Differentiation of Human Embryonal Carcinomas In vitro and In vivo Reveals Expression Profiles Relevant to Normal Development

Embryonal carcinoma is a histologic subgroup of testicular germ cell tumors (TGCTs), and its cells may follow differentiation lineages in a manner similar to early embryogenesis. To acquire new knowledge about the transcriptional programs operating in this tumor development model, we used 22k oligo DNA microarrays to analyze normal and neoplastic tissue samples from human testis. Additionally, retinoic acid–induced in vitro differentiation was studied in relevant cell lines. We identified genes characterizing each of the known histologic subtypes, adding up to a total set of 687 differentially expressed genes. Among these, there was a significant overrepresentation of gene categories, such as genomic imprinting and gene transcripts associated to embryonic stem cells. Selection for genes highly expressed in the undifferentiated embryonal carcinomas resulted in the identification of 58 genes, including pluripotency markers, such as the homeobox genes NANOG and POU5F1 (OCT3/4), as well as GAL, DPPA4, and NALP7. Interestingly, abundant expression of several of the pluripotency genes was also detected in precursor lesions and seminomas. By use of tissue microarrays containing 510 clinical testicular samples, GAL and POU5F1 were up-regulated in TGCT also at the protein level and hence validated as diagnostic markers for undifferentiated tumor cells. The present study shows the unique gene expression profiles of each histologic subtype of TGCT from which we have identified deregulated components in selected processes operating in normal development, such as WNT signaling and DNA methylation.

In many respects, germ cell tumorigenesis resembles early embryogenesis (1). The most common type of human germ cell tumor, testicular germ cell tumor (TGCT) of adolescents and young adults, develops from premalignant and noninvasive intratubular germ cell neoplasia (IGCN, also called carcinoma in situ) and is histologically classified into seminoma, nonseminoma, or a combination of the two (2). Whereas seminomas morphologically resemble IGCN, nonseminomas are a more heterogeneous group of tumors. In addition to the undifferentiated and pluripotent embryonal carcinomas, which have a capacity to differentiate into various lineages, the nonseminomas include yolk sac tumors and choriocarcinomas with extraembryonic differentiation as well as teratomas with somatic differentiation (Fig. 1A and B).

Several cell lines derived from TGCT express features typical of embryonal carcinomas (3). A few of these cell lines, like the pluripotent NTERA2, differentiate extensively in culture in response to morphogens like retinoic acid (RA; ref. 4). When xenografted, they spontaneously differentiate into teratomas (5). Thus, induction of differentiation of NTERA2 cells is a model for the in vivo differentiation of embryonal carcinomas (Fig. 1). However, other established human embryonal carcinoma cell lines, such as the 2102Ep, are relatively nullipotent and remain in the undifferentiated state even after RA treatment (6); they form xenografts of pure embryonal carcinoma histology (7).

To explore the transcriptional programs during germ cell tumorigenesis, we studied samples of each histologic subtype of TGCT, testicular parenchyma from healthy individuals, and samples taken from testis with IGCN. We also tested the RA-induced in vitro differentiation of the two embryonal carcinoma cell lines NTERA2 and 2102Ep. Particular attention was paid on dysregulation of gene categories known to be related to the early embryogenesis, such as the homeobox gene family, genes specifically expressed in embryonic stem cells, genes involved in the WNT, transforming growth factor-β (TGF-β), and Notch signaling pathways, and genes related to DNA methylation and imprinting.

Tissue samples and cell culture. The tumors were histologically classified according to the WHO recommendations (2). Only tissue samples containing a single histologic component from H&E-stained sections of the frozen tissue were selected (Table 1). Included were normal testicular parenchyma (n = 3), premalignant IGCN (n = 3), and the malignant histologic subtypes seminoma (n = 3), embryonal carcinoma (n = 5), yolk sac tumor (n = 4), choriocarcinoma (n = 1), and teratoma (n = 4). Two of the three normal testicular tissue samples were from organ donors (22 and 29 years) with no known history of cancer. The third morphologically normal tissue was obtained adjacent to a TGCT. The present use of the human clinical material is approved by our institution, the Norwegian Radium Hospital, University Division, and the Regional Committee for Medical Research Ethics, The Health Region South, Norway.

Cultures of the TGCT cell lines NTERA2/clone D1 and 2102Ep/clone 2A6 were maintained at high cell densities (>5 × 10⁶ cells per 75 cm² tissue culture flask) and incubated at 37°C under a humidified atmosphere of 10% CO₂ in air in DMEM high glucose formulation (Life Technologies, Paisley, United Kingdom) supplemented with 10% FCS. NTERA2 cells were harvested for subculture by scraping with glass beads (5). 2102Ep cells were harvested using 0.25% trypsin in calcium- and magnesium-free Dulbecco's PBS containing 1 mmol/L EDTA (trypsin-EDTA; ref. 4). To induce differentiation, cells were harvested by using trypsin-EDTA and seeded at 10⁶ per 75 cm² tissue culture flask in medium containing 10⁻⁵ mol/L all-trans RA (Eastman-Kodak, Rochester, NY; ref. 4). For both NTERA2 and 2102Ep, cell pellets for RNA isolation were made at time points 0, 3, and 7 days of culturing with RA.

Gene expression microarrays. We used Agilent Human 1A oligomicroarrays (GEO accession no. GPL885, Agilent Technologies, Palo Alto, CA) containing 60-mer DNA probes synthesized in situ in a 22k format. Of 19,061 spots, 18,086 are noncontrols and there are 17,086 unique transcript sequences, matching to 15,989 unique human genes.⁸

Probe preparations, hybridizations, image generation, and image analyses were done according to the manufacturer's protocol. Briefly, we obtained the total RNA fraction from ground tissue samples (in liquid N₂) and cell culture pellets by using the Trizol reagent (Life Technologies, Rockville, MD). The RNA quality was evaluated by use of the Agilent 2100 Bioanalyzer (Agilent Technologies). Labeled cDNA was synthesized from 20 μg RNA (Fluorescent Direct Label kit, Agilent Technologies) in the presence of cyanine 3-dCTP for the test sample and cyanine 5-dCTP (Perkin-Elmer Life Sciences, Boston, MA) for the common reference, consisting of a pool of 10 human cell lines, including one from embryonal carcinoma (Universal Human Reference RNA, Stratagene, La Jolla, CA). Differentially labeled test and reference samples were mixed with Agilent control targets before hybridization onto the oligomicroarrays for 17 hours at 60°C in a rotating oven.

Data processing and statistics. The fluorescence intensities at the targets were detected by a laser confocal scanner (Agilent Technologies), and resulting images were processed using the Feature Extraction software, version 6.1.1.1 (Agilent Technologies). This included defining the spots, measuring intensities, flagging spots with inadequate measurements, subtracting local background, and LOWESS dye normalization. For spots that were not flagged as having inadequate measurements, ratios (sample over reference) of the processed intensities were used further. The measured ratios of all genes and samples were divided by the median of the ratios of the three normal samples before log₂ transformation. This was done to facilitate the interpretation of the expression values, because now the sign of the value indicates relative up-regulation or down-regulation in the sample compared with the expression level in normal testis.

For the 60-mer DNA sequences that are represented more than once on the array (100 sequences are present in 10 times abundance), the medians were used for further analyses. The total gene expression matrix contained after this adjustment results for 17,086 unique sequences, of which 259 had technically inadequate measurements in >5 of the 29 samples and were excluded from the analyses. Thus, we ended up with 16,817 unique and valid transcript sequences represented by 15,756 individual gene symbols. For further analyses, missing values were estimated by the k-nearest neighbor imputation (k = 10).

Principal component analysis was done by input of the 1,575 genes that were altered in at least three samples to at least a 4-fold deviation from the median across the sample set (J-Express Pro version 2.0, MolMine, Bergen, Norway). Hierarchical clustering for improved visualization in Figs. 2 and 3 was done by average linkage of Euclidean distance measures (J-Express Pro).

Significance analysis of microarrays (SAM; ref. 8) was used to extract significantly differentially expressed genes characterizing each of the histologic subtypes. Here, the same false discovery rate (FDR) for all subtypes could not be used as, for example, a 1% FDR resulted in from 0 significant genes for the IGCN to 780 for the normal samples. Rather, we selected the same number of up-regulated and down-regulated genes from each histologic subtype (the 80 most up-regulated genes and the 20 most down-regulated genes) to the interhistologic subtype-regulated gene list.

In the first step of finding discriminator genes for the undifferentiated and pluripotent phenotype, a SAM analysis of embryonal carcinomas versus differentiated tissues was done (1.84% FDR; d = 0.965), resulting in 217 significant genes. These were again filtered, leaving only genes that were significant in the same direction from a second SAM analysis (10% FDR) comparing undifferentiated cell lines (the three samples from 2102Ep and the untreated sample from NTERA2) in one group and cell lines where differentiation were induced (the two treated NTERA2 samples) in the other.

Functional classes of gene products were assigned according to the Gene Ontology (GO) Consortium database.⁹

The total list of gene symbols with valid DNA microarray measurements and our lists of significant genes were along with the June 2004 version of the GO database uploaded into the GoMiner software (9). Here, the null hypothesis is that genes are evenly distributed among the GO categories represented by the genes on the microarray, and Fisher's exact tests are done to test the significance level for enrichment of genes within each GO category. Also for the pairwise comparisons of the self-curated gene lists (Table 2), Fisher's exact tests were used to estimate the statistical significance levels.

Protein expression on tissue microarrays. In situ protein expression analyses of GAL and POU5F1 were done using immunohistochemistry on tissue microarrays containing 510 testicular tissue cores covering all histologic subtypes and clinical stages of TGCT (10). Tissue microarray sections were stained by immunohistochemistry using the EnVision+ system, peroxidase-horseradish peroxidase by following the manufacturer's protocol (DakoCytomation, Glostrup, Denmark). Briefly, one tissue microarray section for each antibody was deparaffinized and rehydrated, and high temperature antigen retrieval was done by microwave oven at 900 W. The endogenous peroxidase activity was blocked before incubation with the polyclonal antibodies GAL (C-20, sc-16413, 1:100, 2 μg IgG/mL, Santa Cruz Biotechnology, Inc., Santa Cruz, CA) and POU5F1 (H-134, sc-9081, 1:100, 2 μg IgG/mL, Santa Cruz Biotechnology) for 30 minutes at room temperature. Next, the sections were incubated with labeled polymer for 30 minutes and with the peroxidase substrate 3′3-diaminobenzidine tetrahydrochloride for 10 minutes before the sections were counterstained using hematoxylin, dehydrated, and mounted. Negative results on controls using IgG from nonimmunized rabbit and goat were achieved.

For GAL, cases with granular staining in the cytoplasm and/or the extracellular spaces were scored as positive. For POU5F1, tissue cores with staining in >5% of the relevant nuclei were scored as positive.

Gene expression profiles. Testicular tumorigenesis was investigated by gene expression profiling of normal testis, premalignant IGCN, the various malignant histologic subtypes of TGCT (seminoma, embryonal carcinoma, choriocarcinoma, yolk sac tumor, and teratoma), and two TGCT cell lines before and after differentiation (Fig. 1B). DNA microarrays containing oligo 60-mers of 17,086 well-annotated genes and transcripts were used to measure mRNA levels relative to a common universal human reference RNA. Raw data were deposited to the Gene Expression Omnibus public repository for microarray data (GEO accession no. GSE1818)¹⁰

according to the Minimum Information About a Microarray Experiment (MIAME) recommendations for recording and reporting microarray-based gene expression data (11).

To illustrate the relationship between expression profiles of the individual samples, principal component analysis was carried out (Fig. 1C). Here, samples from the same histologic subtypes appear in the vicinity of each other, demonstrating, as expected, that part of the variability in gene expression profiles of TGCTs is dependent on histologic type. A measure of the actual differences between the individual histologic subtypes is shown in the matrix of numbers of genes with expression medians deviating >3-fold between each pair of all combinations of the histologic subtypes (Fig. 1D). For example, there are 292 genes with a >3-fold higher expression value in seminomas than in embryonal carcinomas and vice versa and 182 genes that were more highly expressed by the embryonal carcinomas compared with seminomas. Interestingly, the pluripotent, undifferentiated cell lines have a similar number of genes differentially expressed from seminomas as they have from the embryonal carcinomas, the subtype that they are generally believed to model.

Histologic subtype-specific gene expression. We identified sets of differentially expressed genes within each of the histologic subgroups compared with the rest of the samples (individual SAM analyses; ref. 8). The top 10 up-regulated genes within each subgroup are shown in Fig. 2. From each of the seven histologic subtypes, 80 genes with the most significant up-regulation and the 20 most down-regulated genes were listed as interhistologic subtype-regulated genes (Supplementary Table S1). As 13 genes appeared for more than one histologic subtype, the complete set contained altogether 687 genes and not the expected 700. The GOs of the differentially expressed genes were explored, and several GO categories were significantly enriched within the interhistologic subtype-regulated gene list (Table 3; Supplementary Table S2). Significance levels of enrichment of genes belonging to embryogenesis-related gene categories were also calculated (Table 2). The 687 interhistologic subtype-regulated genes were particularly enriched among embryonic stem cell–specific genes (12 of 83, P = 2 × 10⁻⁴) identified by a DNA microarray study (12).

Comparison of embryonal carcinomas and seminomas. By comparing the data from seminomas and embryonal carcinomas, we extracted a list of 169 genes significantly differentially expressed between these histologic types (SAM, 2% FDR; Supplementary Table S4). Thirty-nine of these genes were specifically expressed in embryonal carcinomas compared with seminomas, and among these, GAL was the most significantly differentially expressed gene. Some of the other significant genes have been linked previously to TGCT [i.e., GDF3 (13, 14), DNMT3B (15), and BCAT (16)]. Among the 130 genes with higher expression in seminomas, PRAME (preferentially expressed antigen in melanoma) was the most significantly different, and the list also included the previously published SLC43A1 (POV1; refs. 17, 18) and KIT (19–21). To ascertain how the TGCT cell lines express these 169 genes, a hierarchical cluster analysis showed that, as might be anticipated, the cell lines cluster more tightly to the embryonal carcinomas than to the seminomas (Supplementary Fig. S1). However, several of the genes with highest expression in the seminomas compared with the embryonal carcinomas are also highly expressed in the cell lines. This suggests that the TGCT cell lines 2102Ep and NTERA2 have a transcriptional phenotype that to some extent lies between those of seminoma and embryonal carcinoma histologic subtypes.

Undifferentiated and pluripotency-specific gene expression. To identify genes that are robustly differentially expressed between undifferentiated and pluripotent embryonal carcinoma and differentiated tissues or cell lines, a stepwise gene selection was implemented. First, SAM analysis was done, comparing the two extreme in vivo groups in terms of differentiation [i.e., the highly undifferentiated embryonal carcinomas (n = 5) versus the differentiated tissue types (n = 12; normal testis, yolk sac tumor, choriocarcinoma, and teratoma)]. This resulted in 176 genes that were most highly expressed in the embryonal carcinomas and 41 genes that were most highly expressed in the differentiated tissues (Table 4; Supplementary Table S5). The same analysis was also done, including only malignant tissue samples, and again, the same five genes, GAL, POU5F1, NANOG, DPPA4, and MT1H, appeared as the most significantly overexpressed in embryonal carcinoma (data not shown). The 217 (176 + 41) genes that were overall differentially expressed between the tissue samples were filtered further, leaving only those genes showing up-regulation or down-regulation in the in vitro models of TGCT differentiation (altered in the same direction after RA treatment for the pluripotent NTERA2 but remained unchanged in the relatively nullipotent 2102Ep). This left us with 68 genes, 58 that are characteristic of the undifferentiated embryonic stem cell–like state and 10 that are characteristic of the differentiated derivatives (Fig. 3). Significantly enriched GO terms assigned to these 68 genes are listed in Table 5. The involvement of these genes in various gene function categories related to developmental biology are found from Table 2. Gene categories, such as DNA methylation genes, homeobox genes, and genes highly expressed in embryonic stem cells, were particularly abundant among the 68 genes specifically expressed in undifferentiated embryonal carcinomas.

Intratubular germ cell neoplasia versus normal testis. A comparison of normal testis and premalignant IGCN yielded 96 significantly differentially expressed genes (SAM, 1.6% FDR; Supplementary Fig. S2). The pluripotency-specific genes NANOG, DPPA4, and POU5F1 as well as the KIT gene were all among the 12 most significantly up-regulated genes in the premalignant IGCN.

Protein expression of GAL and POU5F1. The in situ protein expression of GAL and POU5F1, two genes identified as significantly overexpressed in embryonal carcinoma, were analyzed on tissue microarrays containing 510 tissue samples from normal testis, IGCN, and the various histologic subgroups of TGCT. Both GAL and POU5F1 were negative for all normal testis samples but were positive for 21% and 54% of the overall tumor samples, respectively, and both with significant differences between various histologic subtypes (Fig. 4). Interestingly, the frequency of POU5F1-positive seminomas (63%) was similar to that of embryonal carcinomas (65%), whereas for GAL this was different (8% positive seminomas and 34% positive embryonal carcinomas; P = 5 × 10⁻⁶).

Gene expression studies in human testicular germ cell tumors. We have explored the transcriptional programs altered during the early embryo-like development of TGCT and present gene expression profiles from both an in vivo and an in vitro setting from pluripotent and undifferentiated tumor subtypes to those more differentiated. To our knowledge, this is the first comprehensive DNA microarray study analyzing all the malignant histologic subtypes of TGCT as well as premalignant IGCN and normal testis. The applied oligomicroarrays, covering most annotated genes in our genome, and the universal human reference RNA are commercially available; thus, the current expression data can readily be compared with and supplemented with other expression data. The first study applying microarrays to TGCT reported expression of genes located to chromosome 17 as well as of proto-oncogenes located genome-wide (17). Following this, additional microarray studies of primary TGCT have been published, either focusing into specific genomic regions (16, 22) or concentrating on selected histologic subtypes (15, 18). These studies were all done by use of different platforms, mostly in-house made DNA microarrays and/or reference RNA samples, and cross-comparisons between studies are thus not a trivial issue.

Histologic subtype-specific gene expression and parallels to early embryogenesis. The transcriptional profiling of individual histologic subtypes using SAM analyses revealed mostly novel genes with respect to testicular tumorigenesis. However, some previously known biological and clinical markers also appeared and add support to the importance of the resulting genes, reliability of the technology, and strength of the statistics [e.g., α-fetoprotein (AFP) was highly expressed in yolk sac tumors, and subunits of the chorionic gonadotropin (CGB5 and CGA) were highly expressed in choriocarcinomas (Fig. 2)]. In addition, when comparing normal testis and IGCN (Supplementary Fig. S2), several genes that are expressed in the IGCN samples, such as KIT and the pluripotency markers NANOG and POU5F1, were also identified in a recent DNA microarray study focusing on IGCN (23). Among the novel genes to this study, DPPA4 was specifically expressed in IGCN compared with normal testis, and interestingly, the expression of these pluripotency-specific genes was also seen in seminomas (Fig. 3).

When we searched for gene expression patterns characterizing the pluripotent and undifferentiated phenotype of embryonal carcinomas, we did therefore not include IGCN and seminomas in any of the groups compared to avoid exclusion of the pluripotency-specific markers. The 68 resulting genes with specific expression pattern in the undifferentiated and pluripotent phenotype of embryonal carcinomas included 7 genes that were reported previously as enriched in embryonic stem cells (P = 1 × 10⁻⁷; Table 2; ref. 12), and 2 of these, NANOG and POU5F1, were also abundantly present in IGCN and seminomas. Both these genes are known as key regulators of pluripotency (24–26), and IGCN and seminoma cells might therefore have similar pluripotent capabilities as embryonal carcinoma cells. Also on the protein level, as seen from immunohistochemistry on the tissue microarrays, POU5F1 was present in most IGCN and seminomas. This was not the case for GAL, which was highly associated to embryonal carcinomas compared with seminomas. However, a few seminomas were also positive for GAL, which lends support to the existence of a distinct subgroup of seminomas that resembles embryonal carcinomas on the molecular level.¹¹

For the nonseminomas, morphologic parallels to early embryogenesis are described. In the same way that embryonal carcinomas share many characteristics with the inner cell mass of blastocysts (27), the choriocarcinomas parallel the trophoblastic and syncytiotrophoblastic cells of the placenta, yolk sac tumors parallel the endodermal differentiation giving rise to the yolk sac, and teratomas parallel the differentiation into somatic tissues of all three primary germ layers (2). Hence, transcriptional programs of the early embryogenesis are likely to be altered in testicular tumorigenesis. The enrichment of embryonic stem cell–specific genes in testicular tumorigenesis validates this morphologic link to early embryogenesis also on the gene transcription level. Hence, by presenting gene expression patterns specific to the histologic subtypes of human TGCT (Fig. 2), we also provide novel transcriptional information relevant to the parallel differentiation stages of human development.

Pluripotent embryonal carcinomas are considered to be the malignant counterparts of embryonic stem cells (28). Embryonic stem cells are again potential mediators to regenerative medicine, but as transplants may be rejected from the host's immune system, it would be advantageous if personalized embryonic stem cells could be made from human adult stem cells by dedifferentiation back to the embryonic stage without the need for cloning (29). The identified embryonal carcinoma–specific transcripts (as in Fig. 3) can potentially work as reprogramming factors that lead to dedifferentiation of adult cells into personalized embryonic stem cells. Human embryonic stem cells have thus far been considered pluripotent, rather than totipotent, because of their limited differentiation potency into extraembryonic membranes and tissue types (30). Recently, however, they were shown to have the capability to differentiate into trophoblast cells (31), an observation that may broaden the definition of their developmental potential. However, as TGCTs often contain tissue types with extraembryonic morphology (yolk sac tumor and choriocarcinoma), the current histologic subtype-specific gene expression profiles (Fig. 2) may indeed contain information that can be useful in the search for ways to make the embryonic stem cells differentiate into these lineages; e.g., by overexpression of genes specifically expressed in the TGCT subtype paralleling the desired lineage differentiation.

Gene categories involved in early embryogenesis. Knowledge from the field of embryogenesis may also be used to understand germ cell tumorigenesis, and we have identified subsets of genes that are also known to play a role in early embryonic development (Table 2). For example, all the four most significantly overexpressed genes in the undifferentiated embryonal carcinomas, POU5F1, GAL, NANOG, and DPPA4 (Table 4), are known as pluripotency-associated genes (12). For WNT signaling, we found seven components of this pathway within the interhistologic subtype-regulated genes list (Table 2). Except FRAT2, which was highest expressed in seminomas and embryonal carcinomas, these WNT signaling components (AES, DKK1, FZD4, TCF7L1, and TCF7L2) were highest expressed in yolk sac tumors and teratomas. Further, the three significantly altered target genes of this signaling pathway, FZD7, ID2, and, SLC2A1, were also up-regulated in the yolk sac tumors and teratomas. Thus, activation of the WNT signaling pathway may be a key step in the transition of embryonal carcinomas into differentiated nonseminomas.

High expression of the de novo DNA methyltransferase DNMT3B and its homologue involved in establishment of imprinting, DNMT3L, was significantly associated with the embryonal carcinoma subtype (Fig. 3; Tables 2 and 4). The third homologue, DNMT3A, had a similar expression profile not reaching statistical significance. This discovery of an up-regulated DNA methylation machinery in embryonal carcinomas provide a malignant parallel to the complete epigenetic reprogramming events taking place in the inner cell mass of the blastocysts (32, 33).

In vivo versus in vitro clues to potential drug targets. The RA-induced differentiation of embryonal carcinoma cell lines in this study is used to model the in vivo differentiation of embryonal carcinomas. The shared transcriptional changes between the in vivo and the in vitro systems (Fig. 3) are of special interest as they may be specific and/or necessary to make embryonal carcinoma differentiate and play biologically as well as clinically important roles. In fact, one of the genes sorted out by this approach, NALP7, was recently confirmed to play a functional oncogenic role in TGCT (34). In the treatment of TGCT, it could therefore be advantageous to repress the genes specifically expressed in the undifferentiated cells (e.g., GAL, POU5F1, NANOG, and DPPA4), as growth and malignant potential could be repressed by forcing the cells into terminal differentiation. Further intriguing, cancer drugs targeting genes that in the normal setting are specifically expressed during the blastocyst stage should in the postembryonal body be specific to tumor cells.

The current data could also facilitate cancer drug discovery by use of the genes always expressed in the differentiated tissues but not in the undifferentiated ones (e.g., MGC5576, TTC3, or C9orf154; Table 4), genes that may be used as general markers of differentiation. Their gene products can be used in drug screening programs where the agents causing expression of these selective markers in embryonal carcinoma cells are considered candidate cancer drugs.

In conclusion, we have presented the first comprehensive gene expression profiling of all histologic subtypes of TGCT as well as of precursor lesions and normal testis. This pinpointed specifically expressed genes within all analyzed subgroups. We also uncovered novel transcriptional information about tumorigenesis and its functional relation to normal development. The key genes described, as GAL and POU5F1, will be useful in molecularly assisted diagnosis as well as serve as potential drug targets.

Grant support: Norwegian Cancer Society grant A95068 (R.A. Lothe, R.I. Skotheim, and G.E. Lind), Nordic Academy for Advanced Study (R.I. Skotheim), and Biotechnology and Biological Sciences Research Council and Yorkshire Cancer Research (P.W. Andrews).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

We thank Christine Pigott for the cell culturing work and Tuula Airaksinen for the demonstration of DNA microarray protocols.