Tumorigenesis implies alterations of diverse genetic factors that impact numerous biochemical pathways. Although cancer cell genomes display a high degree of loss of heterozygosity (LOH), only a limited number of known tumor suppressor genes (TSGs) have been found systematically mutated in most sporadic cancer types. In lung cancer, high rates of mutations (>30%) affect only TP53 and RB, whereas for ovarian cancer, TP53 and KRAS are the only genes recurrently mutated.1, 2 Recent findings on epigenetic inactivation through promoter hypermethylation may explain the relatively low rates of mutations in various TSGs.3 However, this phenomenon does not apply to all known TSGs and is often not exclusive of other genetic hits at a given gene locus.4 Thus, the search for genetic alterations associated with the carcinogenic processes in sporadic tumors is required and may provide additional therapeutic targets. Indeed, several recent TSG identifications commenced, at least partially, from homozygous deletion (HD) detection in primary tumors and/or cell lines. Such was the case for WT1, INK4A, BRCA2, PTEN, SMAD4 and SNF5.5, 6, 7, 8
In an attempt to identify tumorigenic pathways, we examined 89 tumor cell lines for HDs throughout the human genome. Although cell lines differ from both normal and cancerous tissues, a recent study has shown that they represent a faithful image of the organs from which they were derived.9 Assuming that the same should be true for tumorigenic processes, we investigated 230 loci either at well-established or putative TSGs. Furthermore, to assess genomic instability, the survey included several fragile sites, telomeric single-tagged sequence (STS) and housekeeping genes. Seventeen different loci were affected by at least one HD. Interestingly, our results demonstrate the occurence of HDs at novel loci, and we provide evidence for previously unknown associations of certain TSGs with particular types of cancers. Finally, we noticed that some subsets of cell lines display HDs at multiple sites, suggesting that, in the corresponding tumors, the carcinogenic process implies several alternative and/or parallel pathways.
MATERIAL AND METHODS
Tumor cell lines and DNA extraction
In total, 89 tumor cell lines were used for HD analysis (Table I). The panel includes 69 cells derived from epithelial tumors and 20 from sarcomas, hematopoietic malignancies, brain tumors and neuroblastomas. The precise source of each of these cell lines is available upon request. After growth in appropriate culture media, cells were harvested and nucleic acids extracted. Genomic DNA was extracted from cells after a treatment with lysis buffer (2% SDS, 10 mM Tris, HCl pH 8,10 mM NaCl, 10 mM EDTA, 10 mg/ml proteinase K) and a single phenol-chloroform. DNA was ethanol precipitated and resuspended in TE. RNA was extracted using the Tri-Reagent® solution (Euromedex, Souffelweyersheim, France).
Table I. Names and Tissular Origins of the 89 Cell Lines Profiled
Tissue of origin
The 37 cell lines of the first panel analyzed for the 246 loci are marked by an asterisk (*).
The analysis of tumor cell lines' genomic DNAs for HD at candidate loci was performed by PCR using the STS listed in Table II. Most oligonucleotide sequences were determined from the published human genome project working draft on the GoldenPath server (http://genome.ucsc.edu/goldenPath/hgTracks.html). For markers aiming at specific genes, when no HD was previously published, coding exons were preferentially chosen. When HDs in a given region were already reported, the marker (exonic sequence or anonymous STS) was chosen in the consensus region of HD overlap. For HOXA5 (7p12), primer sequences detecting indicated STS were found in UniGene (http://www.ncbi.nlm.nih.gov/UniGene). Primers for β-globin (11p15) and β-actin (7p22), commonly used to test genomic DNA quality, have been published.10, 11 When exon size did not exceed 600 bp, both primers were chosen in intronic sequences, whereas for larger exons (>600 bp), one primer was chosen in exonic sequences and the other in a flanking intron. Telomeric markers have been described by Rosenberg et al.12 and Knight et al.13 All primer sequences are available upon request. Oligonucleotides were obtained from Eurogentec (Seraing, Belgium), Genset (Paris, France) and Life Technologies (Cergy-Pontoise, France). Reactions were performed as previously described.14 A step-down amplification was performed in most cases with a minimum annealing temperature ranging from 48–58°C.15 All primer sets mapping at the INK4A/ARF locus were used following the initially reported method of Kamb et al.6 Amplification products were separated on a 2% agarose gel and visualized on a UV transilluminator.
Table II. Candidate Genes and Loci Scanned for Homozygous Deletions
For each gene, the number of the exon analyzed is indicated. The asterisk (*) indicates that homozygous deletions have already been described for this locus in the literature.
Chr. 1 (n = 16)
Chr. 2 (n = 10)
Chr. 3 (n = 10)
Chr. 4 (n = 36)
Chr. 5 (n = 7)
Chr. 6 (n = 8)
Chr. 7 (n =7)
Chr. 8 (n = 16)
Chr. 9 (n = 13)
Chr. 10 (n = 7)
Chr. 11 (n = 10)
Chr. 12 (n = 6)
Chr. 13 (n = 13)
Chr. 14 (n = 3)
Chr. 15 (n = 5)
Chr. 16 (n = 20)
Chr. 17 (n = 17)
Chr. 18 (n = 5)
Chr. 19 (n = 11)
Chr. 20 (n = 5)
Chr. 21 (n = 4)
Chr. 22 (n = 14)
Chr. X (n = 3)
A first panel of 37 cell lines was tested for all 246 loci (Table II). Each negative PCR was performed twice. Bona fide homozygous deletions were confirmed using external, nonoverlapping primer pairs. HD confirmation by duplex PCR was not used as it does not rule out negative PCR attributable to mismatches on primer annealing sequences. The extent of the different HDs were then established step-by-step by amplification of flanking exons or flanking STS.
Reverse transcriptase and PCR
Samples with HD were studied at the RNA level. Five micrograms of total RNA were reverse transcribed at 37°C using 200 units of Superscript reverse transcriptase (Life Technologies) and 300 ng of pd(N)6 random hexameres (Pharmacia, Orsay, France) in 12 μl final. One microliter of the produced cDNA was amplified according to a nested PCR procedure. Amplified fragments were purified from agarose gels (Qiaex II kit, Qiagen, Courtaboeuf, France) and ligated in the TOPO-TA cloning vector (InVitrogen, Cergy-Pontoise, France). Sequencing was performed using Thermosequenase (USB, Amersham, Orsay, France) on an ABI automated fluorescent sequencer (Applied Biosystems, Courtaboeuf, France) at the ESGS sequencing facilities (Evry, France).
Homozygous deletion screen
In a first step, a panel of 37 tumor cell lines was tested for HDs at 246 distinct loci (Tables I and II). Seventeen of these loci showed HDs and were further examined in a second panel of 52 cell lines. We decided to perform HD profiling in various tumor types for two reasons: (i) many TSGs are ubiquitously expressed and active in most human tissues and (ii) some TSGs frequently altered in certain tumors were identified through abnormalities detected in cancer types marginally involved (e.g., BRCA2 was localized in part by an HD in a pancreatic cancer).8 A large proportion of the cell lines studied were established from tumors of epithelial origin (n = 69/89, 77%) with, among them, important subsets of lung (n = 22), ovarian (n = 19) and head and neck squamous cell carcinoma (HNC; n = 13). Other carcinomas included tumors of the breast, prostate, colon and cervix. The remaining cell lines were mostly derived from mesenchymal tumors (blood, muscle, kidney, bone) (Table I). All chromosomes were investigated at 3 to 36 loci. The vast majority (230/246, 92%) of the studied loci were either previously shown to be homozygously deleted or considered as harboring candidate TSGs. The remaining 16 loci can be considered as markers of genomic stability (7 subtelomeric STS, 6 housekeeping gene exons and 3 STS at fragile sites) (Table II). When no amplification product was detected, we performed a confirmatory PCR using an external primer pair to rule out the possibility that the lack of a detectable PCR product was the result of a sequence polymorphism at one of the primer's sequence. In total, 53 HDs (53/9,986, 5.3% PCR) were found at 17 independent loci in 40 cell lines of 89 (44%) (Table III).
Table III. Homozygous Deletions (HDs) in Tumor Cell Lines1
LC, lung carcinoma; OC, ovarian carcinoma; SCC, head and neck squamous cell carcinoma; BC, breast carcinoma; BT/NB, brain tumor/neuroblastoma; LL, leukemia-lymphoma; PC, prostate carcinoma; CC, cervix carcinoma; Misc, miscellanous tumors.
The ratio of HDs for each cell type is indicated in brackets. Homozygously deleted exons are indicated.
2Exons 1 of INK4A and ARF are indicated as exon 1a (alpha) and exon 1 b (beta), respectively.–−/−, HDs at STS on 4q and 17p.
General instability in tumor cell lines
Six housekeeping genes (albumin on 4q, β-actin on 7p, β-globin on 11p, α-tubulin on 2q, GAPDH on 12p, HPRT on Xq) were tested to assess the importance of HDs as potentially nonspecific events in tumor cell lines. No HD was detected at these loci. Unbalanced translocations resulting from terminal deletions and subsequent chromosome fusions represent recurrent alterations in epithelial tumors.16 Therefore, we suspected that in some cases both telomeres may be lost at major targets of LOH. We tested subtelomeric markers (<300 kb from telomeres) at 7 chromosomes (1p, 4q, 6q, 8p, 13q, 16q, 17p) that were previously found deleted in numerous carcinomas.17, 18 Two of these markers, 4QTEL11 and 17PTEL80, were found to be homozygously deleted in 1 (SCC25) and 3 (PEO16, ES2, K562) cell lines, respectively, indicating that homozygous loss of chromosome ends can occur occasionally, albeit at a low frequency (4 HDs of 363 telomeres tested; Table III). Abnormalities at 17PTEL80 were not associated with homozygous deletions of p53. To our knowledge, this is the first time that such alterations are detected by a molecular method in human cells. HD in 4qtel was then studied in detail since 4q35 is a prominent target for loss of heterozygosity in liver cancer, one of our main focuses of interest. According to the last versions of human genome sequence, in SCC25, a 4.7 Mb segment was biallelically removed from the 4q35 region. This relatively gene-poor region contains only a few candidate TSGs, among them the gene encoding the melatonin receptor 1A (MNTR1A) known to exert antiproliferative effects on various tumor cells.19 The 89 fragile sites (FS) mapped on human chromosomes represent additional candidate regions for HD search.20 Three of them were tested in our survey. D7S522 mapping at FRA7G (7q31) did not exhibit any instability in cell lines. In contrast, FRA3B (D3S1300 on 3p14) was homozygously deleted along with exon 5 of the FHIT gene in 5 cases (6%) (Table III). The WWOX gene, a putative tumor suppressor mapping at close range of another common fragile site, FRA16D (16q23), was deleted in 2 samples (2.5%) (Table III).
Variations of HD incidence
A high proportion (9/13, 69%) of HNC cell lines showed at least 1 HD (Table III), whereas the proportion was lower (approx. 40%) for lung or ovarian carcinoma cell lines as well as for malignant blood cells. In ovarian and lung cell lines, however, a multiplicity of targets were involved (7 and 6, respectively), indicating an apparent complexity of carcinogenesis in these tumor types (Table III). Remarkably, 11 cell lines (11/89, 12%) displayed more than 1 HD (as many as 3 in SCC25 and PEO4), whereas the other samples appear more stable. This suggests the existence of variable sensibilities to limited biallelic loss among the different tumors.
Four STS (p14 exon 1β, p16 exonα and p14-p16 exons 2 and 3) mapping to the p16-INK4A/p14-ARF locus were systematically studied in the 89 samples to cover the region in detail. In 23/89 cell lines (approx. 26%), an HD was detected (Table III and Fig. 1).6 A significantly higher prevalence of HDs was detected in leukemia-lymphoma and in HNC cells (41–61%) when compared to ovarian and lung cancer cell lines (18–21%, χ,2p < 0,05). P16-INK4A and p14-ARF genes were co-inactivated in most cases (22/23, 95%) (Fig. 1). However, in the astrocytoma cell line, U373, a lesion was found only in p14-ARF.
The second most frequently deleted target was FHIT in 3p14 (5/89 cases, 5.5% of the cell lines). The double inactivation of this gene, which encodes a diadenosine triphosphate hydrolase involved in stress response signaling, is common in carcinomas.21 However, we found it to be restricted to only 3 specific cell types (HCN, colorectal and cervix cells) (Table III). All HDs were limited to a single exon (exon 5) of FHIT mapping at a close range from FRA3B. Such an alteration results in the removal of the initiation codon and the first 32 amino acids (Table III). Multiple HDs were also observed at the TP53, LRPDIT, PTEN, WWOX and NF2 genes (4, 3 and 2 for the 3 latter, respectively) in cell lines established from several tumor types (Table III). Mutations in the WWOX gene have been reported previously.22 In the case of the TP53 tumor suppressor, initiation codon (exon 2) and most or all of the coding region were homozygously deleted in 4 cell lines of lung, bone and blood origin. The recently identified low-density lipoprotein-related protein 1B (LRPDIT) gene was homozygously deleted in 3 cell lines (Table III).23 The 3 samples (SKMES1, OVCAR3 and U2OS) established from different tissue types were altered through various internal deletions involving exons 4–7 that did not affect the reading frame (Fig. 2). Our findings represent the first reports of HD affecting LRPDIT in ovarian cancer as well as osteosarcoma. The PTEN gene is a potent growth suppressor negatively regulating cell-matrix interactions as well as phosphoinositide-3-phosphate kinase signalling pathways. Initially found mutated in glioblastomas as well as prostate and breast cancers, PTEN mutations were detected later in leukemias and endometrial cancers.24 HDs were detected in the PC3 prostate cancer cells and in the CEM lymphoblastic leukemia. For the already known double deletion of PC3, no PTEN transcript was detectable due to a deletion of all or most of the coding region, whereas in CEM, removal of exons 2–5 resulted in abnormal transcripts with introduction of a premature stop codon (Fig. 2).25 The NF2 gene, which encodes the cytoskeleton-associated merlin protein,26 was homozygously lost in a lung (CALU1) and an ovarian (PEO4) cell line. Exon 1, containing the initiation codon, is removed in PEO4, whereas in CALU1 cells internal deletion of exons 2–3 preserves the frame of the transcript (Fig. 2). Interestingly, a homologous murine mutant form lacking exons 2–3 was shown to be transforming rather than tumor suppressive in transgenic mice.27 HDs reported in the current survey represent the first cases of mutation affecting NF2 in lung or ovarian cancers.28, 29
Single HDs were found at 8 additional loci, RB, SMAD4, MSH2, MAP2K3, MAP2K4, TMPRSS2, STK11 and TP73, genes in ovarian, colorectal, lung, cervix or bone tumor cells (Table III). HDs at the 3 former loci confirmed previous work.30, 31, 32 A double deletion of STK11 was found in HELA cells (Table III). The STK11 gene encodes a serine-threonine kinase interacting with p53 and involved in mediating the p53-dependent apoptotic program.33 It is found mutated in patients suffering from Peutz-Jeghers syndrome as well as in selected sporadic tumors.34 In HELA cells, HD removes the initiation codon together with the 155 N-terminal amino acids, thus likely precluding the production of the wild-type protein. It is well known that cross-contamination of cell lines by HELA is a widespread phenomenon.35, 36 Consequently, we tested 2 HELA cross-contaminants, CHANG-CCL13 and HEP2 for HD at STK11. Both cells lines were homozygously deleted for exons 1–4, indicating that this alteration is not restricted to bona fide HELA cells but rather represents an ancient event preceding cross-contaminations (data not shown). No STK11 transcript was detected in HELA by RT-PCR (Fig. 2). This is the first example of a mutation affecting STK11 in a cervical carcinoma. TMPRSS2, encoding a plasma membrane serine protease, was formerly found deleted in a pancreatic cell line, whereas we detected an HD in the SCC25 HNC cell line in which all exons are removed (Table III).37 The exact functions of TMPRSS2 as well as the other members of the hepsin family of serine proteases remains unknown in the neoplastic process.38, 39 Two HDs affecting the signal transduction machinery were found in OAW28 cells established from an ovarian carcinoma. Both for MAP2K3 and for MAP2K4 all exons were deleted. MAP2K4 is suspected to be a tumor suppressor for pancreatic and breast cancer, where it is found mutated in a small proportion of cases. The role of MAP2K3 in cancer is still largely unknown.40 Our current study is the first to report HD affecting MAPKs in ovarian carcinoma. Finally, an HD affecting TP73 in a small cell lung carcinoma cell line (SC10) was detected. A single mutation among 114 cases of lung cancer screened has been reported previously.41 Significantly, HD affected the 5′ region of the gene encoding the transactivation domain responsible for cell cycle arrest and for the apoptotic properties of p73.42
We found no abnormalities in 40 of 57 loci that had tested positive previously in at least 1 tumor or cell line specimen. The most significant of these negative loci are BRCA2, CASP8, CDH6, MEN1, NF1, BRG1, SNF5, TGFBR1 and WT1. No recurring associations of HDs were noticed among the 11 cell lines displaying at least 2 HDs (Table III).
The search for homozygous deletions in tumors and cell lines is considered a powerful tool to identify TSGs. This prompted us to perform a comprehensive scan for HD in a large panel of cell lines with the aim of revealing genes targeted for mutation in the course of carcinogenesis. We detected a significant number of HDs (n = 53) on 17 different loci of the 246 tested. Some of the data presented here confirm previously published cases of HDs (SMAD4 in FADU, MSH2 in LOVO, RB in SAOS2, both WWOX deletions and some of the p53, FHIT and p16-INK4A cases).22, 32, 43 However, we provide evidence for novel examples of HDs at some of these loci and, more significantly, we report deletions at new loci.
In most cases, TSG loss of function results from a combination of genetic abnormalities including LOH, point mutations or promoter methylation, whereas HD generally represents an infrequent mechanism of TSG inactivation. The only notable exceptions are the p16-INK4A/p14-ARF/p15-INK4B and SMAD4 loci where HDs are frequent. Remarkably, for a given gene, the rate of HDs is not necessarily representative of the global incidence of biallelic inactivation, and certain TSGs appear to be almost never homozygously deleted. For example, SOCS1 and IGF2/M6PR, previously shown to be inactivated by methylation and LOH/point mutation in more than half of the liver cancers studied,44, 45 have as yet not been shown to be affected by HDs. Furthermore, significant variations in HD prevalence among tumor types exist for a given TSG. For example, mutation rates of SMAD4 are similar in pancreatic and colorectal cancers (40–50%). Yet, in pancreatic xenografts, HDs of SMAD4 represent 65% of all mutations, whereas in colorectal cell lines, HDs are found in only 30% of mutated samples.5, 46 Thus, these data suggest that defining a cutoff of significance for HDs at a given TSG may be misleading. Therefore, global biallelic inactivation rate, taking into account LOH, point mutations, methylation and HD, may be the only value providing accurate information about the importance of a TSG in a given tumor type.
Regarding bona fide and candidate TSGs associated with particular neoplasms in our present work, further studies are required to assess the actual percentages of mutations in the corresponding cancers. We already know that there is only a low prevalence of abnormalities at TP73 and MAP2K4 in most tumors studied so far,40, 41 and it is thus possible that this situation also applies to lung, ovarian or head and neck carcinomas for other candidates identified in our present study (NF2, LRPDIT, STK11, MAP2K3). However, in tumorigenic pathways, all intervening partners are not evenly mutated. In lung cancer, for example, RB, p16-INK4A and cyclin D1, which interact at the same step of cell cycle, are altered at different rates.47 Therefore, we can hypothesize that HDs detected in our present survey may represent only fractions of all abnormalities affecting their respective growth control pathways. There is, in addition, a possibility for HDs found in cell lines to be in vitro-selected events. However, most of the works leading to TSG identification associate fresh tumors and established cell lines. In these studies, differences between both kinds of samples rely on the magnitude of mutation rates detected rather than on discrepancies such as absence/presence of mutations.5, 7, 9, 48
Our study revealed the presence of deletions in well-established TSGs not previously associated with certain types of cancers. Here, we describe for the first time HDs at NF2 in lung and ovarian carcinomas as well as at STK11 in a cervical carcinoma. In addition, our present survey disclosed potential roles for LRPDIT, MAP2K3, MAP2K4 and TMPRSS2 in ovarian, bone and head and neck tumors (Table III). NF2 is known to be mutated in familial or sporadic neoplasms of the nervous system.49NF2 has, however, been associated with mutations in seemingly unrelated tumors such as breast and colon carcinomas, melanomas and mesotheliomas.50 According to our data, it is possible that NF2 might also be implicated in lung and ovarian tumors. Therefore, it is reasonable to assume that the spectrum of NF2 mutations is larger than previously believed. STK11 (also termed LKB1) was recently shown to have a key role in p53-induced apotosis or in BRG1-induced growth arrest.33, 51 STK11 is mutated in a restricted set of sporadic tumors (testicular tumors, melanoma, biliary and pancreatic cancer).52, 53, 54 Our data suggest that STK11 may also be involved in cervical carcinoma since HELA cells display a double deletion affecting exons 1–4. Surprisingly, the HD at STK11 locus was hitherto unknown despite the conspicuous absences of STK11 mRNA and protein in previous expression and methylation studies.55, 56LRPDIT encodes for a large plasma membrane receptor protein of 4,599 amino acids. HDs in LRPDIT were found originally in lung, kidney and bladder cancer cell lines and were shown more recently in high-grade bladder cancers.23, 57 Here we provide evidence suggesting that LRPDIT is also an HD target in ovarian carcinoma as well as in osteosarcoma. The impact of LRPDIT deletion on tumorigenesis is unknown, but its preferential alterations in advanced tumors suggest that it is not a rate-limiting event but rather involved in tumor progression. Finally, we describe the first homozygous deletion in TP73. p73 has significant sequence and functional similarities with p53. However, together with the third member of the family, p63, p73 is involved in development rather than acting as a tumor suppressor. The TP73 gene encodes different protein variants, 2 major forms of which differ by the presence or the absence of the first 3 exons and thus display highly divergent activities.41 Significantly, in the SC10 cell line, the 5′ region of p73 has been homozygously lost, thus inactivating the impact of p73 on apoptosis and cell cycle. Our data suggest, therefore, that in selected cases, lung cancer may be partially controlled by p73 activity. Another finding lies in the homozygous losses of telomeres 17p and 4q. It has been recently hypothesized that, preceding telomerase activation in tumors, chromosome attrition may lead, in some circumstances, to chromosome breaks subsequently stabilized through unbalanced translocations.16 According to the present screen, such attrition-driven homozygous loss occurs for more than 1% of telomeres tested. In the panel of cell lines used, chromosome 17p is a prominent target for such phenomenon, being involved in 3 HDs (Table III). In addition, chromosome 17p as a whole appears to undergo a surprisingly wide spectrum of alterations with 3 additional loci affected by HD (TP53, MAP2K3 and MAP2K4).
The main purpose of our present survey was to identify previously unknown associations between genetic pathways and certain types of carcinomas as well as to find novel TSG. Among the 197 different genes tested for HD, only 17 (8.6%) were positive, with all but TP73 previously reported to undergo double deletion. Although we cannot rule out the possibility that HDs may occur at random during tumorigenesis, our data suggest that such alterations preferentially involve unstable regions of the genome or, alternatively, are selected when they affect targets that control cellular growth. The findings of HDs at bona fide TSGs not typically associated with certain tumors suggest that some pathways, thus restricted to specific neoplasms, may have a more general involvement in tumorigenesis (e.g., STK11 in cervical carcinoma). Another salient feature of our current study is the diversity of HD targets in lung, ovarian or even head and neck carcinomas, thus emphasizing the multiplicity of pathways implicated in sporadic cancers. Finally, our data provide novel information for tumor cell lines widely used as in vitro models in cancer research (e.g., CALU1 or HELA). More generally, the present HD scanning of a tumor cell line panel represents a pilot study toward a detailed genetic annotation of common cellular models. These results should thus complement existing and future data on the transcriptome, on the proteome, the allelic status at critical loci as well as on the mutation and methylation status of recurrently altered genes.
We thank Dr. P. Coulie (Université Catholique de Louvain, Faculté de Médecine, Unité de Génétique Cellulaire, Bruxelles, Belgium), Dr. O. Delattre (Laboratoire de Pathologie Moleculaire des Cancers, INSERM U509, Institut Curie, Paris, France), Dr. J.-L. Fischel (Unité d'Oncopharmacologie, Centre Antoine Lacassagne, Nice, France), Dr. H. Gabra (Medical Oncology Unit, Western General Hospital, Edinburgh, United Kingdom), Dr. P. Gauduchon (Laboratoire de Cancérologie Expérimentale, Centre François Baclesse, Caen, France), Dr. S. Jozan (Laboratoire d'Histologie-Embryologie, Faculte de Medecine Toulouse-Rangueil, Toulouse, France), Dr. B.M. Kacinski (Department of Therapeutic Radiology and Dermatology, Yale University School of Medicine, New Haven, CT), Dr. J.-L. Merlin (Laboratoire de Recherche en Oncologie, Centre Alexis Vautrin, Vandoeuvre-les-Nancy, France) and Dr. H. Rochefort (Laboratoire de Biologie Cellulaire, INSERM U148, Faculté de Médecine-CHU, Montpellier, France) for providing us with cell line samples. We are grateful to Dr. J.S. Seeler for careful reading and helpful discussion of the manuscript.