A transgenic mouse model of metastatic carcinoma involving transdifferentiation of a gastric epithelial lineage progenitor to a neuroendocrine phenotype

Abstract

Human neuroendocrine cancers (NECs) arise in various endoderm-derived epithelia, have diverse morphologic features, exhibit a wide range of growth phenotypes, and generally have obscure cellular origins and ill-defined molecular mediators of initiation and progression. We describe a transgenic mouse model of metastatic gastric cancer initiated by expressing simian virus 40 large tumor antigen (SV40 TAg), under control of regulatory elements from the mouse Atp4b gene, in the progenitors of acid-producing parietal cells. Parietal cells normally do not express endocrine or neural features, and Atp4b-Cre bitransgenic mice with a Cre reporter confirmed that the Atp4b regulatory elements are not active in gastric enteroendocrine cells. GeneChip analyses were performed on laser capture microdissected SV40 TAg-expressing cells in preinvasive foci and invasive tumors. Genes that distinguish invasive from preinvasive cells were then hierarchically clustered with DNA microarray datasets obtained from human lung and gastric cancers. The results, combined with immunohistochemical and electron microscopy studies of Apt4b-SV40 TAg stomachs, revealed that progression to invasion was associated with transdifferentiation of parietal cell progenitors to a neuroendocrine phenotype, and that invasive cells shared molecular features with NECs arising in the human pulmonary epithelium, including transcription factors that normally regulate differentiation of various endocrine lineages and maintain neural progenitors in an undifferentiated state. The 399 mouse genes identified as regulated during acquisition of an invasive phenotype and concomitant neuroendocrine transdifferentiation, plus their human orthologs associated with lung NECs, provide a foundation for molecular classification of NECs arising in other tissues and for genetic tests of the molecular mechanisms underlying NEC pathogenesis.

Tumors with neuroendocrine (NE) features arise in tissues located throughout the human body. They exhibit diverse morphologic patterns and range from indolent to highly aggressive (1). This diversity has impeded development of a consensus system for their classification (2, 3). Neuroendocrine cancers (NECs) that have a neuroectodermal origin (“neural-type,” such as those that arise in the adrenal medulla) usually express neurofilaments, whereas those that have an epithelial, most commonly endodermal, origin express cytokeratins. Epithelial-type NECs occurring in organs such as the lung, prostate, or gastrointestinal tract are thought not to have a neuroectodermal origin (1, 4).

In the lung, NECs account for up to a quarter of epithelial tumors. Travis and coworkers (5) have developed a system to classify these NECs on the basis of their morphological phenotype. Such distinctions have important therapeutic and prognostic implications: relatively indolent NE tumors known as “carcinoids” rarely metastasize, whereas small cell carcinomas are almost universally fatal and have commonly metastasized by the time of initial diagnosis (5–7).

Because NECs are typically discovered long after the initiating event, the features of their cell of origin, and the molecular pathways they follow during their progression, are poorly understood. Thus, it would be advantageous if NE carcinogenesis could be modeled in mice. Unfortunately, there are only a few reported mouse models where there is reproducible development of NECs from a known cell of origin. For example, NE cells represent one of the three lineages that constitute the normal mouse prostate epithelium. Forced expression of simian virus 40 large tumor antigen (SV40 TAg) in a subset of these cells produces a metastatic NEC with a rapid and stereotypic pattern of evolution in multiple pedigrees of transgenic mice (8). Intratracheal infection of mice, harboring conditional null alleles of Rb1 and p53, with an engineered adeno-Cre virus leads to somatic inactivation of both genes and development of a metastatic small cell-type NEC, although the cell of origin in this model has not been reported (9).

In the present report, we describe development of a metastatic NEC from a committed gastric epithelial progenitor that normally gives rise to a lineage that expresses no NE features: the acid-producing parietal cell (PC) (10). This transgenic mouse model is informative because it illustrates how NECs can arise by transdifferentiation of a progenitor normally dedicated to a non-NE fate.

Gastric cancers, the second leading cause of cancer deaths worldwide (11, 12), are predominantly adenocarcinomas with two general, occasionally overlapping, phenotypes: those with an intestine-like morphology (“intestinal-type”) and those that infiltrate diffusely (“diffuse-type”) (13). NE tumors have traditionally been considered uncommon in the stomach (≈1% of all gastric cancers; e.g., ref. 14). However, during the last three decades, the prevalence of NE tumors of the carcinoid-type has increased dramatically, either because of an increased incidence of hyperplasia of enteroendocrine cells in response to widespread use of long-term acid-suppressive therapy and/or because more pathologists are recognizing this entity (15). As many as 10% of the tumors diagnosed as intestinal- or diffuse-type may have NE features (16, 17).

The mouse is a good model for studying human gastric tumorigenesis because a considerable amount of information is known about the morphologic features of its epithelial progenitors (18–23). As in humans, the mucosa contains innumerable tubular invaginations known as gastric units (24). The epithelium lining these units is continuously renewed. Renewal is driven by multipotent stem cells functionally anchored in a niche positioned in the middle of each unit (the “isthmus”; see Fig. 4A, which is published as supporting information on the PNAS web site). In the corpus region of the stomach (Fig. 4) this stem cell gives rise to five descendant lineages. Three lineages constitute the majority of the ≈200 cells that populate the gastric unit. Mucus-producing pit cells differentiate as they undergo a rapid upward migration from the isthmus through the so-called pit region to the surface epithelium, where they are removed by exfoliation and/or apoptosis (turnover time, 3 days; ref. 19). Zymogenic cells differentiate during a slower downward migration from the isthmus to the base of the unit (turnover time ≈190 days; refs. 20 and 25). Acid-producing PCs are unique in that they differentiate within the isthmus. They arise from a committed preparietal cell (pPC) progenitor with well described morphologic features and then undergo a bipolar migration to the pit and base (10, 22). The enteroendocrine and caveolated lineages represent the remaining, albeit much less prevalent, cell types (21).

Transcriptional regulatory elements from the mouse gene encoding the noncatalytic β-subunit of H⁺,K⁺-ATPase (Atp4b) have been used to selectively deliver a variety of gene products to pPCs (e.g., refs. 26–28). Our earlier studies of 4- to 11-week-old mice containing a transgene consisting of these regulatory elements linked to SV40 TAg revealed that expression of the viral oncoprotein produces a 50- to 70-fold expansion of pPCs and an accompanying block in their differentiation to mature PCs (27). We now show that members of two pedigrees of these transgenic mice develop metastatic gastric cancer associated with the transdifferentiation of pPCs to a NE phenotype. This evolution was defined by GeneChip profiling of laser-capture microdissected cell populations and was confirmed by multilabel immunohistochemistry, transmission electron microscopy (EM), as well as by hierarchical cluster analyses of our GeneChip datasets with publicly available DNA microarray datasets generated from a variety of human NECs.

Materials and Methods

Details of the methods used to (i) generate and maintain pedigrees of Atp4b-SV40 TAg and Atp4b-Cre transgenic mice, (ii) conduct multilabel immunohistochemical and transmission EM studies, (iii) perform GeneChip analysis of navigated laser-capture microdissection cell populations harvested from Atp4p-SV40 TAg animals and their normal littermates, and (iv) undertake hierarchical clustering studies of DNA microarray datasets are all provided in Supporting Materials and Methods, which is published as supporting information on the PNAS web site.

Results and Discussion

Development of Metastatic Gastric Cancer in Mice That Express SV40 TAg in Their pPCs. Given that expression of SV40 TAg under the control of nucleotides –1035 to +24 of the mouse Apt4b gene produces a 50- to 70-fold increase in the number of normally rare pPCs by 12 weeks of age (Fig. 4 B and C), we postulated that continued expression of the viral oncoprotein would predispose to development of carcinoma. Therefore, we examined transgenic and normal littermates from two pedigrees as they aged from 12 to 52 weeks. Transgenic animals from both pedigrees showed the same phenotype, indicating it was not dependent on the site of chromosomal insertion of their transgene.

Between 12 and 32 weeks of age, pPC hyperplasia produced progressive epithelial thickening as well as cyst formation in all animals studied (n = 32) (Fig. 4 D and E and Table 1, which is published as supporting information on the PNAS web site). There was focal dysplasia in some regions containing the expanded pPC population. Dysplastic cells exhibited nuclear heterogeneity and abnormal polarity and were located in gland profiles featuring cellular stratification and occasional “cribriform” characteristics. At 38 weeks, focally invasive cancer was identified in ≈10% of animals. These lesions were typically located at the apex of the gastric units (i.e., near the lumen) and showed loss of glandular architecture. Invasive cells were organized in loose trabeculae or ribbons, were smaller than hyperplastic pPCs, and had increased nuclear-to-cytoplasmic ratios with granular chromatin (Fig. 4F). Carcinoma encompassing the full thickness of the stomach occurred in 60% of mice by 48 weeks (n = 16). Sheets of small relatively monotonous tumor cells infiltrated the submucosa, often accompanied by histologic evidence of lymphatic–vascular invasion (Fig. 1 A, C, and D). By 1 year of age, all animals studied (n = 53) had invasive gastric cancer. Grossly visible nodular masses projected from the serosal surface of the stomachs of 60% of these mice, and there were associated local lymph node metastases. Macroscopic (≥2-mm) hepatic metastases were present in 20% of animals and were composed of solid sheets of small cells with high nuclear-to-cytoplasmic ratios (Fig. 1B). None of the age- and gender-matched normal FVB/N mice in our colony developed gastric lesions (n = 100).

Fig. 1.

Invasive gastric cancer in Atp4b-SV40 TAg mice. (A) Hematoxylin and eosin (H&E)-stained section from a 44-week-old transgenic mouse showing invasion of tumor into the muscle layers of the stomach and into lymph/vascular spaces (arrows highlight nests and cords of tumor cells expanding vessels). (B) H&E-stained section of a liver metastasis containing a solid mass of small cells with exceedingly scant cytoplasm. (C–F) Serial sections from an 11-month-old Atp4b-SV40 TAg stomach. (C) Low-magnification image of an H&E-stained section demonstrating a tumor spanning the entire mucosa. Tumor cells (lower left, asterisk; compare with tubuloglandular profiles in the upper right) form a nearly solid mass and have scant cytoplasm. (D) Higher-magnification view of the area boxed in C. Note that tumor cells do not form glands: instead, they are organized into cords, loose ribbons, and nests (insulae). (E) Adjacent serial section demonstrating that hyperplastic pPCs in glandular profiles exhibit more intense staining with antibodies to SV40 TAg (purple) compared with invasive gastric cancer cells (iGCs). (F) Adjacent serial section demonstrating Atp4b labeling (purple) of cells in glands and no detectable staining in iGCs. (Bars: 25 μm in A,50 μmin B and D–F, 120 μminC.)

Functional Genomics Analysis Reveals That Development of Invasive Cancer Is Associated with Expression of a NE Phenotype. Multilabel immunohistochemical studies of 48-week-old Atp4b-SV40 TAg mice revealed that whereas hyperplastic pPCs expressed SV40 TAg and the normal PC-specific marker β-subunit of H⁺,K⁺-ATPase (Atp4b), invasive tumor cells in the stomach and liver were weakly positive for SV40 TAg and had no detectable Atp4b (Fig. 1 E and F). Intrigued by these differences, as well as by the morphologic differences observed between hyperplastic pPCs and invasive cells, we conducted a broad survey of their molecular properties.

A navigated form of laser-capture microdissection (see Supporting Materials and Methods) was used to harvest (i) hyperplastic SV40 TAg- and Atp4b-positive cells with clear pPC morphology, (ii) weakly SV40 TAg-positive, Atp4b-negative, locally invasive gastric cancer cells (iGCs), and (iii) the small monotonous metastatic gastric cancer cells (mGCs) in the liver (10,000 cells per mouse per population; n = two 48-week-old mice). Each RNA sample from each mouse was used to generate complementary RNA targets that were subsequently hybridized to Affymetrix MG-U74Av2 GeneChips (all Affymetrix CEL files are available at http://gordonlab.wustl.edu). After designating the pPC GeneChips as a baseline, we used Affymetrix mas version 5.0 software to select for transcripts that had received a “Present” call and an “Increase” or “Decrease” call in duplicate comparisons, and had an average signal log ratio ≥1 (average fold change ≥2) across duplicate arrays. A total of 629 probe sets (592 unique UniGene entries) satisfied these selection criteria when comparing iGCs to pPCs (Table 2, which is published as supporting information on the PNAS web site). Our comparison of mGCs and pPC yielded 794 probe sets representing 752 UniGene entries (Table 3, which is published as supporting information on the PNAS web site), whereas a comparison of the iGC versus mGC populations revealed only 73 differences (Table 4, which is published as supporting information on the PNAS web site).

The majority of tumors arising in the human stomach have clear features of adenocarcinoma. In the intestinal-type, mucus-secreting cells are organized in a glandular fashion (tubes or papillae). In the diffuse type, glandular architecture is less prominent but tumor cells are still mucus filled and thought to be derived from glandular elements that have down-regulated expression of genes that mediate cell–cell adhesion (13, 29). We had expected that forced expression of SV40 TAg in a normal gastric epithelial progenitor would lead to expansion of glandular elements, and thus we were not surprised that pPC hyperplasia initially produced tubuloglandular structures. However, the invasive tumors (iGC and mGC populations) that arose in areas of glandular hyperplasia (Fig. 1 C and D) did not have the typical mucus-glandular features of adenocarcinomas that arise in either the corpus or antral regions of the human stomach. Therefore, we first compared the gene expression profiles of the invasive mouse gastric cell populations (iGCs plus mGCs) with profiles documented in cancers arising in another endoderm-derived tissue where a more diverse mixture of epithelial tumors commonly form: the human lung.

Hierarchical Clustering with DNA Microarray Datasets Obtained from Human Pulmonary Neoplasms. A large number of global gene expression profiles for human lung carcinomas are available in the public domain. Because we needed to compare gene expression across different species and different GeneChip platforms, we developed a two-step process that involved initial extraction of a molecular signature of genes that clearly differentiate the iGC and mGC populations (invasive cancers) from SV40 TAg-expressing pPCs and gastric epithelium from normal mice. We then used the human orthologs of these mouse genes to cluster a previously published group of 203 GeneChip profiles of lung carcinomas with diverse phenotypes, as well as normal lung tissue (30). The first step was performed by clustering the pPC, iGC, and mGC GeneChip datasets described above with three GeneChip datasets obtained from normal adult total gastric unit epithelium (harvested by laser-capture microdissection from the corpus region of the stomach; ref. 31). A cluster of 399 unique probe sets was identified representing transcripts that differentiated iGCs and mGCs from their presumptive pPC precursors and members of all other normal gastric epithelial lineages (see boxed area with an asterisk in Fig. 5A and Table 5, which are published as supporting information on the PNAS web site). For the second step, we performed a hierarchical cluster analysis that was limited to the human orthologs of these mouse invasive gastric cancer-associated genes to determine whether this subset of orthologs could also help differentiate human lung tumors from normal pulmonary epithelium, and whether they are preferentially expressed in particular types of human lung carcinoma.

In this cross-platform analysis, we first matched the 399 mouse U74Av2 GeneChip probe set identifiers to human U95Av2 GeneChip probe set identifiers by using UniGene Gene Symbols corresponding to their respective probe sets. Of the 399 genes, 199 were found to be common to both the human and mouse arrays (Table 6, which is published as supporting information on the PNAS web site). This 199-member list was then used as a filter for hierarchical clustering of U95Av2 GeneChips representing NE tumors of the carcinoid type (n = 20), NE small cell lung carcinoma (n = 6), squamous cell carcinoma (n = 21), adenocarcinomas (n = 127), and normal lung control specimens (n = 17). All comparisons were performed using the hierarchical clustering algorithm incorporated in dchip software (32).

Among the transcripts detected by these 199 probe sets, 48 (24%) were enriched in carcinoids (green dendrogram labeled LCa-1 in Fig. 5B). Eight of these 48 transcripts were detected almost exclusively in carcinoids, where they had uniformly high levels of expression (red in Fig. 5B), and low to undetectable expression (green-black) in the 171 other tumors or normal lung controls. They included mRNAs encoding AMPA2 inotropic glutamate receptor (GRIA2), dopa decarboxylase (DDC), transthyretin (TTR), proprotein convertase 1 (PCSK1), secretogranin II (SCG2), chromogranin A (CGA), chromogranin B (CGB), and secretory granule neuroendocrine protein 1 (SGNE1).

The 199-gene filter identified five other tumor clusters: (i) 12 pulmonary adenocarcinomas that had NE characteristics [dendrogram labeled LCa-2 in Fig. 5B; genes responsible for defining this cluster included calcitonin (CALCA), as well as those whose expression was enriched in carcinoids (e.g., DDC, PCSK1)]; (ii) a group of pure squamous cell carcinomas (LCa-3); (iii) a group of pure small cell lung carcinomas (SCLC; LCa-4); (iv) a group of squamous cell cancers and adenocarcinomas (LCa-5); and (v) a group of 17 adenocarcinomas and one squamous cell carcinoma (LCa-6) that had high levels of expression of 9 genes that function predominantly in the synthesis, maintenance, or remodeling of extracellular matrix (ECM) components [e.g., lysyl oxidase (LOX), fibronectin (FN1), osteonectin (SPARC), and multiple collagens]. (Additional information concerning clustered lung samples is provided in Fig. 5C.)

Bhattacharjee and colleagues (30) provided only survival data for adenocarcinomas in their study. Using this information, Kaplan–Meier analysis (winstat excel plug-in software from A-Prompt, Lehigh Valley, PA) disclosed that, compared with all other tumor samples in the dataset, only the LCa-2 cluster (adenocarcinomas with NE features) had a significantly less favorable outcome (P = 0.002).

Together, these results demonstrate that the great majority of genes that differentiate the invasive gastric carcinoma cell populations (iGCs, mGCs) in Atp4b-SV40 TAg mice from pPCs and other epithelial lineages in the normal mouse stomach also help segregate human lung carcinomas that have NE features.

Immunohistochemical and Transmission EM Evidence That pPCs Undergo Transdifferentiation to iGCs with NE Phenotypes. To confirm that invasive gastric cancer cells acquire NE gene expression, follow-up immunohistochemical studies were performed using stomachs from 48-week-old Atp4b-SV40 TAg mice and antibodies directed against three well established NE biomarkers: dopa decarboxylase (Ddc; increased 50-fold in the GeneChip comparison of iGC vs. pPC populations and highly expressed in lung carcinoids), chromogranin A (CgA; increased 20-fold and also expressed at high levels in carcinoids), and tryptophan hydroxylase (Tph1; increased 830-fold). All three proteins were prominently expressed in areas of focal invasive neoplasia but not in adjacent normal-appearing epithelium or in hyperplastic epithelium containing SV40 TAg and Atp4b-positive pPCs (e.g., Fig. 2).

Fig. 2.

The transition from hyperplasia to neoplasia is characterized by increased expression of NE markers and loss of a PC marker. (A) Multilabel immunohistochemical study of serial stomach sections from an 11-month-old Atp4b-SV40 TAg mouse. Chromogranin A (CgA, green cytoplasm), a marker for enteroendocrine cells, is expressed in cells that have weak SV40 TAg staining (red, nuclear; e.g., arrowheads). Arrows point to larger, more intensely staining enteroendocrine cells that are SV40 TAg-negative. (B) Adjacent section demonstrating cells with prominent dopa decarboxylase (Ddc, green) staining, and weak SV40 TAg expression. Note the sharp demarcation between the clusters of strongly and weakly staining SV40 TAg-expressing cells. (C) The β-subunit of H⁺,K⁺-ATPase (Atp4b; green) is limited to the intensely SV40 TAg-positive pPC population. (Bar: 25 μm.)

These results establish that NE biomarkers are expressed in neoplastic SV40 TAg-positive cells. At least two scenarios could account for this observation: initiated pPCs undergo transdifferentiation to a NE phenotype, or SV40 TAg expression is initiated, or becomes more pronounced, in members of the gastric stem cell's descendant enteroendocrine lineage as animals age and invasive cancer appears. The latter scenario implies that nucleotides –1035 to +24 of Atp4b are active in enteroendocrine cells. One piece of evidence arguing against such a scenario is that our immunohistochemical studies did not detect coexpression of SV40 TAg and CgA in serially sectioned stomachs harvested from young (9-week-old) transgenic animals, whereas there was invariable coexpression of SV40 TAg and PC-specific glycans recognized by Dolichos biflorus agglutinin (Fig. 3A).

Fig. 3.

Invasive gastric carcinoma associated with transdifferentiation of pPCs to a NE cell phenotype. (A) Multilabel study showing that SV40 TAg (red) colocalizes with Dolichos biflorus agglutinin (DBA) binding (blue; a PC lineage marker) but does not localize with the enteroendocrine cell marker CgA (green) in the stomach of a 9-week-old Atp4b-SV40 TAg transgenic mouse. (B and C) Section from the stomach of a 7-week-old Atp4b-Cre, R26R bitransgenic mouse. B shows a combined bright-field and immunofluorescence image demonstrating that none of the β-galactosidase (β-gal)-expressing cells (blue, due to Cre-mediated recombination of R26R) belong to the CgA-positive (green) enteroendocrine lineage. C shows the section presented in B viewed with immunofluorescence. pPCs that appear blue in B appear red after staining with DBA. CgA-positive enteroendocrine cells remain green. Nuclei are stained dark blue with bisbenzimide. (D and E) Transmission EM of a region of focal neoplasia in the stomach of a 14-month-old Atp4b-SV40 TAg mouse. D presents a high-power view of four cells with a range of morphological features. Cell 1 is representative of a normal pPC on the basis of the presence of multiple large mitochondria (M), tubulovesicular structures (tv), and long microvilli (L). Cell 4 is representative of a normal enteroendocrine cell on the basis of the presence of many electron-dense secretory granules (eg). Cells 2 and 3 have a mixture of large mitochondria, tubular vesicular structures, and electron-dense secretory granules, indicating transdifferentiation from a pPC toward a NE phenotype. E shows a lower-magnification view at the border of a focus of neoplasia. Cells with the pPC phenotype located in the upper left lack electron-dense secretory granules; enteroendocrine cells at the bottom right have abundant granules. The remaining cells have a variable number of secretory granules, indicating that they are acquiring NE features. (Bars: 20 μm in A–C; original magnifications: ×15,600 in D, ×3,200 in E.)

To further rule out the possibility that nucleotides –1035 to +24 of Atp4b may be active (or become active) in non-PC types, we generated two pedigrees of FVB/N transgenic mice that contained these regulatory elements linked to Cre recombinase. Recombination was scored in bitransgenic mice possessing the Cre-containing transgene and the R26R Cre reporter [R26R contains a floxed “stop sequence” located directly in front of an Escherichia coli lacZ (β-galactosidase) ORF deposited in the otherwise ubiquitously expressed Rosa26 locus (33). Expression of Cre induces lacZ expression through excision of the stop sequence].

Analysis of multiple tissues harvested from Atp4b-Cre, R26R bitransgenic mice confirmed that E. coli β-galactosidase expression [detected with 5-bromo-4-chloro-3-indolyl β-d-galactoside (X-Gal)] was limited to PCs, and not detectable in any CgA-positive or Ddc-positive gastric enteroendocrine cells (e.g., Fig. 3 B and C). (X-Gal staining of tissue sections also demonstrated that Cre was not expressed in adult kidney, small intestinal, and colonic epithelial lineages, or in liver, spleen, and lung; data not shown.)

These results support the scenario that SV40 TAg-expressing pPCs undergo transdifferentiation to a NE phenotype as they become invasive. Subsequent transmission EM studies documented this transdifferentiation (Fig. 3 D and E). Normal pPCs are characterized by multiple large mitochondria and rudimentary canaliculi within their cytoplasm, whereas enteroendocrine cells have distinct, electron-dense cytoplasmic secretory granules. Most cells in areas of invasive neoplasia (iGCs) contained both of these features. The relative proportion of expressed pPC versus endocrine features varied between cells: i.e., within an area of focal neoplasia, a gradient of mixed-feature cells was evident. This spectrum in aggregate captures the transdifferentiation events that pPCs undergo to a NE-like cell.

Mechanistic Considerations. We sought to identify factors that may be potential regulators/mediators of this transdifferentiation. As noted above, transcripts recognized by 399 genes on the mouse U74Av2 GeneChips were enriched in the iGC/mGC populations compared with hyperplastic pPCs and all normal gastric epithelial cell types. Gene Ontology (GO) terms were used to extract the 32 genes that encode transcription factors and DNA-binding proteins from the dataset (Table 7, which is published as supporting information on the PNAS web site).

Some of the extracted transcription factors are expressed in pancreatic endocrine cells. For example, neurogenic differentiation 1 (NeuroD1), paired box gene 6 (Pax6), and Nkx2.2 are key regulators of insulin gene expression (34). Studies of mice deficient in NeuroD1 and Pax6 have established that these factors are necessary for development of alpha, beta, and delta cells in pancreatic islets (35, 36). NeuroD1, which interacts with the promoter region of Nkx2.2 (37), is also expressed in differentiated small intestinal enteroendocrine cells (38).

In addition to transcription factors known to effect endocrine cell differentiation, other factors were identified that normally maintain neuronal precursors in a relatively undifferentiated state. SRY-box 2 (Sox2), a member of the SoxB1 family, maintains neural progenitor characteristics in the CNS: inhibition of Sox2 expression in these progenitors causes them to exit the cell cycle and differentiate (39). Another transcription factor in the list, Hes-related with YRPW motif 1 (Hey1), also maintains the neural precursor phenotype in brain: it negatively regulates basic helix–loop–helix (bHLH) transcription factors, such as mouse achaete scute homolog 1 (mAsh1) and mouse atonal homolog 3 (Math3) that induce neuronal differentiation (40). (The bHLH transcription factor mAsh1 is essential for differentiation of NE cell lineages; mAsh1^–/– mice die at birth and lack NE cells in their lungs and thyroid (41–43)). Thus, the induction of Sox2 and Hey1 as pPCs transdifferentiate to NE-like iGCs may be important in maintaining their relatively undifferentiated/invasive growth properties.

Both p53 and Rb are inactivated in up to 90% of small cell lung cancers (SCLCs; ref. 44). As noted in the Introduction, conditional inactivation of pRB and p53 by adenovirus-mediated delivery of Cre to the pulmonary epithelium of mice results in metastatic small cell type lung NECs. This is not an effect of adenoviral infection per se: when this approach was used to engineer over-expression of K-ras, an oncogene associated with lung cancer, the resulting tumors had the histopathologic features of adenomas (papillary subtype) and adenocarcinomas, rather than SCLCs (45). Because SV40 TAg is known to inactivate p53 and pRB, an obvious question is whether NE features are commonly expressed in other epithelial neoplasms produced in other mouse models by expression of this viral oncoprotein [e.g., stomach (46), prostate (8, 47, 48), thyroid (49), mammary gland (50), lung (51), and retina (52)]. While tumor cells in some of these mouse models have morphologic features of NECs (i.e., poorly differentiated with high nuclear-to-cytoplasmic ratios, clustering together to form focal (or multifocal) solid (nonglandular) lesions (e.g., ref. 49), the results of global gene expression profiling of SV40 TAg-expressing tumors originating in cells that are not neurons or members of a NE lineage have not been deposited in publicly accessible databases. If these datasets are forthcoming, the 399-element filter described above may be useful for cluster analyses to determine whether SV40 TAg commonly produces carcinomas with the molecular properties of NECs.

NE Features in Human Gastric Cancers. If acquisition of NE markers is more common than generally thought in gastric epithelial carcinomas, it could have important implications for diagnosis and/or treatment. Currently, the incidence, classification, and prognosis of NECs are far better described in the human lung than in the stomach. Older studies have concluded that gastric carcinoids represent only 1% of stomach cancers, and 3% of all carcinoids, whereas a more recent analysis concluded that gastric carcinoids represent up to 10–30% of all carcinoids (15). The paucity of information about the incidence of NE features in gastric (or intestinal) adenocarcinomas may reflect the absence of a robust set of molecular biomarkers for systematic analysis.

With this in mind, the signature of 399 transcripts enriched in mouse mGC and iGC populations was used to probe human gastric cancer cDNA microarray datasets (http://genome-www5.stanford.edu) originally published by Leung et al. (53). These datasets were generated from 90 primary gastric adenocarcinomas with intestinal, diffuse, mixed, and indeterminate phenotypes, plus 14 lymph node metastases from 14 of the 90 primary cancers, and were accompanied by datasets from 22 nonneoplastic gastric mucosal samples. The 399 transcripts were matched to 341 cDNAs on the microarray (Table 8, which is published as supporting information on the PNAS web site). Hierarchical clustering with dchip and this 341-member panel disclosed distinctive molecular signatures in a subset of gastric carcinomas (Fig. 6, which is published as supporting information on the PNAS web site). The GCa-1 cluster was enriched for genes associated with neural/endocrine cells: e.g., SOX2, DDC (dopa decarboxylase), GAD1 (glutamic acid decarboxylase), SNAP25 (synaptosomal associated protein), SYT1 (synaptotagmin 1), ENO2 (neuronal enolase), and RTN1 (reticulon 1). Tissues in the GCa-2 category were enriched in genes involved in extracellular matrix remodeling [e.g., LOX (lysyl oxidase) and multiple collagens], paralleling the molecular signature of the extracellular matrix-enriched lung cancers shown in Fig. 5B).

Using clinical data provided by Leung and coworkers, (http://genome-www.stanford.edu/Gastric_Cancer), we performed three individual Kaplan–Meier analyses comparing each cluster (GCa-1, GCa-2, or GCa-3) to all other gastric carcinomas in the dataset. No statistically significant differences (i.e., P < 0.05) were observed in patient survival. In addition, our hierarchical clustering of the microarray datasets without the 341-gene filter failed to segregate gastric cancers on the basis of their histopathology (diffuse, intestinal, mixed, or indeterminate; data not shown). Intriguingly, while no correlation was observed between tumor stage and the clustering of a sample into the GCa-1 or GCa-2 categories, all samples in the GCa-3 category were categorized in the American Joint Committee on Cancer (AJCC) classification scheme as IIIA or IIIB (invasive gastric cancers locally metastatic to regional lymph nodes). This cluster represented 16% (7/43) of the stage IIIA and IIIB samples. Genes in this cluster encode membrane proteins involved in cytoskeletal reorganization, cell motility, adhesion, and antiapoptotic functions [e.g., ena/VASP-like protein (EVL; ref. 54), Wiskott–Aldrich syndrome protein interacting protein (WASPIP or WIP; ref. 55), cytoplasmic FMR1 interacting protein 2 (CYFIP2; ref. 56), and the tetraspanin CD53 (57)]. These functions may support an aggressive, invasive phenotype in these tumors.

In summary, our cluster analysis has established that the NE molecular signature obtained from a comparison of iGCs and mGCs versus pPCs and normal epithelium is directly applicable to human lung NECs, and it may prove useful for further subcategorization of human gastric cancers.

Prospectus. Our finding that an epithelial lineage progenitor, which normally gives rise to a nonneural nonendocrine cell type, can transdifferentiate to metastatic cells with a NE phenotype has implications for the molecular pathogenesis of neuroendocrine cancers in various endoderm-derived epithelia. The list of genes whose expression changed during transdifferentiation (e.g., Sox2, Hey1, NeuroD1) provides a starting point for designing hypothesis-based gain-of-function or loss-of-function genetic tests of their contributions to this process. The Atp4b transcriptional regulatory elements identified as active in pPCs, as well as the Atp4b-Cre mice described in this report, should facilitate such studies.