Introduction

Top Abstract Introduction Materials & Methods Results Discussion References

Diagnosis of bronchogenic carcinoma currently is made by microscopic examination of fixed and stained cell and/or tissue samples. Although this method is highly sensitive and specific for the malignant phenotype, it does not provide answers regarding mechanisms of malignant transformation, provide markers that predict sensitivity to chemotherapeutic agents or radiation, or predict tendency to metastasize. Such information will be necessary before advances in treatment of bronchogenic carcinoma can take place, either by developing new therapeutic agents on the basis of the mechanisms responsible for each tumor or by selecting the best currently available treatment for each tumor on the basis of markers.

Because of recently developed, highly sensitive, quantitative reverse transcriptase polymerase chain reaction (RT-PCR) methods for measuring large numbers of genes simultaneously (1), it now is feasible to associate empirically patterns of gene expression with specific cellular phenotypes using nanogram quantities of RNA. The sensitivity of these methods is such that the amount of tissue obtained from fine-needle biopsies is enough for hundreds of gene expression measurements.

To identify gene expression patterns that characterize tumors in practical ways (e.g., sensitivity to chemo- and radiotherapy) on the basis of assessment of samples obtained through fine-needle biopsies, it will be necessary first to identify a gene expression pattern that is predictive of malignant phenotype. Because tumors are heterogeneous with respect to the amount of normal stroma within them, the fraction of tumor cells within any particular fine- needle aspirate (FNA) biopsy will vary. Furthermore, any sample used for cytologic analysis will not be available for gene expression analyses. Thus, after the biopsy sample is processed for RNA extraction, and before other gene expression patterns are evaluated, it will be necessary to ensure that a gene expression pattern specific for malignancy is present.

An effort to identify gene expression patterns specific for malignant phenotype is likely to be easier in cultured cells, under defined conditions, in the absence of normal stroma. However, there will be uncertainty regarding the relevance of such findings to primary in situ normal and malignant lung tissues. Thus any gene expression pattern associated with malignant phenotype in cultured cells will need to be validated in primary tissues.

A large number of genes involved in controlling proliferation of eukaryotic cells have been identified in the last ten years (5). Many of these genes are associated with malignant transformation of bronchial epithelial cells (BEC) as a result of either qualitative alteration (mutation in the coding region) or quantitative alteration in expression (7). We hypothesized that the relative messenger RNA (mRNA) expression of multiple genes involved in cell proliferation homeostasis would better discriminate between normal and malignant cells than the expression levels of individual genes. In an effort to test this hypothesis, recently developed methods for quantitative RT-PCR measurement of gene expression (1) were used to simultaneously evaluate 15 cell-cycle control genes in cultured normal and malignant BEC lines, all in exponential growth equilibrium. The gene expression index c-myc × E2F-1/ p21 associated with malignant phenotype in these samples was then evaluated in primary normal and malignant BEC samples, as well as normal parenchyma surrounding the tumor samples.

	Materials and Methods

Top Abstract Introduction Materials & Methods Results Discussion References

Analysis of Cultured Cells

The reagents, primers, competitive template mixtures, and cell populations were previously described (3, 4). Additional normal BEC lot numbers 17714, 6F0395, and 7F0075 (Clonetics, San Diego, CA), and tumor cell lines H661, H460, and H322 (American Type Culture Collection, Rockville, MD) were cultured as described (4).

Analysis of Primary Tissues

Normal BEC, normal lung parenchyma, and bronchogenic carcinoma tissues were obtained from patients at the time of surgery. During anesthesia, and before thoracotomy, the patients underwent bronchoscopy as standard procedure to rule out endobronchial lesions. During the bronchoscopy, bronchial brush biopsies were obtained from secondary or tertiary bronchi as described previously (2, 3). Following bronchoscopy the tumor was resected, along with sufficient adjacent normal parenchymal tissue to ensure tumor-free margins, and sent for immediate frozen section assessment by the pathologist. The tissue that remained after pathologic assessment was made available for this research project. These studies were approved by the Medical College of Ohio Internal Review Board, and all patients consented to use of their tissues in the manner described.

Tissues received from the pathologist were processed as follows. First, the areas of normal parenchyma or tumor were identified with the assistance of the pathologist, and areas with greater than 50% viable tumor as opposed to stroma or necrosis were selectively dissected, placed in separate 50-ml conical tubes on wet ice, and transferred to the laboratory. In the laboratory, each of the samples was placed independently in a stainless-steel mortar cooled with liquid nitrogen, resulting in rapid freezing. The frozen samples were pulverized with a large, liquid nitrogen-cooled porcelain pestle. The pulverized sample was placed into TRI-REAGENT in a 50-ml conical tube. Approximately 1 ml of TRI-REAGENT was used for each 100 mg of pulverized tissue. RNA was extracted according to the manufacturer's instructions, and as previously described (3, 4). Reverse transcription and quantitative PCR also were according to previously described methods (3, 4). In some cases BEC were collected by first obtaining a section of bronchus that had been resected from the tissue sample and determined by the pathologist to be free of tumor. The bronchial epithelial tissue lining the bronchus was scraped with a sterile scalpel blade into 15 ml of media in a 100-mm petri dish. The cells then were pelleted and RNA was isolated as described (4).

Data Analysis

All statistical analyses were carried out using the SAS software (Version 6.11, 1996; SAS Institute, Cary, NC). The natural log of all expression values was used to normalize all data for statistical comparison. This was necessary because of the wide inter-individual variation in gene expression within the same population groups (see Table 2 in Reference 4; e.g., c-myc in small-cell carcinoma, E2F-1 in normal, and p21 in populations with small-cell or squamous cell carcinoma).

Identification of a Malignancy Index

The approach we used in an attempt to identify an expression index associated with malignancy was first to evaluate expression of genes known to be mechanistically related to cell-cycle control and then determine which ones were significantly increased or decreased in malignant compared with normal cells. It was hypothesized that phenotype (with respect to cell-cycle control) results from an interaction among several variables in the form of gene expression levels. Expression of each of these genes was related to the malignant phenotype in univariate (Student's t test, analysis of variance) analysis using the Statistical Analysis System software (SAS version 6.12).

Next, we developed quantitative models, referred to as indices, by (1) multiplying the expression value (in units of mRNA/10⁶ -actin mRNA) of one or more cell-cycling genes that had activities positively associated with one or more histologic subtypes of carcinoma (small-cell, squamous, and/or adenocarcinoma), and (2) dividing these products by the multiplicative product of the expression values of genes that were negatively associated with malignancy. Indices derived in this way were considered models that needed to be validated through testing of additional samples.

	Results

Top Abstract Introduction Materials & Methods Results Discussion References

Single-gene Analysis

When cultured cells were evaluated under optimal proliferating conditions (normal BEC incubated in BEC growth medium, carcinoma cell lines incubated in RPMI 1640 containing 10% fetal bovine serum), no cell-cycling gene was significantly altered (P < 0.05) in all three subtypes of carcinoma cell lines evaluated relative to normal proliferating BEC populations (see Table 2 in Reference 4). Increased expression of c-myc is associated with proliferation in many cell types and malignant transformation of some bronchogenic carcinoma cells (6). Expression of c-myc mRNA was increased in the small-cell carcinoma cell lines studied here, however, the mean level of c-myc expression in adenocarcinoma and squamous carcinoma cell lines (870 and 1,023 mRNAs/10⁶ -actin mRNAs, respectively) was lower than that of cultured normal cells (1,137 mRNAs) (Table 1; see also Table 2 in Reference 4). Similarly, because increased E2F-1 expression is associated with increased cell proliferation (5), we hypothesized that it would be expressed at higher levels in tumor compared with normal cells. As expected, the mean level of E2F-1 expression for all three types of carcinoma cell lines (10,843, 3,570, and 5,146 mRNAs for small-cell, adenocarcinoma, and squamous cell lines, respectively) was higher than for all cultured normal cell populations (468 mRNAs) (Table 1; see also Table 2 in Reference 4). However, the E2F-1 level for BEC lot 6F0450 was greater than that for both A549 and H2126 cell lines, and thus no significant difference was observed between the means of normal cultured proliferating BEC and adenocarcinoma cell lines (Table 1; see also Table 2 in Reference 4). Increased p21 expression is associated with inhibition of cell proliferation (5) and might be expected to be expressed at higher levels in normal compared with tumor cells. Whereas this was true with respect to adenocarcinoma and squamous carcinoma cell lines, the mean level of p21 in the small-cell lines was higher than that in normal cells (27,589 and 18,005 mRNAs, respectively) (Table 1; see also Table 2 in Reference 4). Thus, for c-myc, E2F-1, and p21, the level of expression of each gene in a series of normal cell populations overlapped the levels observed in a series of malignant cell populations.

Similarly, in primary normal and malignant tissues no single gene distinguished all malignant from normal samples. Although E2F-1 expression did distinguish the three matched pairs of normal bronchial epithelial tissue and bronchogenic carcinoma, it did not distinguish normal parenchyma from bronchogenic carcinoma (Table 2).

Derivation of a Malignancy Index

Indices were derived by (1) multiplying the expression value (in units of mRNA/10⁶ -actin mRNA) of one or more genes that were expressed at significantly increased levels in at least one of the three subtypes of carcinoma cell lines initially evaluated (c-myc, E2F-1, cyclin E, and CDK4 PCNA), and (2) dividing these products by the product of the values of one or more genes that were expressed at significantly decreased levels in at least one of the subtypes (p21, p53) (see Tables 2 and 3 in Reference 4). Genes that did not significantly vary in expression among the normal and tumor samples studied (cyclin D3, cdc2, mad, max p21, max p22, and rb) were excluded from consideration (see Table 2 in Reference 4). In addition, because cyclin D2 was not expressed at measurable levels in several tumors (see Table 2 in Reference 4), it could not be assessed in the indices. Each of the derived indices was empirically evaluated for predictive value (sensitivity and specificity for tumorigenic phenotype).

The c-myc × E2F-1/p21 Index Distinguishes All Normal from All Malignant Cultured BEC

In contrast to expression of c-myc, E2F-1, and p21 genes individually, an index of all three genes (c-myc × E2F-1/ p21) clearly distinguished cultured normal cells from each of the tumor cell lines (Table 1). In the closest approximation of a cultured normal BEC to a cultured bronchogenic carcinoma cell population, the index value of 77 for BEC lot 6F0450 was nearly 2-fold less than that of 151 for adenocarcinoma cell line A549 (Table 1). Thus, a value of 100 separated all cultured malignant cells from all cultured normal BEC tested thus far. In other words, it detected all carcinoma cells as malignant (no false negatives), did not detect any of the normal cell populations as malignant (no false positives), and thereby had a predictive value of 100% in all samples evaluated thus far. Inclusion of cyclin E, CDK4, PCNA, and/or p53 gene expression in the index did not improve, or decreased, the predictive value. After thorough statistical analysis of the data, no other combination of cell-cycle control genes discriminated between normal and tumor cells as well as c-myc × E2F-1/p21.

Effects of Differentiating and Confluent Conditions on the Index

Although the appropriate comparison between cultured normal and malignant BEC was considered to be cells cultured in their respective optimal growth media under exponentially proliferating conditions, cells cultured in exactly the same conditions were also evaluated. Incubation of BEC populations in the same medium used to culture the carcinoma cell lines (RPMI-1640 containing 10% FBS) for 16 to 18 h prior to RNA extraction as described (4) resulted in a reduction in the malignancy index and therefore greater discrimination between normal and tumor cells (data not shown). Under these conditions the malignancy index for lot 6F0450 was 0.75 and the highest malignancy index for a normal BEC population was 3.08, observed in lot 10525, which is approximately 50-fold lower than the value of 151 observed in A549. Furthermore, growth to confluence decreased the malignancy index in all normal cell populations but one, and the difference between normal and malignant cells increased (data not shown). Under confluent conditions the highest malignancy index for a normal BEC population was 7.87, obtained for lot 17378. This was nearly 25-fold lower than the lowest value obtained for a carcinoma cell line under confluent conditions, which was 191 for adenocarcinoma cell line A427.

Mechanisms of Index Elevation in Malignant Cell Lines

In normal cells when c-myc was expressed at high levels relative to the mean for all normal BEC populations (as in lot 17684), an increase in p21 expression and a decrease in E2F-1 expression took place and preserved the homeostatic equilibrium associated with a nontumorigenic population. The gene or genes responsible for the increase in the index in carcinoma compared with normal cells varied from one cell line to another (Table 1). In addition, c-myc expression was increased in some tumors relative to the mean for normal BEC populations but not in others. For tumors in which c-myc was not increased, E2F-1 was increased (e.g., SW900) and/or p21 decreased (e.g., H520, H2126, A549).

Validation of the c-myc × E2F-1/p21 Malignancy Index in Primary Bronchogenic Carcinoma Tissue

The c-myc × E2F-1/p21 index described previously must be considered a model based on preliminary data, and must be tested to attain validity. This index was first hypothesized to be associated with malignant phenotype after assessing only five normal BEC and five carcinoma cell lines. Since that time, the index has been assessed in an additional three normal BEC cultures and seven carcinoma cell lines (Table 1), three matched primary normal bronchial epithelial and malignant samples, and three matched normal parenchyma and bronchogenic carcinoma tissues (Table 2). Data for all eight normal BEC and 12 carcinoma cell populations are presented in Table 1, and data for all primary tissue samples are presented in Tables 2 and 3. An index value of 100 was 100% sensitive and specific for distinguishing normal from malignant phenotype in both cultured and primary BEC and tissues (Tables 1 and 2). In addition, an index value of 300 distinguished all five primary bronchogenic carcinoma tissues from the three normal parenchymal samples evaluated thus far (Table 3).

As described in MATERIALS AND METHODS, the tumor samples studied were selected for > 50% viable tumor content. It is expected that there will be a loss of index sensitivity for malignant phenotype as the amount of normal stroma increases. The degree of sensitivity relative to fraction of tumor cell content is being evaluated. Of note, a sample that contained less than 5% tumor cells had a normal index value.

	Discussion

Top Abstract Introduction Materials & Methods Results Discussion References

Simultaneous expression measurement of many cell-cycling genes in carcinoma cell lines and their progenitor cells under the same culture conditions has provided data that allow new opportunities for hypothesis-testing. In particular we tested the hypothesis that simultaneous analysis of multiple relevant genes in carcinoma cell lines would reveal expression patterns with better correlation than single gene expression or other current predictors with malignancy, and the data obtained support that hypothesis.

Although individual cell-cycling genes were altered in a significant proportion of tumors, no single gene discriminated all malignant from normal samples. However, by comparing relative expression of genes related to proliferation control, the c-myc × E2F-1/p21 index was identified and does discriminate all 12 carcinoma cell lines tested from all eight normal cultured BEC populations, and all five primary bronchogenic carcinoma tissues from all three normal bronchial epithelial tissues, as well as all three normal lung parenchymal samples. This finding emphasizes the importance of comparing relative expression of many different genes to understand the mechanistic basis for a particular phenotype. Based on the limited studies reported here, the index will be lower in primary normal bronchial epithelial tissues compared with normal lung parenchyma. Whether this difference is significant will require further studies. A difference is likely due to a different pattern of cell-cycle control gene expression in different normal cell types. This difference in index values is not due to the alveolar macrophages (AM) in lung parenchyma, because we have evaluated many samples of bronchioalveolar lavage cells that were greater than 90% AM and the value in those cells is significantly lower than that in normal BEC (data not shown). Thus, the difference is likely due to one of the other cell types in lung parenchyma, such as alveolar epithelial, endothelial, fibroblast, muscle, or inflammatory cells.

The close association of elevated c-myc × E2F-1/p21 index values with malignant phenotype provides additional opportunities for hypothesis testing. For example, it should be feasible to measure this index in FNA samples from primary bronchogenic carcinoma and determine whether there is a significant proportion of tumor cells represented in the sample. Any sample with a c-myc × E2F-1/ p21 index elevated to a level consistent with malignant phenotype will qualify as a suitable sample for correlation with other gene expression patterns, such as those that predict sensitivity to chemo- and/or radiation sensitivity.

The ability to measure quantitatively many genes simultaneously allows for development of indices that are likely to have more mechanistic significance than measurement of individual genes one at a time. Although the index presented here was derived empirically, the fact that each of the genes in the index is mechanistically associated with malignant transformation supports the hypothesis that the association of the index with malignancy is mechanistic. This hypothesis is testable through in vitro antisense targeting of genes expressed at higher levels in tumor cells (e.g., c-myc and E2F-1) and transfection of expression vectors to supplement genes expressed at lower levels in tumor cells (e.g., p21).

We hypothesize that the primary genetic mechanisms underlying alterations in the index value are mutations in one or more cell-cycle regulatory genes (e.g., p53, rb). Such mutations may prevent cells from maintaining a pattern of gene expression mechanistically associated with normal, division-limited homeostatic equilibrium.

The results described here have several implications for understanding BEC biology and pathology, including the following. First, although the index presented has a high predictive value for distinguishing normal from tumor cells, the individual genetic alterations that resulted in alteration of the index were different for each cell line. Thus effective gene therapy will require targeting different genes in each tumor. Second, a gene expression index that has high specificity for malignant tissue, such as the one described here, will provide a necessary quality control when attempting to relate other gene expression patterns to malignant phenotype of small (e.g., FNA) biopsy specimens. Third, it is expected that the index will be improved as new data are obtained from repeated testing of additional genes and samples. For example, we plan to evaluate additional genes involved in control of cell proliferation. It is possible that upon analysis some of these genes will be altered and, if incorporated into the index, will result in even wider separation between normal and malignant lung cells and tissue. In addition, it may be possible to identify tumors that are not clearly distinguished from normal on the basis of the c-myc × E2F-1/p21 index. Further analysis of those tumors may identify particular genes that will improve the index when included. Fourth, it is expected that similar indices will be developed to define better other BEC phenotypes, including mucous, ciliated, and squamous metaplastic differentiation, cystic fibrosis, asthma, and bronchitis.

As additional samples are studied, it is possible that higher index values will be observed in normal samples and lower values will be observed in malignant samples. It is recognized, therefore, that it may be necessary in the future to reevaluate the optimal index value on the basis of incidence of false positives and false negatives, and on the application for which it is being used. For example, although an index value of 100 distinguished all cultured and primary normal from all malignant BEC and tissues, a value of 300 better distinguished all normal lung parenchyma from bronchogenic carcinoma tissues. These indices will have different applications. The index value that distinguishes normal from malignant samples may be used to address mechanistic questions, such as: Will altering the index to a low level in a malignant cell line by introducing a vector that contains either sense for p21 or antisense for c-myc or E2F-1 result in loss of the malignant phenotype? The index that distinguishes normal parenchyma from bronchogenic carcinoma tissue will be used to confirm that an FNA sample contains a significant amount of tumor specimen, and that it is suitable for other gene-expression studies. Such studies may include, for example, an effort to identify a gene expression index that predicts sensitivity to chemotherapy.