02, hazard ratio 2.62; 0.95% confidence interval 1.23–5.56; Table 2).
Table 1. Clinical Characteristics of Lung Adenocarcinoma Phenotypes Identified Based on Expression of the 6-gene Basal Cell Smoking-induced (BC-S) hESC Signature.
Abbreviations: BC-S hESC-signature, a cluster of six coexpressed human embryonic stem cell signature genes upregulated in airway basal cells of healthy smokers compared with those from healthy nonsmokers; CI, confidence interval; COPD, chronic obstructive pulmonary disease; HR, hazard ratio for overall survival.
BC-S hESC-Signature Is Associated with the TP53-Inactivation Molecular Phenotype
We then asked whether AdCa subjects overexpressing BC-S hESC genes exhibit distinct patterns of mutations. Although there was no significant difference in the frequency of mutations of EGFR or KRAS (Table 1), or STK11, BRAF, and PTEN (not shown) between high- and low-expressors, AdCa subjects with high BC-S hESC-signature expression exhibited significantly higher frequency of mutations of the tumor suppressor gene TP53 (p < .0002; Table 1).
Consistently, the presence of TP53 mutations in AdCa was associated with higher overall expression of BC-S hESC-signature genes (Fig. 4A). In AdCa-smokers with TP53 mutations, expression of these genes was strongly positively correlated with the expression of a subset of genes known to be upregulated after TP53 mRNA silencing [] (“TP53-inactivation signature”; Fig. 4B). Consistently, the NCI-H522 and NCI-HI299 lung carcinoma cell lines with TP53-inactivating mutations exhibited significantly higher expression of the BC-S hESC-signature genes than TP53-wild-type lung cancer cell lines A549 and NCI-H838 (Fig. 4C).
We next analyzed whether the TP53-inactivation molecular phenotype is present in BC-S. PCA revealed that, based on the expression of both BC-S hESC-signature and the TP53-inactivation signature, BC-S, but not BC-NS, shared a similar distribution as AdCa subjects with TP53 mutations (Fig. 4D), indicating that BC-S and AdCa with TP53 mutations share a similar TP53-inactivation molecular pattern. Next, we analyzed the effect of smoking on the TP53-inactivation signature expression in the healthy airway epithelium. No significant differences were detected between the complete LAE-NS and LAE-S (Fig. 4E, upper panel), whereas there was a dramatic upregulation of the TP53-inactivation signature genes in BC-S versus BC-NS (Fig. 4E, lower panel) indicating that smoking selectively induces TP53-inactivation phenotype in the BC population. Finally, there was a very strong correlation between the hESC- and TP53-inactivation signatures induced by smoking in airway BC (Figs. 4F–4G).
Various Types of Human Lung Cancer Overexpress BC-S hESC-Signature
To evaluate BC-S hESC-signature expression in different subtypes of lung cancer, independent lung cancer datasets were analyzed [[18, 19, 31]]. Similar to the original AdCa cohort, all three independent AdCa datasets exhibited preferential upregulation of the BC-S-induced hESC-signature genes, with remarkable overlap between individual datasets (Fig. 5A). Notably, non-BC-S hESC-signature genes CYP26A1, HESX1 and NANOG, associated with airway epithelial differentiation, were downregulated in AdCa datasets (Fig. 5A). Preferential upregulation of the BC-S hESC signature genes with a pattern similar to that induced in lung AdCa was also observed in two independent lung SCC datasets as well as in small- and large-cell lung carcinomas (Fig. 5A). Overall BC-S hESC-signature gene expression in lung SCC was considerably higher than in lung AdCa (Fig. 5A).
Genome-wide PCA analysis revealed that airway BC from healthy individuals exhibit higher similarity to hESC with BC-S distributed closer to lung cancer samples (Fig. 5B). Based on the entire hESC-signature expression, a subset of AdCa samples and the majority of SCC shared with BC-S, but not BC-NS, a similar distribution with a notable shift toward hESC (Fig. 5C). Further restriction of the analysis to the 15-gene BC-S hESC-signature revealed similarity of the SCC samples and a subset of the AdCa samples to both BC-S and hESC (Fig. 5D). This spatial pattern was effectively reproduced using the dataset containing six coexpressed, prognostically relevant BC-S hESC-signature genes (Fig. 5E), but not the non-BC-S hESC-signature genes (Fig. 5F). Finally, SCC and a subset of AdCa samples clustered together with BC-S and hESC based on expression of the TP53-inactivation signature (Fig. 5G), suggesting that acquisition of the transcriptome features of TP53 inactivation is coupled to the reprogramming toward a common hESC-like phenotype shared by BC-S and lung cancer.
Accumulating evidence indicates that a number of human epithelial cancers display activation of genes and associated regulatory networks previously ascribed to hESC [[4-6]]. These observations lead to the hypothesis of this study that, under the influence of the chronic carcinogenic stress of cigarette smoking, airway BC, the adult stem/progenitor cell population of the airway epithelium, acquire hESC-like molecular traits similar to those induced in lung cancer as an early step toward malignant tissue derangement.
Cigarette smoking, the major risk factor for lung cancer, induces all steps of preneoplastic progression, inducing BC proliferation resulting in hyperplasia, skewing airway epithelial differentiation toward metaplasia, followed by dysplastic changes that precede the development of invasive carcinoma [[9-11]]. These morphologic changes are preceded and/or accompanied by genomic alterations, epigenetic modifications, and transcriptome reprogramming []. This study demonstrates that smoking induces a unique hESC-like transcriptome program in the airway BC of healthy individuals, which considerably overlaps with the hESC-like program present in all major subtypes of human lung cancer, including AdCa, SCC, large-, and small-cell carcinomas, and is associated with a distinct, more aggressive phenotype of lung carcinomas. Several themes relevant to the molecular and cellular origins of human lung cancer emerge from these observations.
First, the data of this study emphasize the role of airway BC, the airway stem/progenitor cells, as the major target of smoking-induced reprogramming of the airway epithelium toward a lung cancer-relevant molecular phenotype. Smoking is known to induce contrasting effects on different cell populations of the airway epithelium. In the airways of smokers, there is loss and functional defects of ciliated cells accompanied by increased proliferation of BC []. But how do BC, located below the layer of differentiated and columnar cells, sense cigarette smoke? Previous studies showed that the airway epithelial junctional barrier, which separates the luminal and basolateral epithelial compartments, is compromised by cigarette smoking [], making the BC compartment accessible to components of cigarette smoke. In addition, BC may directly sample luminal content by extending their processes across the epithelial layer []. In favor of such model, we found that direct exposure of BC from healthy nonsmokers to cigarette smoke extract in vitro resulted in the acquisition of the hESC-signature similar to that induced in BC-S in vivo. Finally, smoking-induced hESC-like BC phenotype was maintained in culture, suggesting that stable changes to the BC genome and/or epigenome induced by smoking in vivo allow them to maintain their phenotype after they have been removed from the in vivo microenvironment. The overall hESC-signature expression markedly decreased following BC differentiation into the ciliated epithelium in vitro, suggesting that regulatory mechanisms controlling the expression of these genes in vivo were also largely preserved in vitro, and the observed increased hESC-signature gene expression in BC-S versus BC-NS was due to their in vivo smoking-induced reprogramming. Furthermore, the same set of BC-S hESC genes was induced in BC-NS after exposure to the cigarette smoke extract in vitro, indicating it is smoking, not the culture itself, that is the factor responsible for the induction of the BC-S hESC-signature genes in BC-S. The smoking-dependent nature of the BC-S hESC-signature expression was further supported by the in vivo observation that a subset of AdCa patients who quit smoking display significantly lower expression of the BC-S hESC-signature genes than actively smoking AdCa patients. This suggests that both irreversible and reversible genomic and epigenomic changes may be responsible for smoking-induced acquisition of the BC-S hESC transcriptional program in both BC and lung carcinomas.
Second, results of this study provide new insights into the cellular origin of human lung cancer. Airway BC have been regarded as putative cell-of-origin for SCC [[10, 34]], but not for other types of lung cancer. The remarkable similarity of the hESC-signature induced in BC-S to that overexpressed in four histologically distinct types of human lung cancer identified in this study suggests that reprogramming toward a hESC-like molecular phenotype in these types of lung cancer likely represents a common early molecular process driven by smoking-induced changes in airway BC. Expansion of the smoking-reprogrammed hESC-like BC clones in susceptible individuals might provide potential explanation for progressive dedifferentiation associated with the development of smoking-associated lung carcinomas. Indeed, patches of clonally related cells harboring a uniform set of molecular alterations identical to those present in lung cancer have been found in the histologically normal airway epithelium of smokers without cancer [[35, 36]] and the cells expressing BC markers CK5 and CK14 are predominant in SCC-related potentially preneoplastic lesions in smokers' airways []. Another intriguing finding in this study is that, although the BC were from the LAE, the smoking-induced hESC-signature in these cells contributed to the molecular phenotype of both predominantly proximally derived lung carcinomas such as SCC, SCLC, and large-cell lung carcinomas, as well as AdCa, which is thought to originate in peripheral airways []. It is known that smoking creates a field of cancer-related molecular changes throughout the airway epithelium []. In support of this model, multiple clonal outgrowths of molecularly altered cells have been found widely distributed in the airway epithelium of a smoker [] and smoking-induced changes in the LAE transcriptome have been used to predict lung cancers located at a distance from the sampled LAE [].
Clinical relevance of the BC-S hESC signature identified in this study was further demonstrated by the observation that overexpression of this signature in AdCa is associated with a distinct, more aggressive clinical/pathologic phenotype. These individuals are predominantly smokers, have a higher comorbidity with COPD and decreased lung function parameters FEV1 and DLCO, more advanced pathological stage, larger tumors, markedly poorer differentiation grade, higher recurrence frequency and, most strikingly, a 79-month shorter median overall survival than lung AdCa patients not expressing this signature. Importantly, high expression of the BC-S hESC-signature predicted poorer survival in lung AdCa independent from other covariates such as age, gender, pathologic tumor stage and COPD. By contrast, high expression of the non-BC-S hESC-signature genes was not associated with shorter survival of AdCa patients. In a recent study, overexpression of a distinct set of the hESC-related genes was associated with poor survival of lung AdCa, but not SCC []. It is likely that, compared with lung AdCa, which is characterized by variable expression of the BC-S hESC-signature genes allowing categorization of patients into “high” and relatively “low” expressors with markedly different survival, SCC with its uniformly high expression of the BC-S hESC-signature does not exhibit such clinically detectable heterogeneity. Thus, smoking-induced hESC-signature in BC population of healthy airway epithelium carries valuable information related to both early pathogenesis and clinical phenotypes of lung carcinomas, including those not previously thought to originate from BC.
Third, this study sheds light on the early molecular mechanisms associated with acquisition of lung cancer-relevant features in the stem/progenitor cells of otherwise normal airway epithelium chronically exposed to the oncogenic stress of cigarette smoking. The observation of this study of a significantly higher incidence of TP53 mutations in AdCa patients highly expressing the BC-S hESC-signature, suggests two possible mechanistic models whereby smoking might reprogram airway BC toward cells with lung cancer-relevant molecular phenotype. First, TP53-inactivation might be required for acquisition of the hESC-like transcriptome features. TP53 is a tumor suppressor gene encoding phosphoprotein p53, which suppresses tumor formation by promoting apoptosis, activating cell cycle checkpoints and inducing senescence []. In addition to these classic functions, recent studies have documented a critical role for TP53 in maintaining ESC genomic stability, inducing their differentiation [] and suppressing pluripotency [[42, 43]]. TP53 mutations, a known biomarker of cigarette smoke exposure in lung cancer [], represent the most common mutation in lung carcinomas, including SCC, AdCa and SCLC, with a frequency varying between 40% and 75% depending on smoking status []. This study provides several lines of evidence in favor of the model, in which activation of the BC-S hESC signature is associated with the TP53 gene inactivation. First, we found that different lung carcinoma cell lines harboring TP53 gene mutations overexpress hESC-signature genes with a pattern similar to that induced in BC-S. Second, AdCa patients with TP53-mutations exhibited significantly higher expression of BC-S hESC-signature genes. Third, transcriptome analysis revealed a selective induction of genes associated with the TP53 inactivation in BC, but not in the complete airway epithelial population of healthy smokers. The molecular pattern of TP53 inactivation in BC-S was similar to that present in hESC, AdCa with TP53-mutations and the majority of SCC samples. Finally, overall expression levels of the hESC and TP53-inactivation signatures in airway BC strongly correlated.
Thus, it is possible that BC carrying inactivated TP53 acquire the hESC-like phenotype, gain a selective growth advantage and eventually play a role in tumor initiation and propagation, thereby contributing to the development of poorly differentiated aggressive lung carcinomas. In support of this scenario, a widespread distribution of epithelial cells bearing a single point mutation in TP53 codon 245, a codon which is frequently mutated in lung cancer, has been detected in the airways of a smoker with dysplastic changes [], suggesting that a single clone of smoking-reprogrammed TP53-mutant progenitor cells might populate relatively large and distant areas of the airway epithelium prior to the formation of overt cancer. Furthermore, loss of heterozygocity at the TP53 locus and overexpression of the mutant p53 protein have previously been found in the dysplastic bronchial epithelium of smokers without lung cancer []. Although the mechanism causing TP53 inactivation in the BC-S is beyond the scope of this study, epigenetic modifications that may occur in response to environmental factors, such as cigarette smoke, can repress gene function without changes in the DNA sequence []. Alternatively, DNA replication stress induced by cigarette smoking in proliferating BC might select for TP53-inactivation as a response to ongoing DNA damage []. Consistent with this concept, CHEK2, the central component of the DNA damage response [], was among the hESC-signature genes induced in BC-S. When the function of p53 is lost, BC can escape its “genome guardian” functions, acquire the cancer-relevant hESC-like phenotype and the precancerous lesion can become malignant. Indeed, DNA replication stress leading to genomic instability and selective pressure for p53 mutations has been described as an early mechanism of lung cancer development [].
In summary, expression of a surprisingly similar pattern of prognostically relevant hESC-signature genes in BC-S and human lung carcinomas, observed in this study, provides transcriptome-based evidence for a novel model of lung cancer development in which selective smoking-induced reprogramming of airway BC toward the hESC-like phenotype represents a common molecular event in the pathogenesis of all four major types of lung cancer, contributing to the molecular phenotype of aggressive lung carcinomas.
We thank B-G. Harvey, R.J. Kaner, A.E. Tilley, M.W. Butler, and M. O'Mahony for help in obtaining the large airway epithelium samples; M. Ladanyi for making the Memorial Sloan-Kettering Cancer Center adenocarcinoma dataset available to us; P. Karp and M. Welsh, University of Iowa, for the protocol on culturing primary airway epithelial cells; J. Fuller for coordinating the cancer sample database, B. Ferris and B. Witover for their help in basal cell characterization; M. Al-Hijji, P. Bonsu, and T. Fukui for help with in vitro experiments; D. Dang and M. Teater for sample processing for microarray analysis; N. Mohamed and D.N. McCarthy for help in preparing this manuscript. These studies were supported in part by P50 HL084936 and R01 HL107882, Starr Foundation/Starr Cancer Consortium, and UL1-RR024996. R.S. was supported in part by the Parker B. Francis Foundation.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.