Transcriptional Gene Expression Profiling of Small Cell Lung Cancer Cells

INTRODUCTION

SCLC 4 is an aggressive disease, which is generally disseminated at the time of diagnosis. Initially the cancer is responsive to chemotherapy, but almost always recurs in a resistant form resulting in a 5-year survival rate of <5%. SCLC is generally correctly identified by pathological means, wherefore identification of new markers for classification of this tumor type is not pertinent. This is in contrast to the situation for discrimination of subclasses of other lung tumor forms, such as adenocarcinomas, for which therapeutic response and survival rates can differ markedly despite similar pathology. Differences in expression profiles, which can distinguish between these subclasses, have been revealed by microarray analyses (1, 2, 3) .

Because of the aggressive behavior of SCLC and the very poor outcome of present treatments, new therapeutic methods for systemic treatment of SCLC are in high demand. Using global gene expression analysis we have searched for genes that are highly and/or specifically expressed in all or most of the tumor cells with the aim to identify novel potential targets for the development of new therapeutic agents. These could be surface molecules for direct targeting in radio-, toxin, or gene therapy, or molecules to which development of cancer vaccines could be used. Other potential targets are molecules involved in maintenance of the malignant phenotype, such as oncogenes and antiapoptotic molecules, to which inhibitors can be applied or lost activity restored.

A characteristic of all of tumors and their metastases, both between patients and within a tumor, is their heterogenicity, making the development of therapeutic strategies difficult, as some cells invariably can escape the treatment. Sufficient material of SCLC tumors from patients is extremely difficult to obtain both in number of specimens and sufficient amounts for microarray analysis. Therefore, we used an alternative approach and used the expression profiles of 21 SCLC cells lines obtained from five different laboratories and 8 xenografted tumors from these cell lines to compare to the expression profiles of 17 normal adult tissues. By this analysis, we identified several genes highly and specifically expressed by all or most of the SCLC cell lines, xenografts, and 6 ressected tumors with no or little expression in normal tissues, which could be candidates for therapeutic targeting. In addition, the analysis clearly divided the SCLC cell lines into two distinct subclasses with different expression profiles.

MATERIALS AND METHODS

Cell Culture.

The following human SCLC cell lines were used: CPH 54A, CPH 54B (4) , GLC-2, GLC-3, GLC-14, GLC-16, GLC-19, GLC-26, GLC-28 (5, 6, 7) , DMS 53, DMS 79, DMS 92, DMS 114, DMS 153, DMS 273, DMS 406, DMS 456 (8) , NCI H69, NCI N417 (9) , MAR 24H and MAR 86MI (10 , 11) . CPH 54 A and B were propagated in MEM (Eagle), all of the GLC, NCI, MAR cell lines and DMS 79 were propagated in RPMI 1640, and all DMS (except DMS 79) were propagated in Waymouth medium, all supplemented with 10% FCS. All of the serum and media were obtained from Invitrogen (Tåstrup, Denmark).

Xenografts.

Cells (0.5–1.2 × 10⁷) from the cell lines CPH 54A, GLC-3, GLC-14, DMS 273, NCI H69, NCI N417, and MAR 24H were inoculated bilaterally in the flanks of 12–13-week-old BALB/c nude mice. The mice were sacrificed, and the xenografted tumors were harvested when one of the tumors had reached a maximal diameter of 1 cm. The cell line CPH 136A was only propagated in nude mice by inoculation of a 2-mm tumor block. Necrotic tissue was removed, and the tumors were either processed immediately or stored in RNAlater (Ambion, Cambridgeshire, United Kingdom) for RNA extraction.

Isolation of RNA.

Total RNA from normal, human tissues were obtained from either Clontech (Brøndby, Denmark; fetal brain, brain, lung, kidney, heart, trachea, adrenal gland, prostate, salivary gland, and thyroid) or from Ambion (lung, liver, brain, pancreas, spleen, small intestine, skeletal muscle, colon, stomach, and testis). Only one sample was analyzed in duplicate (lung RNA from Clontech and Ambion) and one in triplicate (two different batches brain RNA from Clontech and one from Ambion). The duplicates and triplicates showed similar expression profiles, and only the results using the RNA from Ambion are shown here. RNA from cell lines in exponential growth was harvested (after trypsinization for adherent cells), and total RNA from ∼10⁷ cells was isolated using the RNeasy kit (Qiagen) according to the manufacturer’s instructions. RNA from xenografted tumors were homogenized in TRIzol (Invitrogen) and purified according to the manufacturer’s instruction. The TRIzol isolated RNA was additionally purified using the RNAeasy kit (Qiagen, Albertslund, Denmark).

Affymetrix Oligonucleotide Array.

The preparation of biotin cRNA was prepared essentially as described in the Affymetrix Expression Analysis Technical Manual. Briefly,10 μg of RNA was used as template to generate double-stranded cDNA using a T7-(dT)₂₄ primer (Genset, Paris, France) using SuperScript RnaseH⁻ Reverse Transcriptase and subsequent second-strand synthesis (Invitrogen). The cDNA was transcribed into biotin-labeled cRNA using the BioArray, High Yield RNA transcript labeling kit (Enzo Diagnostics, Farmingdale, NY). Fragmentation, hybridization, and scanning were according to the Affymetrix protocol using the Human U95Av2 array (Affymetrix, Santa Clara, CA) and the antibody amplification protocol. The data were analyzed using Affymetrix Microarray Suite version 5. Data from each chip was scaled to a global scaling of 100. In figures displaying the microarray signals (array signal) as bar diagrams, only genes with expression scored as present (P) are included.

Gene Selection and Hierarchical Clustering.

As the basis for the clustering analysis, genes were selected that had a sum of signals across all of the samples >5000, and a SD threshold of 200 expression units to select the 1620 most variable transcripts. We used the CLUSTER and TREEVIEW programs (12) for clustering and visualization of the dataset. Median centering and normalization of the data were performed before clustering.

RT-PCR.

Semiquantitative RT-PCR was performed using cDNA from all of the normal tissues, cell lines, and xenografts prepared as described above. The PCR was performed using cDNAs from 150 ng of total RNA using oligonucleotide primers for INSM1, CDKN2A, PTTG1, and ASCL1 (Official Gene Symbol annotations) with 25 cycles of PCR amplification.

Western Blots.

Whole cell lysates were prepared from cell lines and xenografts by homogenization in ice-cold 20 mm Tris-Cl (pH 7.5), 2% Triton X-100 containing Protease Inhibitor mixture set II and III (Calbiochem, Albertslund, Denmark). Western blots were performed on 5–15 μg of lysate. For probing with anti-NCAM1 antibodies, the lysates were pretreated for 5 min at 37°C with 40 ng/μl recombinant EndoN-HIS (gift from E. Bock, Protein Laboratory, University of Copenhagen, Denmark) to remove polysialylation residues. The antibodies used were mouse monoclonal anti-NCAM1 clone 123C3 (Santa Cruz, Århus, Denmark), polyclonal anti-mGluR8 (GRM8; Upstate Biotechnology, Frederikssund, Denmark), goat polyclonal anti-integrin αE (N-19; Santa Cruz), mouse monoclonal anti-GluR2 clone 3A11 (PharMingen, Brøndby, Denmark), and goat polyclonal anti-NPTXR (C-17; Santa Cruz).

RESULTS AND DISCUSSION

Cluster Analyses.

The expression profiles of 21 SCLC cell lines, 8 xenografted tumors, and 18 normal tissues (17 adult tissues and 1 fetal brain) were analyzed. The xenografted tumors were included in the analysis to be able to identify genes expressed preferentially because of the conditions of cell culture rather than the more physiological condition of propagation in nude mice.

Of the 12,000 genes represented on the U95Av2 array, a selection of differentially expressed genes was performed on 1,620 genes for cluster analysis. A two-dimensional hierarchical cluster analysis of all of the median levels of expression of SCLC cell lines, xenografts, and normal tissues was performed. In the analysis, it was possible to include the expression data of 6 ressected SCLC tumors from Bhattacharjee et al. (1) , as their analyses were performed on the same Affymetrix Microarray Chips. Comparison of the genes highly expressed in normal lung in Bhattacharjee et al. (1) with the expression levels measured for lung in this analysis showed good agreement, thus justifying a direct comparison of the data. The total cluster analysis is shown in Fig. 1 ⇓ . All of the normal tissues clustered together, and all of the SCLC lines, except CPH 54A and B, clustered together. The cell lines were distributed in 4 subclasses (A, B, C, and D). The cluster D only contains the CPH 54 cell lines and has an expression pattern distinctly different from the remainder of the SCLC lines. The 6 ressected tumors clustered closely together, and closest to the normal tissues and the CPH 54 lines. It has been shown previously that these SCLC tumors cluster distinctly apart from other forms of lung cancers and normal lung (1) .

The cell lines GLC-14, -16, and -19 were established from the same patient during longitudinal follow-up. These cell lines have been compared with the biopsies from which they were derived, and showed a good match among morphological, biochemical, and immunohistological findings (6) . GLC-14, established from a tumor before treatment, was in the same subcluster as GLC-16 and -19, which were established after relapse after chemotherapy and on reoccurrence after radiotherapy, respectively. GLC-26 was established from 1 patient from the primary tumor and GLC-28 from a metastasis (13) , and were found in different subclusters. The line CPH 136A, which has only been passaged in vivo as xenografts, clustered with the cell lines in cluster A.

A fetal brain sample was included in the analysis for comparison, as this sample would be expected to have the expression pattern of a proliferative, neuronal phenotype with many similarities to the endocrine and oncofetal characteristics typical for SCLC. In fact all of the SCLC cell lines and tumors clustered closest to fetal brain, adult brain, and testis, thus confirming the neuroendocrine characteristics of SCLC. The clustering close to testis presumably reflects the proliferative status of testis, and that a large variety of cancer cells express cancer/testis antigens, such as NY-ESO-1 (CTAG1), MAGEs, and GAGEs G antigens (14) . All of the xenografts clustered closest to the corresponding cell line, showing that the conditions of cell culture does not significantly change the expression pattern compared with the more physiological growth condition as xenografts.

Many distinct clusters were observed. Several of these are tissue-specific and contain genes specifically expressed by, e.g., brain, liver, adrenal gland, skeletal muscle, and heart. Several sets of clusters are highly expressed in normal tissues and the SCLC tumors, but not the cell lines or xenografts, as indicated (Normal tissues) in Fig. 1 ⇓ . Some of these gene clusters may reflect the similar expression profiles between SCLC and bronchial epithelial cells as found by Anbazhagan et al. (15) , and some reflect the influence of the tumor microenvironment on gene expression. In addition, there are several gene clusters containing genes of the immune system and extracellular matrix proteins, which probably reflect some infiltration by immune and stromal cells. One tumor (no. 6) has distinctly less expression of these genes but does not differ significantly in the SCLC specifically expressed genes.

There is a large group of genes exclusively expressed by the cell lines and xenografts, but not by the tumors. Many of these gene products are involved in replication, cell cycle regulation, and proliferation. Therefore, these may not be cancer relevant, but rather artifacts of established cell lines and are, therefore, not potential targets or markers. There is a distinct cluster of genes expressed by the brain but not in other tissues of which there is expression in most of the tumors and cell lines. A part of this cluster is shown in detail in Fig. 2 ⇓ (Box 3). In addition, there is a cluster of neuronal or neuroendocrine genes expressed by many of the tumors and cell lines. A section of this cluster is shown in Fig. 2 ⇓ (Box 2). Finally there is a cluster of genes expressed by almost all of the tumors and some cells lines, and only in few normal tissues and, therefore, to some extent SCLC-specific. A part of this cluster is shown in detail in Fig. 2 ⇓ (Box 1). The total cluster analysis clearly demonstrates the neuronal or neuroendocrine phenotype of SCLC. The analysis reveals that there is some heterogeneity in expression of the cell lines. The cell lines have a set of expressed genes, which are not expressed in the tumors, and the tumors have expression of contaminating normal tissues and cells, but by comparing both sets of data, SCLC-specific gene expression can be identified.

SCLC Cell Line Subclasses.

The grouping into four distinct clusters of SCLC cell lines and xenografts (cluster A, B, C, and D in Fig. 1 ⇓ ) reveals that there may be distinct subclasses of these cell lines. Cluster D is discussed below. SCLC cell lines have been divided previously into two classes: the variant type and the classic type. The variant type is associated with higher growth rates and a more aggressive phenotype. Indeed, the growth rates of cell lines in cluster B (average doubling time 27 h) is higher than for clusters A and C (72 and 67 h, respectively), whereas there is no obvious distinction between the morphology of different cell lines in the clusters in vitro. The classic type express DDC and bombesin-like immunoreactivity, such as GRP, whereas the variants do not (9 , 16) . Almost all of the cells in cluster C, which contains the classic type NCI H69, express mRNA for GRP and DDC, whereas cells in cluster B, to which the variant form NCI N417 belongs, do not (Fig. 3) ⇓ . All of the cells in cluster A also express DDC and GRP. The expression in the tumors indicate that these may include both types. Another suggested marker for distinction between classic and variant types is expression of the SGNE1, which has been found expressed in all of the classic, but few variant cell lines. SGNE1 is expressed in most cell lines of clusters A and C, but few in cluster B (data not shown). Therefore, it is probable that clusters A and C contain the classical type and cluster B the variant type of SCLC, and this analysis identifies many genes differentially expressed in these subclasses of tumors.

In the selected clusters shown in Fig. 2 ⇓ it is clear that cluster B has a distinctly lower expression of many tumor, brain, neuronal, and neuroendocrine-specific genes. Some of the typical neuroendocrine markers such as chromogranin C (SCG2), ENO2, and NCAM1 are reduced in cluster B, whereas others such as synaptophysin (SST) and PGP 9.5 (USHL1) are not. A large group of clusters of genes relatively specific for the cell lines containing many genes involved in replication, cell cycle regulation, and proliferation do not differ between the cell line clusters. An exception is a small cluster, which appears more prominent in the cluster B and contains many ribosomal proteins and the molecular chaperone HSP90, and may, therefore, reflect the higher proliferation rate of cell lines in cluster B.

It should be mentioned that the grouping in classical and variant types of SCLCs has been performed mainly on cell lines, and it is not quite clear how these relate to the mixed small cell-large cell carcinoma or variant type of SCLC-combined described in the new WHO classification of lung tumors (17) . As this present analysis now reveals many new, intracellular markers to which antibodies are available, it would be possible to make a retrospective analysis on historical paraffin sections to determine whether the grouping has clinical significance.

Characterization of the CPH 54 Cell Lines.

Two cell lines, CPH 54A and B clustered distinctly apart from the other SCLC cell lines and closest to normal tissues (Figs. 1 ⇓ and 2 ⇓ , cluster D). These two cell lines are subclones from the same original tumor 54A (4) and would, therefore, be expected to cluster together, although they differ in DNA index and sensitivity to radiation therapy (13) . Although the cell lines express several neuronal genes, the expression pattern was distinct from the other SCLC lines and xenografts. A cluster of genes highly expressed by the CPH 54 lines and a xenograft thereof, but not any other SCLC line or tumor is indicated in Figs. 1 ⇓ and 2 ⇓ . It is noticeable that in the cluster of genes highly expressed by many SCLC cells and tumors, the expression in the CPH 54 lines is low or absent (Fig. 2) ⇓ . In the cluster (Fig. 1) ⇓ specific for the CPH 54 cell lines there is high expression of GAGE 1, 2, 3,4, 5, 6, and 7, which are known tumor-associated testis-specific antigens (18) . Only one other cell line (DMS 153) expresses GAGEs at high levels, thus demonstrating that the CPH 54 lines differ markedly from the majority of the other SCLC lines. In the CPH 54 cluster there is also high expression of extracellular matrix proteins, such as a large variety of collagens, fibronectin, and laminin, which are not expressed to the same level by the rest of the SCLC cell lines. This indicates that the cell lines may be of fibroblastoid origin rather than SCLC, although the xenografted tumors from the CPH 54A cell line preserved the pathologically determined features of SCLC (4) . In addition, these are the only cell lines that do not have mutated p53 5 or loss or mutated pRB, two of the characteristics of SCLC (19, 20, 21, 22) .

SCLC, Brain, and Neuroendocrine Clusters.

There are three clusters of genes either predominantly expressed by the SCLC cell lines and tumors or also expressed in the brain. A section of these are displayed in Fig. 2 ⇓ with the gene annotations included. In Fig. 1 ⇓ is indicated the localization of the three clusters. In Table 1 ⇓ ⇓ the genes are listed according to functions. Cluster 1 contains genes, which are expressed primarily by SCLC tumors and cell lines. This cluster contains many genes known to be highly expressed by a variety of tumor cells. Many of these are genes involved in cell proliferation, signal transduction, cell division, or high cell motility. The cluster also contains genes reflecting high metabolic activity. Noteworthy is the presence of ASCL1 in this cluster, because this is regarded as a specific neuroendocrine tumor marker negatively regulated in other tissues (23) . The function of ASCL1 is not clarified, but its expression is highly correlated to the snail family of transcription factors involved in cell migration and, therefore, may contribute to the invasive phenotype (24) . High expression of ASCL1 was found in the 6 ressected tumors in Bhattacharjee et al. (1) and 5 ressected tumors in Garber et al. (2) . The expression levels of ASCL1 in the cell lines and normal tissues were verified by semiquantitative RT-PCR (Fig. 4D) ⇓ confirming the expression levels on the array and showing low expression only in fetal and adult brain in normal tissues. There is only very low expression in one line of cluster B (variant type) and very high expression in all but one cell line in clusters A and C (classic type). Therefore, ASCL1 is probably one of the best novel markers for distinction between the variant and classic type of SCLC. ASCL1 is not expressed by the CPH 54 cell lines.

Cluster 2 contains many neuroendocrine genes, including typical neuroendocrine markers, neuronal developmental markers, and neurotransmitters. This cluster also contains other genes often associated with tumor cells coding for proteins involved in cell proliferation, signal transduction, cell adhesion, and motility. Cluster 2 contains NCAM1, a molecule used as a marker for endocrine tumors and cell lines (25 , 26) , and is primarily expressed in brain in the adult. The analysis was verified by Western blotting revealing that all but one of the SCLC cell lines express the two predominantly fetal isoforms (M_r 140,000 and M_r 180,000 isoforms) of NCAM1 (Fig. 5) ⇓ , thus confirming earlier studies performed on several of the cell lines (27) . Cell lines in clusters A and C (the classical type) have high expression, whereas cell lines in cluster B (variant type) have low expression. Therefore, the level of expression of NCAM1 may also be a new marker for determination of which subclass a cell line or tumor belongs to. However, as most SCLCs express NCAM1 (although at different levels), it has been a frequent candidate for targeted therapy against SCLC. Indeed, 5 of the 6 of the tumors from Bhattacharjee et al. (1) also show high expression of NCAM1. Attempts have been made to use NCAM1 for radioimmunotherapy (28, 29, 30, 31) or for immunotoxin therapy (32, 33, 34, 35) , but none of these have reached clinical use. A number of small peptides, which bind NCAM1, have been identified recently (36) , and these could alternatively be tried for use in NCAM1-targeted therapy against SCLC.

Cluster 3 demonstrates the neuronal phenotype of SCLC, as this cluster contains molecules almost exclusively expressed by SCLC, and fetal and/or adult brain. Many of the molecules are involved in neurite outgrowth or neuronal development. A neuronal molecule not associated previously with SCLC is the GRIA2, an excitatory neurotransmitter receptor, is included in Cluster 3. The array analysis identified expression of GRIA2 by >50%, and RT-PCR verified 86% (data not shown) of the SCLC lines or xenografts and among normal tissues only expressed in brain. Validation by Western blotting showed the presence of protein in ≥35% of the cell lines or xenografts (Fig. 5) ⇓ . Another glutamate receptor, the GRM8, was also identified as expressed by 30% of the SCLC cell lines by the array analysis, whereas Western blotting revealed expression in all of the cell lines (Fig. 5) ⇓ . The presence of glutamate receptors on SCLC provides potential new targets for radiotherapy, because there are many specific, small molecular weight ligands (agonist or antagonists) for these receptors (37 , 38) and receptor-specific antibodies. As many of these ligands do not penetrate the blood brain barrier, therapy could be limited to areas outside the brain, thus avoiding adverse effects on glutamate receptors expressed in the brain.

Other neuroendocrine markers not included in the clusters displayed in Fig. 2 ⇓ were also found highly expressed in all or most SCLC lines and tumors. Of these can be mentioned ubiquitin carboxyl-terminal esterase L1, which is a neurospecific peptide that protects against targeted protein degradation; secretonin I (chromogranin B), a neuroendocrine secretory protein, and creatine kinase BB, a supposedly brain-specific enzyme, although the array analysis showed high expression also in many normal tissues. A newly identified marker highly expressed exclusively in neuroendocrine cells and tumors is INSM1, a transcription factor involved in pancreatic and neuronal development (39) . INSM1 was highly expressed by most cell lines, and by the 6 SCLC tumors from Bhattacharjee et al. (1) and the 5 tumors from Garber et al. (2) . The array analysis for expression of INSM1 in the cell lines and normal tissues was verified by RT-PCR (Fig. 4C) ⇓ . INSM1 was not expressed by the CPH 54 lines.

Other Potential Target Molecules Identified by the Microarray Analysis.

Specific expression of other surface molecules not associated previously with SCLC were identified by the analysis, and some of these are potential candidates as surface targets for therapy. The NPTXR was found expressed by the array analysis in more than half of the SCLC cell lines, and only in brain and prostate in normal tissues. The expression of the protein by the SCLC cell lines was confirmed by Western blotting (Fig. 5) ⇓ , which demonstrated expression in >90% of the cell lines and xenografts. This receptor is normally only expressed in synapses, where it mediates uptake of extracellular material via one of its two ligands, neuronal pentraxin 1 and 2 (40) . Another receptor, the apolipoprotein E receptor 2 (LRP8), was also expressed in many cell lines with no detectable expression in normal tissues except for brain and testis (data not shown). This receptor internalizes lipoproteins associated with ApoE in addition to a specific ligand, reelin (41 , 42) . These two receptors are internalizing and, therefore, good candidates for a number of therapeutic approaches involving natural ligands, synthetic ligands, or antibodies. Another candidate surface molecule which has not been identified previously as expressed by SCLC is the ITGAE. ITGAE interacts with the β7 subunit forming an E-cadherin receptor and is normally only expressed on intraepithelial lymphocytes (43 , 44) . The expression of the protein by the cell lines was verified by Western blotting, which demonstrated expression by at least 76% of the cell lines (Fig. 5) ⇓ . Other receptors such as the nicotinic acetylcholine receptor α 5 and GRP49, a glycoprotein hormone receptor, were also found highly expressed in most SCLC lines and xenografts.

Cancer-related Gene Expression.

A newly identified oncogene, PTTG1 (securin), expressed by various cancers was highly expressed in all of the SCLC lines, tumors, and testis, and weakly expressed in spleen, thyroid, and trachea. The analysis was verified on the cell lines and normal tissues by RT-PCR (Fig. 4A) ⇓ , which revealed low expression in more normal tissues. The gene product has transforming activity in vitro and tumorigenic activity in vivo (45 , 46) . Several different functions have been attributed to this molecule relating to its oncogenic nature. One function is as an inhibitor of chromatid separation causing chromosomal instability (47) and a different function as a transcription factor. The latter is believed to induce expression of c-myc, which subsequently induces proliferation and expression of basic fibroblast growth factor (FGF2), which, in turn, stimulates angiogenesis (45) . Indeed, increased serum levels of basic fibroblast growth factor have been found in patients with SCLC (48) . The effect of the expression of PTTG1 on cell proliferation, vascularization, and chromatin stability in SCLC should therefore be performed to assess whether drugs targeted to this molecule may offer new approaches for therapy.

Protooncogenes of the myc family are highly expressed by many cancer types, in particular lung cancers. Elevated expression of at least one of the three mycs (c-myc, n-myc, or l-myc) has been found in many SCLC cell lines (13 , 49 , 50) . The high levels of expression and the form of myc expressed as determined by the array analysis correlated completely with the data published previously for the cell lines. The high expression has in many cases been correlated to amplification of the genes, but is generally more prevalent after establishment as xenografts (51) . This is in concordance with the observation that the ressected tumors showed expression of c-myc but not at the same high level as the cell lines and xenografts.

A family of cancer-related molecules are the MAGEs, which are expressed exclusively in testis in normal tissue, and often highly expressed in cancer cell lines and tumors (14 , 52) . They are targets for many trials using immunotherapy, as peptides from these molecules are displayed by cytotoxic T lymphocytes (53) . Only few studies of MAGE expression in lung cancer have been performed (54) . As the array contains probe sets for all 11 of the known MAGEs, a more complete profiling of MAGE expression in SCLC could be determined. The analysis revealed very high expression of many members of the MAGE family in most SCLC cell lines with high expression of MAGE-2, -3, and -6, and intermediate expression of MAGE- 1, -5a, and -12. Expression of MAGE-3 is shown in Fig. 6 ⇓ . However, in all of the cases the tumors had a markedly lower expression, supporting the theory that cell culture may lead to activation of MAGE genes (53) . Therefore, MAGE-based immunotherapy may not appear to be the first choice for alternative treatment of SCLC.

A group of cancer invasion-associated molecules are the MMPs and their inhibitors (TIMPs), as these can supply the extracellular proteolysis necessary for vascularization, tumor invasion, and tumor dissemination. Expression of several MMPs and TIMPs has been found in several cancer types including SCLC (55 , 56) . However, it is well established that components of extracellular protease systems can be expressed by either stromal and tumor cells or both, and involve cellular interactions in the microenvironment of the tumor (57) . Indeed, the array analysis showed that neither the SCLC cell lines nor xenografts have elevated MMP or TIMP expression compared with normal tissues, whereas the tumors had elevated expression of many MMPs and TIMPs (in particular MMP-9, MMP-11, MMP-12, MMP-14, and TIMP-1), thus demonstrating one of the limitations of using cell lines and xenografts for this type of analysis.

The array analysis shows that of the multidrug resistance and resistance-associated protein genes, only ABC C5 (MRP5) is highly expressed by all of the SCLC cell lines and tumors. High expression of an unspecified MRP type in SCLC has been reported previously (58) . The knowledge of the most prominently expressed multidrug resistance gene could affect the choice of anticancer drugs for the disease.

Proliferation and Replication-related Molecules.

The analysis revealed high and relatively specific expression in most or all of the cell lines and xenografts of a number of molecules involved in cell proliferation, cell cycle control, or chromosome separation; all molecules that are potential targets for therapy. A number of these are listed in Table 2 ⇓ . All of the shown genes are also expressed in at least 50% of the tumors. Some of the genes have not been associated previously with SCLC but have found to be up-regulated in other tumors. One puzzling novel observation is the high expression in SCLC of mitotic arrest deficient, a component of the mitotic spindle assembly checkpoint, which is generally regarded as a tumor suppressor. Reduced expression has been reported in some cancers (59) , although elevated expression has been reported for gastric cancer. The microarray analysis did not indicate reduced expression of its downstream component, cdc20.

However, this is the case for another tumor suppressor, CDKN2A (p16INK4), which is a negative regulator of cell proliferation by stabilizing the tumor suppressor pRB. Deletions, reduced expression, or mutations of p16INK4 are observed frequently in a variety of tumors. However, the microarray analyses showed high expression by most SCLC lines and tumors, which is consistent with previous observations of high expression in SCLC tumors (60 , 61) and cell lines (62) . p16INK4 leads to cell cycle arrest in the presence of functional pRB. pRB has been found to be absent or mutated in the majority of SCLC tumors and cell lines (20 , 22 , 63 , 64) . In fact, absence of p16INK4 has been found to be restricted to lung cancer cell lines that retain wild-type pRB (65) . There are two major forms of CDKN2A, p16INK4 and p14ARF, derived by alternative splicing. p14ARF is likewise regarded a tumor suppressor by stabilizing p53, but also has p53-independent cell cycle regulatory functions (66) . The probe sequences on the microarray will recognize both forms. RT-PCR using a p14ARF specific primer set showed expression levels similar to the array analysis (Fig. 4B) ⇓ demonstrating the presence of this mRNA. p14ARF transcription is stimulated by E2F transcriptions factors (67) , and the analysis shows high expression of E2F1 and E2F3 in all of the SCLC cell lines, as has been found previously for most SCLC tumors by immunohistochemistry (68) . However, an immunohistochemical analysis has shown that despite high levels of mRNA expression, the p14ARF protein was not present in the majority of SCLC tumors, indicating a regulation at the translational level, whereas p16INK4A protein was present in most tumors (60) . This observation demonstrates the importance of validating results obtained by array analysis on a protein level.

In the search for new therapeutic targets for therapy and prognosis of SCLC, an expression screening of many SCLC cell lines and xenografts was performed using oligonucleotide microarrays. The analysis revealed several novel potential targets and confirmed the expression of the majority of known targets. These genes include surface receptors, oncogenes, antiapoptotic genes, and other cancer-related genes, which all may be targets for a wide variety of therapeutic approaches. Comparison with the published expression profiles of 6 ressected SCLC tumors (1) demonstrated that the expression profiles identified using the cell lines and xenografts in most cases are similar to the profiles of the tumors, although the expression profiles of the tumors are to some extent blurred by the presence of stromal tissues, vascular tissues, and immune cell infiltration. On the other hand, the cell lines had expression of a number of genes, which were not expressed by the tumors. The analysis clearly demonstrates that for expression analysis it is not necessary to use xenografted tumors, as cell lines in culture show the same expression profiles for the majority of genes, both for cancer markers and genes involved in proliferation and cell division. One cell line, which has only been propagated as a xenografted culture, did not differ in expression profile from the cells propagated in culture. Therefore, this expression analysis presumably can justify the use of SCLC cells in culture for additional characterization of the tumor markers and for testing the initial effect of potential therapeutic approaches. However, it is necessary to use many cell lines with different characteristics for the analysis to obtain reliable results.

The microarray analysis was found to be a very effective means of distinguishing between subclasses of SCLC cells, a method that has also been used for defining subclasses of lung adenocarcinomas (1, 2, 3) .

Genes identified as highly and specifically expressed in SCLC, but which are not potential direct targets, could be used for other purposes such as tumor-specific expression for gene therapy. Several tumor-specific promoters have been tested previously, such as the carcinoembryonic antigen promoter (69) , the GRP promoter (70) , and the ENO2 promoter for SCLC (71) . Although these promoters confer cancer specificity, the activity of these promoters has been found insufficient for gene therapy. ASCL1, which appears to be down-regulated in all of the non-neuroendocrine cells because of the presence of specific repressors (23) , and INSM1, which is regulated by transcriptional activators in neuroendocrine cells (72) , are two potential candidates conferring high expressional activity.

In addition to genes with known or proposed functions the analysis also revealed high and specific expression of many genes with no known function or homologies. Many of these genes are KIAA genes isolated from a human immature myeloid cell line and often found highly expressed in a variety of cancers (73 , 74) . Therefore, when functions are assigned for these genes, many more potential targets may appear.