Gene Expression Profiling of Cultured Human Bronchial Epithelial and Lung Carcinoma Cells

Gary M. Hellmann,1, Wanda R. Fields and David J. Doolittle

Biological Research, Bowman Gray Technical Center, R. J. Reynolds Tobacco Company, Winston-Salem, North Carolina 27102

Received August 23, 2000; accepted February 5, 2001


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Lung cancer is a complex collection of diseases that is thought to begin with single mutated progenitor cells and culminates in any of several clinically described pathologies. Our knowledge of the molecular events that lead to different lung cancer types—small cell carcinoma, squamous cell carcinoma, adenocarcinoma, and large cell carcinoma—is incomplete. Nonetheless, it is evident that genetic changes that impact multiple molecular networks are involved in the generation of each specific phenotype. Due to the obvious complexity of these processes, the simultaneous quantitative monitoring of changes in the expression of genes that define these networks can provide mechanistic information to increase our understanding of the molecular basis for human pulmonary carcinogenesis. To this end, we have employed a commercially available human cDNA array (AtlasTM Human Array, Clontech Laboratories) to systematically screen for alterations in the expression of 600 genes in normal human bronchial epithelial (NHBE) cells as well as in several lung carcinoma lines. Studies on the reproducibility and variability of array results indicate that a 2-fold or greater difference in the expression of a particular gene could be considered a real difference in transcript abundance. Accuracy of gene expression as measured in the array was verified by comparing mRNA levels of the proto-oncogene c-myc in the array with results obtained by traditional Northern blot analysis and by quantitative RT-PCR. Gene expression profiles were compared within and among cell types. The differential expression of 17 genes, including downregulation of MRP8 and MRP14 and upregulation of CYP1B1, was observed in all four carcinoma lines compared to NHBE cells. The direction of all 17 gene expression differences, either upregulation or downregulation relative to NHBE cells, was the same for all four carcinoma lines, underscoring their common molecular features. Each lung tumor line also exhibited a number of unique differences compared to both normal cells and the other tumor cell lines. These differences may be due to differences in the cellular origin and/or pathology of the cell lines studied.

Key Words: gene expression; array; lung carcinogenesis; mRNA; expression profiling; human cell culture; bronchial epithelial cells..


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
It is now evident that most diseases, including cancers, are associated with a series of changes in the expression of specific genes. Lung cancer, one of the leading causes of cancer death in the world, is believed to be the result of a series of genetic alterations that accumulate over a period of time in pulmonary epithelial cells. These alterations can lead to changes in the expression of both mutated and nonmutated genes. The term "lung cancer" is a generalized term describing a heterogeneous disease manifest as a number of different malignant neoplasms that can occur throughout the pulmonary tree (Minna et al., 1989Go). The disease is generally subdivided into two large groups: small cell lung cancer (SCLC) which accounts for about 20–25% of bronchogenic carcinomas, and non–small cell lung cancer (NSCLC), which accounts for the remainder. Small cell carcinomas are believed to arise via the malignant transformation of neuroendocrine cells (Carney and Minna, 1982Go; Gould et al., 1983Go). NSCLC can be further divided into three major subtypes: large cell lung carcinoma (LCLC), adenocarcinoma, and squamous cell carcinoma. Large cell lung carcinoma is considered an indeterminate diagnosis and may represent highly undifferentiated squamous carcinomas and adenocarcinomas (Robbins et al., 1984Go). Adenocarcinoma can be either bronchial derived or bronchioalveolar in origin (Robbins et al., 1984Go). In contrast, squamous cell carcinoma results from malignant transformation of epithelial cells lining the bronchi. Although each of these types share a common set of characteristics, each is believed to arise and diverge from a different pathway of mutation and gene expression changes.

Cell lines derived from each of the major tumor types are in wide use as experimental models in cancer biology, both for their utility in the investigation of new drugs and for insights they can provide in understanding the molecular changes that occur during tumorigenesis (e.g., Jiang and Yang, 1999; Kalemkerian et al., 1994; Katabi et al., 1999; Varsano et al., 1998). We have assembled lung tumor cell lines that are representative of each of the four major lung cancer types and have used them in a comparative study with normal human bronchial epithelial cells.

Recent advances in human genomics have provided a large and growing repertoire of gene sequences for use in assessing mRNA levels, an important parameter for measuring gene expression status. This new information has been used in the construction of cDNA arrays that now permit the simultaneous monitoring of the expression of genes involved in such functions as cell cycle regulation, DNA synthesis and repair, cell adhesion, cell–cell communication, and stress response. Monitoring changes in the expression of these types of genes can provide mechanistic clues for all stages of a disease process, from the initial response of cells to environmental insult to permanent genetic change, as in the case of carcinomas. We have employed a commercially available human cDNA array system to screen for alterations in the simultaneous expression of 600 genes and to monitor the relevant differences.

The objective of this study was threefold: a) to examine overall gene expression differences between normal bronchial epithelial cells and lung tumor cells; b) to document gene expression differences shared among the lung tumor lines, and c) to identify differences unique to each of the tumor lines. The results of this study will assist in elucidating the molecular and/or genetic alterations that may result in the development of specific types of human lung tumors.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Lung cells and cell lines.
Normal human bronchial epithelial (NHBE) cells from a 20-year-old Caucasian male donor were obtained from BioWhittaker, Inc. (San Diego, CA). The four immortalized human lung tumor cell lines used in this study were obtained from ATCC (Rockville, MD). The small cell lung carcinoma line (NCI cell line designation NCI-H82; ATCC catalog number HTB-175) was established from tissues obtained from a 40-year-old Caucasian male. This line is classified as a variant SCLC line by virtue of its rapid doubling time, its overexpression of c-myc due to DNA amplification, and low levels of neuroendocrine markers (Kalemkerian et al., 1994Go). The adenocarcinoma line (ATCC catalog number CCL-185; ATCC line designation A549) was established from tissues obtained from a 58-year-old Caucasian male. The adenosquamous cell carcinoma line (NCI: NCI-H596, ATCC: HTB-178) was established from tissues obtained from a 73-year-old Caucasian male. The large cell carcinoma line (NCI: NCI-H460, ATCC: HTB-177) was established from tissues obtained from a 40-year-old Caucasian male.

Cells were typically grown in T75 flasks or 100-mm dishes and maintained in humidified incubators at 37°C and 5% CO2. The suppliers' suggested media were used to maintain optimal growth conditions for each cell type. A549 cells were cultured using Kaighn's Modified F12 medium (JRH Biosciences, Lenexa, KS) supplemented with 10% fetal bovine serum (FBS), 50 µg/ml gentamycin and 2 mM glutamine. The NCI-H82, NCI-H460, and NCI-H596 cells were cultured using RPMI medium (JRH Biosciences) supplemented with 10% FBS, 50 µg/ml gentamycin, and 2 mM glutamine. Finally, NHBE cells were cultured using a modified LHC-9 medium (BioWhittaker, Inc.) supplemented with 52 µg/ml bovine pituitary extract, 0.5 µg/ml hydrocortisone, 0.5 ng/ml human epidermal growth factor, 0.5 µg/ml epinephrine, 10 µg/ml transferrin, 5 µg/ml insulin, 0.1 ng/ml retinoic acid, 6.5 ng/ml triiodothyronine, 50 µg/ml gentamycin and 50 ng/ml amphotericin-B. The media was replenished every 2–3 days for all cell types, and the cells were harvested in logarithmic growth phase. Because a number of gene expression changes occur during the process of differentiation (Freytag, 1988Go; Olsen et al., 1995Go; Shichiri et al., 1993Go), NHBE cells were strictly cultured in the growth medium (BioWhittaker, Inc.) described above to help maintain the cells in an undifferentiated state (Lechner and LaVeck, 1985Go) and used no later than passage six. When cultures reached 60–80% confluence, cells were harvested by trypsinization, pelleted, then frozen at –80°C prior to RNA isolation.

RNA isolation.
Isolation of total RNA using TRIClon C reagent was conducted according to the manufacturer's recommendations (Clontech Laboratories, Inc., Palo Alto, CA). Briefly, 1 ml of TRIClon C reagent was used to homogenize approximately 2 x 106 cells. Samples were lysed by rapid up and down pipetting and subsequently incubated at room temperature for 5 min. The samples were then extracted with 0.2 ml chloroform by inversion mixing. Phase separation was achieved by placing the samples at room temperature for 3 min, then centrifuging at 12,000 x g and 4°C for 15 min. The aqueous layer was mixed with 0.5 ml of isopropanol to precipitate the RNA. Samples were incubated at room temperature for 10 min, then centrifuged at 12,000 x g and 4°C for 10 min. The RNA pellet was washed with 1 ml of 75% ethanol, then centrifuged at 7,500 x g and 4°C for 5 min. The pellet was air dried, resuspended in nuclease-free water and quantified using a Beckman DU-650 spectrophotometer (Beckman Instruments, Inc., Fullerton, CA). Formaldehyde-agarose gel electrophoresis was performed to assess the integrity of the RNA preparations.

cDNA synthesis and array hybridization.
32P-labeled cDNAs were synthesized using reagents and protocols provided with the AtlasTM Human cDNA Expression Array I system (Clontech) except as noted. The array consists of 609 individual addresses: 588 experimental sequences, 9 housekeeping sequences, 3 nontarget DNA sequences printed in triplicate, and 3 blank addresses. Briefly, for each sample, 5 µg total RNA was annealed to a mixture of 600 cDNA primers, one for each different cDNA sequence immobilized on the array membrane, by incubation at 70°C for 2 min. cDNA synthesis was performed in 20-µl reactions containing 1X RT reaction buffer (50 mM Tris pH 8.3, 75 mM KCl, 3 mM MgCl2), 5 mM DTT, 200 units MMLV-RT (Promega Corporation, Madison, WI), 500 µM each dCTP, dGTP, dTTP, and 50 µCi [32P]dATP (3000 Ci/mmol; NEN Life Science Products, Boston, MA) with incubation at 50°C for 30 min. 32P-cDNA products were separated from unincorporated nucleotides by chromatography on Chromaspin-200 columns. Fractions containing [32P]-cDNA products were determined by scintillation counting, pooled, then mixed with 5 µg Cot-1 DNA to minimize nonspecific hybridization, and denatured at 98°C for 2 min. AtlasTM Human cDNA Expression arrays were prehybridized in a hybridization oven (Robbins Scientific, Sunnyvale, CA) at 68°C using two successive 30-min incubations. Each incubation was performed with 9-ml portions of ExpressHybTM (Clontech) buffer containing 100 µg/ml denatured salmon testes DNA (Sigma, St. Louis, MO). Following the second prehybridization period, the buffer was discarded and replaced with 6 ml of fresh buffer. The denatured cDNA mixture was added and incubation continued overnight. Membranes were washed extensively under low and high stringency conditions according to the supplier's recommendations. After the final wash, the membranes were washed in 0.3 M sodium chloride/0.03 M sodium citrate, pH 7.0, allowed to dry thoroughly, and exposed to storage phosphor screens (Molecular Dynamics, Sunnyvale, CA) for 5 days.

Quantitation and comparison of gene expression profiles.
Exposed storage phosphor screens were scanned at 50 micron resolution using a Storm 860 phosphorimager (Molecular Dynamics). Using ImageQuant 5.0 software (Molecular Dynamics), a series of grids was constructed and superimposed over the blot such that each duplicate immobilized DNA target was contained within a cell. Total counts for each cell in the grid were determined. The numerical output was exported to Microsoft Excel for comparisons and identification of differentially regulated genes. Blots were normalized to the sum of counts in all cells without background subtraction, and the ratio of tumor line/NHBE determined for each tumor line and each gene. The log base 2 of each value was determined in order to equalize the magnitude of deflection of upregulated and downregulated genes, and gene expression differences were ranked based on absolute values. Genes indicating a difference (up or down) of 2-fold or greater were verified by visual inspection of a printout of the phosphorimage. A list of the genes included on the array can be found on Clontech's web site (http://atlas.clontech.com/).

Northern blot and TaqManTM RT-PCR analysis.
To compare the quantitation of the cDNA arrays with other more widely used methods, the level of c-myc mRNA was determined by two alternate methods: Northern blot analysis and a quantitative RT-PCR technique (TaqManTM).

Northern blots were prepared using 3 µg per lane of the same total RNA preparations used for array analysis. Gels, transfers, and hybridizations were performed with NorthernMax-Plus reagents and protocols (Ambion, Inc., Austin, TX). A gel-purified, 477-bp radiolabeled c-myc cDNA fragment used as probe was prepared using the Random Primers Labeling Kit (Gibco/BRL, Gaithersburg, MD). Membranes were exposed overnight to storage phosphor screens and were scanned at 50 micron resolution using a Storm 860 phosphorimager (Molecular Dynamics). Signal quantification was performed using ImageQuant 5.0 software (Molecular Dynamics).

TaqManTM analysis of c-myc mRNA levels was performed as described (Fields et al., 1999Go). Briefly, total RNA (10 ng) was amplified using Access RT-PCR Core reagents (Promega) in the presence of 200 nM each c-myc forward and reverse primer (Life Technologies, Inc., Grand Island, NY) and 50 nM c-myc fluorogenic probe (Applied Biosystems, Foster City, CA). Agarose gel analysis was used to confirm the generation of the 478 and 295 RT-PCR products for c-myc and ß-actin, respectively. Gene-specific detection and quantitation of the amplicon was performed with a TaqManTM–Luminescence Spectrometer (LS50B) Fluorescence Detection System (Applied Biosystems).

The TaqManTM probe consists of an oligonucleotide (20–30 bp) homologous to a portion of the target gene and contains covalently attached reporter (FAM: 6-carboxy-fluorescein) and quencher (TAMRA: 6-carboxy-tetramethyl-rhodamine) dyes. FAM and TAMRA excite at a wavelength of 488 nm; however, the dyes emit at different wavelengths—FAM at 518 nm and TAMRA at 588 nm. The level of a given mRNA was determined by the annealing of a gene-specific probe to the cDNA copy of the mRNA. During the extension phase of PCR, the fluorescence signal was liberated by the 5'-3' exonuclease-induced release of the reporter dye from the probe and was subsequently quantified spectrophotometrically. The value determined for c-myc mRNA was divided by that determined for ß-actin mRNA to yield a relative value for c-myc mRNA levels. Analysis of the data was performed by two-way ANOVA using values from at least three separate experiments.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Analysis of Array Reproducibility
Two tests were used to determine the reproducibility of results obtained from the arrays. In the first test, we compared expression profiles of duplicate blots probed with aliquots from the same pool of [32P]-labeled cDNA. A second test was performed by preparing separate blots from total RNA preparations obtained from NHBE cells harvested roughly 2 weeks apart and representing passage numbers 4 and 6. In each test, single arrays were performed for each sample.

Reproducibility was analyzed by normalizing control (NHBE) and experimental (tumor cell line) blots to the sum of the values in each of the 609 cells, then determining the ratio of each individual gene from each of the two blots in the absence of any background subtraction. The first test examined variability due to the printed arrays themselves, as well as variability contributed by all steps from hybridization through quantitation. Of the 600 genes analyzed, 591 (98.5%) displayed variation of less than 30% between the two blots. No genes displayed greater than 2-fold variation (data not shown).

The second test examined variability in mRNA levels between cell culture passage numbers 4 and 6, any unintended variability in the culturing of the cell lines, and variability in RNA preparation, cDNA synthesis, etc. Of the 600 genes analyzed, 96% displayed variation of 30% or less. A single gene (for Interleukin I receptor type II) displayed variation greater than 2-fold (2.32-fold), while approximately 4% of the genes displayed variation between 30% and 2-fold. An x,y scatterplot of the data revealed a high degree of correlation between expression levels obtained for RNA obtained from the two passages (Fig. 1Go). In all cases, the relatively low printing density of the array provided results that were easily verified by visual inspection, and any spurious background signals were easily checked by examination of phosphorimager printouts whose overall intensity levels had been normalized based on the sum of the values in all cells. Based on the results of these reproducibility and variability studies, we set a minimum variation threshold of 2-fold (absolute value of log2 of ratios = 1.0) to flag genes with sufficient difference in expression level to be considered biologically significant.



View larger version (20K):
[in this window]
[in a new window]
 
FIG. 1. Scatterplot analysis of variation in cell culture gene expression data. Comparison of NHBE gene expression levels from RNA samples obtained from two separate cell passages are shown. Values (represented as log10 of phosphorimager numerical output) from one sample are plotted on the x-axis; values of corresponding genes from the other sample are plotted on the y-axis. A regression line indicates the degree of correlation between the two samples. Broken lines indicate 2-fold upregulation or downregulation.

 
Comparison of Array Quantitative Data with Northern Blotting and TaqManTM Analysis
In order to evaluate the accuracy of the quantitative predictions of the array system, we compared relative levels of c-myc mRNA determined by the array with values obtained by both traditional Northern blot analysis and a quantitative technique referred to as TaqManTM RT-PCR methodology. Results indicate reasonably good agreement among the three methods (Fig. 2Go). Comparisons between methods revealed differences in c-myc levels varying by not more than 30% across the three methods for the high c-myc–expressing NCI-H82 line. It should be pointed out that normalization is performed in a different manner for each of the three methods. The TaqManTM RT-PCR methodology used here expresses the levels of genes relative to the level of ß-actin mRNA; the Northern blot provides values relative to the mass of total RNA (equal amounts of total RNA loaded onto each lane), while the array is normalized to the sum of the expression levels of 600 genes. Differences in normalization schemes could, in some instances, lead to significantly different results.



View larger version (21K):
[in this window]
[in a new window]
 
FIG. 2. Quantitation of c-myc mRNA levels by three different methods. Levels of c-myc mRNA for each of the lung tumor lines are relative to the level in NHBE cells determined by the same method. Scale indicates fold change.

 
Lung Tumor Cell Lines Exhibit Significant Differences from NHBE Controls
Profiles obtained for each of the four lung tumor lines were compared with the profile obtained from NHBE cells. In general, there was a surprisingly large number of differences observed (Table 1Go). In the A549 line, 56 genes (9.3% of the 600 genes analyzed) displayed a greater than 2-fold difference relative to levels in NHBE cells, with 10 of those genes displaying a difference greater than 5-fold. Almost half (44.6%) of the genes showing differences from NHBE cells were downregulated in the A549 cells. The largest number of expression differences occurred in the category that included cell–cell communication, cytokine, and interleukin genes. The other cell lines, NCI-H460, NCI-H82, and NCI-H596 appeared even less similar to NHBE cells than did the A549 line. Each displayed a similar number of differences from NHBE cells with 153, 132, and 130 genes, respectively, exhibiting a difference greater than 2-fold. Compared to the A549 line, each had a higher percentage of genes exhibiting a greater than 5-fold difference from NHBE cells, and a lower percentage of genes that were downregulated from the levels observed in NHBE cells. Each of these three lines displayed the greatest number of differences in the category that included apoptosis-related, DNA synthesis, repair, and recombination-related genes. The marked differences in mRNA expression levels is best illustrated in x,y scatterplot comparisons of gene expression values between NHBE cells and each of the four lung tumor lines (Fig. 3Go).


View this table:
[in this window]
[in a new window]
 
TABLE 1 Summary of Gene Expression Differences between NHBE Cells and Lung Tumor Lines
 


View larger version (43K):
[in this window]
[in a new window]
 
FIG. 3. Scatterplot analysis of variation between lung tumor line and NHBE gene expression data. Log10 of expression levels are shown. A. Comparison of NHBE gene expression levels (x-axis) to those from adenocarcinoma line A549 (y-axis). B. Comparison of NHBE gene expression levels (x-axis) to those from adenosquamous carcinoma line NCI-H596 (y-axis). C. Comparison of NHBE gene expression levels (x-axis) to those from SCLC line NCI-H82 (y-axis). D. Comparison of NHBE gene expression levels (x-axis) to those from LCLC line NCI-H460 (y-axis). Broken lines indicate 2-fold upregulation or downregulation.

 
Lung Tumor Lines Possess Many Shared Differences from NHBE Cells
It might be expected that lung tumor lines of different histopathological origin would share a number of similar molecular characteristics. Genes showing greater than 2-fold differences from controls were compared across tumor lines. Seventeen gene expression differences were identified that were common to all four lines (Fig. 4Go). The direction of gene expression difference for each gene, either upregulation or downregulation relative to NHBE cells, was the same for all four lines, underscoring common molecular features of the tumor lines. The largest magnitude of difference observed was the significant downregulation of the two calgranulin genes, MRP-8 and MRP-14. Another cytoskeleton-associated gene, integrin beta-4, also displayed significant downregulation in all tumor lines. The downregulation of these genes may be manifestations of general perturbations of the cytoskeletal framework in tumor cells.



View larger version (53K):
[in this window]
[in a new window]
 
FIG. 4. Gene expression differences from NHBE cells common to all tumor lines. Values indicate difference in mRNA level (fold) relative to NHBE cells. Black numerals in white boxes, downregulated; white numerals in dark gray boxes, upregulated.

 
Among the nonstructural gene expression differences shared by all of the tumor lines was upregulation of cytochrome P450 CYP1B1, a dioxin-inducible form of the cytochrome P450 family of xenobiotic metabolizing enzymes. In addition, cyclin-dependent kinase inhibitor 1A (p21; Cip-1; WAF-1) was significantly downregulated in all tumor lines.

In addition to the 17 gene expression differences common to all four tumor lines, 75 gene expression differences were common to three of four lines, with 65 of those differences common to the NCI-H82, NCI-H460, and NCI-H596 lines (Fig. 5Go). The largest differences were upregulation of RACH1 and M-CSF1 and the downregulation of integrin beta-1. Seventy-one of the 75 genes exhibited change in a common direction. Of the 75 genes, 10 were downregulated and 65 were upregulated.



View larger version (118K):
[in this window]
[in a new window]
 
FIG. 5. Gene expression differences from NHBE cells common to three of four tumor lines. Values indicate difference in mRNA level (fold) relative to NHBE cells. Black numerals in white boxes, downregulated; white numerals in dark gray boxes, upregulated.

 
To further investigate relationships among lines based on this set of 600 genes, cluster analysis was performed. Average linkage clustering was performed using SAS® (SAS Institute, Cary, NC) software incorporating the correlation metric described by Eisen et al., 1998. The analysis identified NCI-H82, NCI-H460, and NCI-H596 as forming a cluster separate from the A549 and NHBE cells (Fig. 6Go). Interestingly, expression levels for the A549 cell line were more highly correlated with those of normal cells than with those of the first cluster, and thus the analysis combined the A549 and normal cells in a separate cluster.



View larger version (9K):
[in this window]
[in a new window]
 
FIG. 6. Dendrogram showing hierarchical clustering of lung tumor cell lines and NHBE cells. Clustering was based on the gene expression profiles for 588 genes using average linkage clustering and a correlation metric according to the method of Eisen et al. (1998). The scale represents the distance metric, 1-r, where r is the Pearson correlation coefficient.

 
Different Lung Tumor Lines Exhibit Unique Differences from NHBE Cells and from Each Other
An examination of tumor line gene expression differences from controls (NHBE cells) revealed a number that were unique to a given cell line. The four genes exhibiting the greatest magnitude of difference in each cell line are listed in Fig. 7Go.



View larger version (45K):
[in this window]
[in a new window]
 
FIG. 7. Gene expression differences unique to a given cell line (four genes displaying largest magnitude of difference only). Values indicate difference in mRNA level (fold) relative to NHBE cells. Black numerals in white boxes, downregulated; white numerals in dark gray boxes, upregulated.

 
In the adenocarcinoma line A549, neuroleukin mRNA levels exhibited a 4.3-fold upregulation. Also upregulated were mRNAs for contactin, a neuronal adhesion molecule found highly expressed in human neuroblastoma cell lines (Reid et al., 1994Go), and fibroblast growth factor receptor-1, the receptor for basic fibroblast growth factor.

The small cell lung carcinoma line NCI-H82 similarly displayed a number of unique differences, including upregulation of c-fms proto-oncogene and downregulation of MIP2{alpha} mRNAs.

The LCLC line NCI-H460 displayed the lowest percentage of unique differences (19.6%), none of which were greater than 3-fold different from NHBE controls. The TGF-ß–inducible follistatin-related protein mRNA was upregulated approximately 2.8-fold.

Among differentially regulated genes unique to the NCI-H596 cell line, the largest magnitude of difference in expression level was a 4.6-fold upregulation displayed by interferon regulatory factor-1 (IRF-1). A 4-fold downregulation of the mRNA for 60S ribosomal protein L6 was also observed, although this protein has been reported to be upregulated in rapidly growing thyroid cells (Ohta et al., 1994Go).


    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Considering the complexity of pulmonary carcinogenesis, methods that allow simultaneous analysis of multiple molecular events offer much promise in providing insight into the complex network of cellular regulatory pathways and how they participate in the generation of tumor phenotypes. The parallel approach offered by cDNA expression array technology provides such a method, one which has already found use in a variety of applications from gene discovery to toxicology (Farr and Dunn, 1999Go). Recent years have witnessed an explosion in developments in array technology itself and its use in examining a number of biological processes including cancer development (Szallasi, 1998Go; Zhang et al., 1997Go; Zhang et al., 1998Go).

Taken as a whole, the data obtained from gene expression profiling of the four lung tumor lines relative to that of normal lung bronchial epithelial cells illustrate a complex picture of lung tumor biology. Even among the limited number of genes included on the arrays used in these experiments, perturbation of a significant percentage was observed. Sources for gene expression differences include the inherent genetic complement of the donor, genetic mutation, DNA amplification, and downstream effects where gene expression levels are influenced by upstream changes. An impressive list of differences was common to the lung tumor lines, reflecting shared functional and organizational differences from normal cells. These genes fell into a number of functional classes, underscoring the diverse changes that occur in tumors.

The genes exhibiting the greatest magnitude of change, MRP-8 and MRP-14, are related to cytoskeletal organization. The protein products of these genes form calcium-binding heterodimers and may be involved in the reorganization of cytoskeletal filaments (Goebeler et al., 1995Go). It is well known that cancerous cells display a perturbed cell morphology and that cytoskeletal rearrangements are coincident with metastatic transmigration (Voura et al., 1998Go). Similarly, downregulation of integrins beta-4 and beta-1 have been identified as hallmarks of cervical cancer (Shim et al., 1998Go). Changes in integrin-mediated cell adhesive interactions and cytoskeletal organization have been implicated in processes such as tumor cell growth and metastasis (Gille and Swerlick, 1996Go). The observation of such a dramatic downregulation of MRP8 and MRP14 raised the possibility that, rather than observing downregulation of these genes (as well as integrin ß1) in the tumor lines, upregulation of these genes was occurring in the NHBE controls. Significant upregulation of these genes has been reported in keratinocytes when cultured in media containing 10% fetal calf serum, conditions that favor differentiation (Olsen et al., 1995Go). However, NHBE cells were specifically cultured in serum-free media and used no later than passage 6 to help minimize such differentiation. In addition, other genes reported as being aberrantly regulated by differentiation-inducing media, such as CD41, were either unresponsive or upregulated in tumor cells. Finally, gene expression profiles of uncultured total lung RNA revealed levels of MRP8 and MRP14 comparable to those observed in NHBE cells (data not shown). As the gene products of MRP8 and MRP14 organize in the cell as heterodimers, it was reassuring to observe similar levels of difference for these two genes. Likewise, the observed consistency in the direction of change for virtually all perturbed genes lends similar confidence to the results. This highlights one of the powers of arrays—to provide internal molecular network cross-checks.

Other observed differences between tumor lines and NHBE cells were consistent with what is known regarding molecular dysfunction in tumor cells. For example, CYP1B1, upregulated in all four tumor lines, has been found to be expressed at high levels in a number of human cancer types including lung cancer (Murray et al., 1997Go). In addition, cyclin-dependent kinase inhibitor 1A (p21; Cip-1; WAF-1), which was downregulated in all tumor lines, is a negative regulator of the cell cycle and has been shown to be involved in G1 arrest as well as suppression of tumor growth (Gartel et al., 1996Go). Downregulation of the p21 gene may represent a compromise in cell cycle controls characteristic of the tumor phenotype.

Expression differences unique to a given line were also consistent with previous reports of aberrant gene regulation in tumors. Neuroleukin mRNA, upregulated only in the adenocarcinoma line A549, encodes a tumor-secreted cytokine that stimulates cell migration in vitro and metastasis in vivo. It is also considered a cancer progression marker (Watanabe et al., 1996Go). mRNA levels for neuroleukin were observed to be elevated 4-fold in human fibroblast sarcoma cells relative to embryonal fibroblasts (Niinaka et al., 1998Go). Elevated expression of fibroblast growth factor receptor-1 was observed in epithelial tumors including lung adenocarcinoma (Morikawa et al., 1996Go). Unique to NCI-H82, c-fms proto-oncogene overexpression has been correlated with increased cell proliferation and aggressive tumor formation (Keshava et al., 1999Go), and MIP2{alpha}, also termed gro-beta, is an angiogenesis inhibitor shown to inhibit tumor growth in nude mice (Cao et al., 1995Go). Finally, interferon regulatory factor-1, upregulated in the NCI-H596 line, is a key regulator of the interferon system and has antiproliferative and tumor suppressor functions (Vaughan et al., 1997Go).

Based on global expression differences among the tumor lines, A549 might be considered most like NHBE cells, as it exhibited the fewest overall number of differences. Even though this line exhibited a fewer number of differences from NHBE cells, a high percentage (35%, not shown) of these differences were not shared among the other tumor lines, making A549 most unlike the other lines on the basis of the surveyed genes. NCI-H82, NCI-H460, and NCI-H596 displayed a striking relatedness to each other given the number of gene expression differences in common. Different views of the nature of these relationships were afforded by three separate methods: the x,y scatterplot analysis exhibited in Figure 3Go, a tabular analysis of gene expression differences (Figs. 4 and 5GoGo), and cluster analysis (Fig. 6Go). Each of these methods provided a consistent picture of the relatedness of each tumor line to the others and to NHBE cells. In spite of numerous similarities, each line possessed a number of unique characteristics to make tumor type fingerprinting a possibility.

In a number of instances, the direction of change in the expression of a given gene was not anticipated based on current knowledge of the role of that gene in tumorigenesis. For example, follistatin-related protein, thought to be involved in negative regulation of cell growth (Mashimo et al., 1997Go), was upregulated in NCI-H460, and interferon regulatory factor 1 (IRF-1) was upregulated in NCI-H596. It should be pointed out that there is a less than perfect correlation between the level of a given mRNA and the ultimate biological activity of its gene product. Moreover, IRF-2 is an antagonist of IRF-1 and can repress activation of the interferon-beta gene by IRF-1, thus exerting fine control over cell growth. (cDNA sequences for the IRF-2 gene were not present on the array.)

In the current study, the cDNA expression array platform produced reproducible results and exhibited quantitative power comparable to other techniques commonly used to determine mRNA levels. Others have reported similar agreement among cDNA array, Northern blot, and RT-PCR–based quantitative data (Backert et al., 1999Go). The merits and limitations of this type of platform (membrane-based, medium gene density, 32P-labeling of cDNAs) as compared to alternatives such as glass slides with fluorescent cDNA labeling or DNA chips with immobilized DNA oligonucleotides, have been examined extensively (Bertucci et al., 1999Go; Granjeaud et al., 1999Go) and so will not be revisited here.

Currently, there is no single universally adopted systematic method for the analysis of array data. A conservative approach was adopted for this study in which backgrounds were not subtracted from gene expression values prior to comparative analysis. This necessitated the use of blots possessing a uniform background for quantitation. Obviously, genes with low mRNA abundance are most affected by inclusion of background, the effect being to dampen out differences between small numbers. This choice runs the risk of underreporting differential expression among genes with low mRNA abundance, but prevents erroneous reporting of large differences between expression levels visible only after the subtraction of a large background. In addition, blots were normalized to the sum of expression values for all genes as a more accurate equalizer of unperturbed expression levels than the housekeeping genes included on the arrays. As has been previously noted, many genes commonly used for normalization, in fact, vary considerably under different experimental treatments (Spanakis, 1993Go). It should be noted that among differences common to all four cell lines were two of the so-called housekeeping genes- ubiquitin and phospholipase A2.

In spite of their utility in cancer biology, care must be taken in interpreting gene expression differences in immortalized lines. Profiles obtained for these four lung tumor lines may not necessarily be duplicated in primary tumors even of the same histopathological type. A recent study has indicated that immortalized lines can retain many of the properties exhibited from the parental primary tumor (Wistuba et al., 1999Go). However, it seems reasonable that through many generations, traits that make a line amenable to ease of manipulation in culture will be selected during repetitive subculturing, and these traits may represent drift from the original cell population. It is also likely that some gene expression differences may be attributed to factors that may not be direct reflections of the tumor phenotype. Such variables as the age, health, sex, and individual genetic polymorphisms of the donor will influence the expression profile. For example, NCI-H82 is considered an SCLC line but is classified as a variant cell line due to low levels of neuroendocrine markers, rapid doubling time, and c-myc gene amplification and overexpression. Additionally, it should be pointed out that the tumor lines are by and large genetically stable and composed of a homogeneous population of cells—features generally not shared with primary tumors.

An important next step will be a survey of primary tumors, and even more importantly, successive profiles of each stage of tumor development. Nonetheless, the identification and understanding of the functions of gene expression differences shared by the tumor lines give insight into general features of the tumor phenotype, while expression differences unique to a given line give insight into the genetic changes that have resulted in the generation of that particular tumor type. The characterization of gene expression differences observed in these four tumor lines, both shared and unique, give promise for a better mechanistic understanding of tumor biology and tumor progression.


    ACKNOWLEDGMENTS
 
The authors are deeply grateful to Mr. Walter T. Morgan for expert assistance with cluster analysis and advice regarding statistical analysis of gene expression data, and to Mr. Joseph Desiderio for superb technical assistance with culturing of cell lines.


    NOTES
 
1 To whom correspondence should be addressed at R. J. Reynolds Tobacco Company, Environmental and Molecular Toxicology Division, Winston-Salem, NC 27102. Fax: (336) 741-1321. E-mail: hellmag{at}rjrt.com. Back


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Backert, S., Gelos, M., Kobalz, U., Hanski, M. L., Bohm, C., Mann, B., Lovin, N., Gratchev, A., Mansmann, U., Moyer, M. P., Riecken, E. O., and Hanski, C. (1999). Differential gene expression in colon carcinoma cells and tissues detected with a cDNA array. Int. J. Cancer 82, 868–874.[ISI][Medline]

Bertucci, F., Bernard, K., Loriod, B., Chang, Y.-C., Granjeaud, S., Birnbaum, D., Nguyen, C., Peck, K., and Jordan, B. R. (1999). Sensitivity issues in DNA array-based expression measurements and performance of nylon microarrays for small samples. Hum. Mol. Genet. 8, 1715–1722.[Abstract/Free Full Text]

Cao, Y., Chen, C., Weatherbee, J. A., Tsang, M., and Folkman, J. (1995). gro-beta, a -C-X-C- chemokine, is an angiogenesis inhibitor that suppresses the growth of Lewis lung carcinoma in mice. J. Exp. Med. 182, 2069–2077.[Abstract]

Carney, D. N., and Minna, J. D. (1982). Small cell cancer of the lung. Clin. Chest Med. 3, 389–398.[ISI][Medline]

Eisen, M. B., Spellman, P. T., Brown, P. O., and Botstein, D. (1998). Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. U.S.A. 95, 14863–14868.[Abstract/Free Full Text]

Farr, S., and Dunn, R. T., II. (1999). Concise review: gene expression applied to toxicology. Toxicol. Sci. 50, 1–9.[Free Full Text]

Fields, W. R., Bombick, D. W., Desiderio, J. G., and Doolittle, D. J. (1999). Comparison of c-myc expression in normal human bronchial epithelial cells and lung cancer cells lines using a quantitative fluorescence-based reverse transcriptase-polymerase chain reaction mechanism. Toxicol. Methods 9, 173–188.[ISI]

Freytag, S. O. (1988). Enforced expression of the c-myc oncogene inhibits cell differentiation by precluding entry into a distinct predifferentiation state in G0/G1. Mol. Cell. Biol. 8, 1614–1624.[ISI][Medline]

Gartel, A. L., Serfas, M. S., and Tyner, A. L. (1996). p21- negative regulator of the cell cycle. Proc. Soc. Exp. Biol. Med. 213, 138–149.[Abstract]

Gille, J., and Swerlick, R. A. (1996). Integrins: Role in cell adhesion and communication. Ann. N. Y. Acad. Sci. 797, 93–106.[Abstract]

Goebeler, M., Roth, J., van den Bos, C., Ader, G., and Sorg, C. (1995). Increase of calcium levels in epithelial cells induces translocation of calcium-binding proteins migration inhibitory factor-related protein 8 (MRP8) and MRP14 to keratin intermediate filaments. Biochem. J. 309, 419–424.[ISI][Medline]

Gould, V. E, Linnoila, R. I., Memoli, V. A., and Warren, W. H. (1983). Neuroendocrine components of the bronchopulmonary tract: Hyperplasias, dysplasia, and neoplasms. Lab. Invest. 49, 519–537.[ISI][Medline]

Granjeaud, S., Bertucci, F., and Jordan, B. R. (1999). Expression profiling: DNA arrays in many guises. Bioassays 21, 781–790.[ISI][Medline]

Jiang, H., and Yang, L. Y. (1999). Cell cycle checkpoint abrogator UCN-01 inhibits DNA repair: Association with attenuation of the interaction of XPA and ERCC1 nucleotide excision repair proteins. Cancer Res. 59, 4529–4534.[Abstract/Free Full Text]

Kalemkerian, G. P., Jasti, R. K., Celano, P., Nelkin, B. D., and Mabry, M. (1994). All-trans-retinoic acid alters myc gene expression and inhibits in vitro progression in small cell lung cancer. Cell Growth Diff. 5, 55–60.[Abstract]

Katabi, M. M., Chan, H. L., Karp, S. E., and Batist, G. (1999). Hexokinase type II: A novel tumor-specific promoter for gene-targeted therapy differentially expressed and regulated in human cancer cells. Hum. Gene Ther. 10, 155–164.[ISI][Medline]

Keshava, N., Gubba, S., and Tekmal, R. R. (1999). Overexpression of macrophage colony-stimulating factor (CSF-1) and its receptor, c-fms, in normal ovarian granulosa cells leads to cell proliferation and tumorigenesis. J. Soc. Gynecol. Invest. 6, 41–49.[ISI][Medline]

Lechner, J. F., and LaVeck, M. A. (1985). A serum-free method for culturing normal human bronchial epithelial cells at clonal density. J. Tissue Cult. Methods 9, 43–48.

Mashimo, J., Maniwa, R., Sugino, H., and Nose, K. (1997). Decrease in the expression of a novel TGF beta1-inducible and ras-recision gene, TSC-36, in human cancer cells. Cancer Lett. 113, 213–219.[ISI][Medline]

Minna, J. D., Pass, H., Glatstein, E., and Ihde, D. C. (1989). Cancer of the Lung. In Cancer, Principles and Practice of Oncology, 3rd edition. pp. 591–705. J. B. Lippincott Company, Philadelphia.

Morikawa, Y., Ishihara, Y., Tohya, K., Kakudo, K., Seo, M. K., and Matsuura, N. (1996). Expression of the fibroblast growth factor receptor-1 in human normal tissues and tumors determined by a new monoclonal antibody. Arch. Pathol. Lab. Med. 120, 490–496.[ISI][Medline]

Murray, G. I., Taylor, M. C., McFadyen, M. C., McKay, J. A., Greenlee, W. F., Burke, M. D., and Melvin, W. T. (1997). Tumor-specific expression of cytochrome P450 CYP1B1. Cancer Res. 57, 3026–3031.[Abstract]

Niinaka, Y., Paku, S., Haga, A., Watanabe, H., and Raz, A. (1998). Expression and secretion of neuroleukin/phosphohexose isomerase/maturation factor as autocrine motility factor by tumor cells. Cancer Res. 58, 2667–2674.[Abstract]

Ohta, K., Endo, T., Gunji, K., and Onaya, T. (1994). Isolation of a cDNA whose expression is markedly increased in malignantly transformed FRTL cells and neoplastic human thyroid tissues. J. Mol. Endocrinol. 12, 85–92.[Abstract]

Olsen, E., Rasmussen, H. H., and Celis, J. E. (1995). Identification of proteins that are abnormally regulated in differentiated cultured human keratinocytes. Electrophoresis 16, 2241–2248.[ISI][Medline]

Reid, R. A., Bronson, D. D., Young, K. M., and Hemperly, J. J. (1994). Identification and characterization of the human cell adhesion molecule contactin. Brain Res. Mol. Brain Res. 21, 1–8.[ISI][Medline]

Robbins, S. L., Cotran, R. S., and Kumar, V. (1984). In Pathologic Basis of Disease, 3rd edition, pp. 753–754. W. B. Saunders Company, Philadelphia.

Shichiri, M., Hanson, K. D., and Sedivy, J. M. (1993). Effects of c-myc expression on proliferation, quiescence and the G0 to G1 transition in nontransformed cells. Cell Growth Diff. 4, 93–104.[Abstract]

Shim, C., Zhang, W., Rhee, C. H., and Lee, J. H. (1998). Profiling of differentially expressed genes in human primary cervical cancer by complementary DNA expression array. Clin. Cancer Res. 4, 3045–3050.[Abstract]

Spanakis, E. (1993). Problems related to the interpretation of autoradiographic data on gene expression using common constitutive transcripts as controls. Nucleic Acids Res. 21, 3809–3819.[Abstract]

Szallasi, Z. (1998). Bioinformatics. Gene expression patterns and cancer. Nat. Biotechnol. 16, 1292–1293.[ISI][Medline]

Varsano, S., Rashkovsky, L., Shapiro, H., Ophir, D., and Mark-Bentankur, T. (1998). Human lung cancer cell lines express cell membrane complement inhibitory proteins and are extremely resistant to complement-mediated lysis; a comparison with normal human respiratory epithelium in vitro, and an insight into mechanism(s) of resistance. Clin. Exp. Immunol. 113, 173–182.[ISI][Medline]

Vaughan, P. S., van Wijnen, A. J., Stein, J. L., and Stein, G. S. (1997). Interferon regulatory factors: Growth control and histone gene regulation–it's not just interferon anymore. J. Mol. Med. 75, 348–359.[ISI][Medline]

Voura, E. B., Sandig, M., Kalnins, V. I., and Siu, C. (1998). Cell shape changes and cytoskeleton reorganization during transendothelial migration of human melanoma cells. Cell Tissue Res. 293, 375–387.[ISI][Medline]

Watanabe, H., Takehana, K., Date, M., Shinozaki, T. and Raz, A. (1996). Tumor cell autocrine motility factor is the neuroleukin/phosphohexose isomerase polypeptide. Cancer Res. 13, 2960–2963.

Wistuba, I. I., Bryant, D., Behrens, C., Milchgrub, S., Virmani, A. K., Ashfaq, R., Minna, J. D., and Gazdar, A. F. (1999). Comparison of features of human lung cancer cell lines and their corresponding tumors. Clin. Cancer Res. 5, 991–1000.[Abstract/Free Full Text]

Zhang, L., Zhou, W., Velculescu, V. E., Kern, S. E., Hruban, R. H., Hamilton, S. R., Vogelstein, B., and Kinzler, K. W. (1997). Gene expression profiles in normal and cancer cells. Science 276, 1268–1272.[Abstract/Free Full Text]

Zhang, M., Martin, K. J., Sheng, S., and Sager, R. (1998). Expression genetics: A different approach to cancer diagnosis and prognosis. Trends Biotechnol. 16, 66–71.[ISI][Medline]