This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Hu, M. Articles by Shivdasani, R. A. Search for Related Content PubMed PubMed Citation Articles by Hu, M. Articles by Shivdasani, R. A.

Overlapping Gene Expression in Fetal Mouse Intestine Development and Human Colorectal Cancer

¹ Dana-Farber Cancer Institute and ² Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, Boston, Massachusetts

Requests for reprints: Ramesh A. Shivdasani, Dana-Farber Cancer Institute, 44 Binney Street, Boston, MA 02115. Phone: 617-632-5746; Fax: 617-632-4471; E-mail: ramesh_shivdasani{at}dfci.harvard.edu.

	Abstract

Top Abstract Introduction Materials and Methods Results Discussion References

 
Pathways relevant to cancer are well known to overlap with fetal development, as reflected in reactivation of embryonic genes in tumors. However, molecular evidence for this notion has gathered in piecemeal fashion, and systematic approaches have rarely been applied to gauge the extent and global characteristics of the overlap in gene expression between developing tissues and cancer. The fraction of genes that is expressed aberrantly in a given cancer and also developmental in primary function is unknown, and the tissue specificity of recapitulated gene expression remains unexplored. We developed a statistical method to relate expression profiles from human colon cancer and diverse nonintestinal tumors to transcripts that decline in expression with epithelial differentiation in the fetal mouse gut. For genes that are overexpressed in colon cancer, we computed 8% to 19% likelihood that they were expressed transiently during epithelial morphogenesis in intestine development. Among genes dysregulated in other tumors, the corresponding likelihood fell between 1% and 6%. Similarly, low probabilities were obtained when we compared genes not overexpressed in colon cancer with transcriptional profiles in intestine organogenesis. Genes that increase after fetal gut epithelial differentiation were not differentially represented between cancerous and normal colon. Our findings systematically characterize the global extent and tissue specificity of developmental expression programs in colorectal cancer and illustrate the use of such an approach to identify candidate biomarkers and therapeutic targets.

	Introduction

Top Abstract Introduction Materials and Methods Results Discussion References

 
Molecular mechanisms of embryonic development are recognized to correlate with those in cancer, and a growing body of evidence highlights various signaling, transcriptional, and metabolic pathways that are shared between organogenesis and malignant tumors (1–3). Additionally, important properties of tumors including tissue invasion, viability at distant sites, and drug resistance, correlate strongly with the degree of histologic differentiation in resected specimens. Such considerations have popularized the idea that tumor cells represent reversion to a primitive state, although definition of such states is imprecise in both concept and molecular characterization. The supporting evidence is largely anecdotal, and systematic approaches have rarely been applied to gauge the extent and global characteristics of the overlap in gene expression between developing tissues and cancer. Questions such as what fraction of the genes expressed aberrantly in a given cancer reflect reactivation of a developmental program, or whether recapitulated gene expression is characteristic to the affected tissue, remain unanswered. Besides their relevance to tumor biology, these questions have practical implications. First, oncofetal markers can serve as useful tools in cancer diagnosis and in monitoring response to therapy (4, 5). Second, cancer treatments often are limited by severe toxicity that reflects expression of the drug target in unaffected tissues. If tumors depend on the expression of some proteins that are absent in adult tissues, then as therapeutic targets, such proteins might confer a wide therapeutic window.

Computational analysis of suitable mRNA expression data sets could potentially yield systematic approximations of the true underlying statistics on overlapping gene expression between developing and cancerous tissues. One such study (6) interrogated microarray-based gene expression profiles across human medulloblastomas and mouse cerebellar development. The authors used a pattern classification (singular value decomposition) approach to reveal that transcripts increased in expression in human medulloblastoma significantly reflect early mouse cerebellar development, whereas transcripts reduced in this disease correspond to a later, complementary program of gene expression.

The intestinal mucosa is organized into crypts, which house replicating progenitor cells, and villous projections lined by post-mitotic epithelial cells with differentiated morphology and functions (7, 8). In a significant developmental transition, the gut endoderm first acquires villous character between 13 and 15 days in mouse gestation. We have assembled over 65,000 serial analysis of gene expression (SAGE; ref. 9) tags (representing over 10,000 unique mRNAs) from the mouse small intestine at 12, 13, and 15 days post-coitus. The complete data set, found at http://genome.dfci.harvard.edu/GutSAGE, shows nearly twice as many significant changes in gene expression between 13 and 15 days post-coitus than between 12 and 13 days post-coitus, revealing notable modulation of gene activity in conjunction with epithelial histogenesis. Genes with reduced expression following the villus transition may serve biological functions that are confined to the period of organogenesis. We recently showed that many such transcripts are absent from the adult gut, and for a small group of developmentally repressed genes, we presented experimental evidence that up to one fifth of the transcripts may be reexpressed in human colorectal tumors (10). Here, we extend these limited findings by applying a computational strategy toward a larger number of transcripts.

We mapped all existing human colon cancer gene expression profiles and a diverse set of expression profiles from nonintestinal tumors onto our data set of mouse intestine development. We applied a conditional probability method to approximate the degree to which human colon cancers show tissue-specific recapitulation of developmentally regulated genes. For genes that are overexpressed in colon cancer, our statistical analysis yields 8% to 19% likelihood that they were expressed transiently during gut epithelial morphogenesis in development. Among genes overexpressed in other malignancies, the corresponding probability falls between 1% and 6%. Our findings systematically estimate the extent to which cancer gene dysregulation coincides with the developmental expression program and show the tissue specificity of this process.

	Materials and Methods

Top Abstract Introduction Materials and Methods Results Discussion References

 
Treatment of human colon cancer data sets. To capture a comprehensive collection of matched tumor and normal tissue expression data, we queried Oncomine (http://141.214.6.50/oncomine/main/index.jsp) for all results on matched, well-characterized colon cancer compared with normal tissue (11). This search yielded two adequate microarray and two SAGE expression data sets. Starting with SAGE libraries (12) NC_1 versus Tu_98 (14,300 unique SAGE tags) and NC_2 versus Tu_102 (7,400 tags), we first mapped every SAGE tag against all nonredundant human Unigene entries (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=unigene) and isolated transcripts (CAN) showing >2-fold increase in tumor compared with normal tissue (P < 0.005) or a control set (NC) for which tumor/normal tissue expression ratios fall between 0.8 and 1.2 (P > 0.6). A second filter applied HomoloGene (http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?db=homologene) criteria and retained only those human genes with a reliable murine homologue. Finally, consideration was limited to transcripts with at least one tag entry in our fetal intestine SAGE database (http://genome.dfci.harvard.edu/GutSAGE), and genes were removed arbitrarily from the NC set with the dual purpose of bringing the two final sets (CAN_INT and NC_INT) to a comparable size and to introduce additional randomness. For both the SAGE and following microarray data sets, Table 1 lists transcript numbers after application of the filters used to obtain CAN_INT.

The second microarray study (15) probed adenocarcinomas and normal tissue from three patients for expression corresponding to 14,000 Genbank accession nos. (Affymetrix Hu6800 and Hu35KsubA gene chips). Unlike the above example, the data are available on a raw expression scale, without P for comparisons. We therefore set a more stringent variable and isolated transcripts showing >2.5-fold mean expression in tumors over the matched normal tissue across all sample pairs; the resulting gene set is denoted mArray_GCM_colon. Filters similar to the above treatment were used to arrive at the corresponding sets CAN_HOMOL and CAN_INT. To generate the control, noncancer data set NC_INT, we scanned the paired samples for transcripts with similar mean expression in tumor and normal tissues (ratio between 0.9 and 1.1) and randomly selected 200 of 1,300 such transcripts to bring the noncancer and cancer sets to comparable size.

Treatment of data sets from human nonintestinal cancers. The principle of applying successive filters is similar to that described above for colon cancer expression data sets. We first extracted all transcripts that are overexpressed in tumors compared with matched normal tissue. The actual variables differ slightly for each data set, as described below. Where available (liver and plasma cell), we accepted the variables for differential expression from the original reports; for the remaining three data sets (prostate and breast), we chose expression cutoffs and Ps that are common across the microarray literature and also returned a data set sufficiently large for statistical analysis. Owing to the heterogeneity of these large data sets and our demand of differential expression across >15 or >10 profiled tumors, we chose 1.5- and 1.75-fold cutoffs for the prostate and breast data, respectively. The numerical readouts in microarray analysis are well known to underestimate, often significantly, the differential expression between cancer and normal tissues (15, 16), and higher arbitrary cutoffs yielded too few genes for further analysis; to reduce chance events, we avoided demanding altered gene expression in fewer tumors.

A second filter eliminated genes that lack a reliable mouse homologue, as specified in LocusLink and annotated in HomoloGene. The final filter retained only genes with a reliable entry in our mouse fetal intestine SAGE database. The resulting groups are denoted as nonintestinal cancer (NIC) gene sets, and Table 2 lists the sizes of the intermediate sets resulting from application of each filter.

Liver tumors. Data were extracted from a study that characterized mRNA expression using 17,400-gene chips on 82 hepatocellular carcinoma and 74 nontumorous liver samples (18). For the first filter, we took the reported list of 1,640 most differentially expressed transcripts between tumor and normal samples (Bonferroni corrected P < 0.01; see Supplementary Section of the original report; ref. 18) and considered genes with a mean tumor/normal expression ratio of >2.

Prostate cancer. One study (Prostate1) compared gene expression from 52 adenocarcinomas and 50 normal samples using microarrays with probes for 12,600 genes (19). Owing to heterogeneity of samples in the large cohort, we took transcripts with a "Present" call (expressed at detectable levels) in >15 of the 52 tumor samples and mean tumor/normal expression ratio of >1.5. For the data (Prostate2) from paired SAGE libraries (http://cgap.nci.nih.gov/SAGE), constructed from a microdissected tumor and adjacent normal epithelium, we required a ratio of tumor/normal SAGE tags of >1.5 (P < 0.01).

Plasma cell tumors. The data derive from a study that used Affymetrix microarrays to compare gene expression profiles from nine patients with multiple myeloma and eight myeloma cell lines with those of nonmalignant plasma cells (20). In our first filter, we took the list of 250 genes that the authors reported as being significantly overexpressed in malignant plasma cells (>2-fold, P < 0.05).

	Results

Top Abstract Introduction Materials and Methods Results Discussion References

 
Statement of the problem and data set resources. The main question and approach are depicted in Fig. 1. We define the rate of developmental gene recapitulation in cancer as the likelihood that a given gene that is activated in colon cancer was expressed at the highest level during gut epithelial morphogenesis. This is cast as the conditional probability Pr(DEV|CAN), or the probability of a given gene belonging to the developmental program, conditioned on its overexpression in cancer. The unknown variable, the true likelihood of recapitulation, is approximated by empirically determining the underlying Pr(DEV|CAN) (the ratio of the size of the subset x to that of the set CAN in Fig. 1). This value is computed for multiple sets of genes (CAN) that are overexpressed in colon cancer, derived from different experiments and profiling methods. We can thus empirically arrive at an approximation for Pr(DEV|CAN) on each set, resulting in a distribution of likelihood values (right curve in Fig. 1). A similar approximation of developmental gene recapitulation in normal (noncancer, NC) tissues gives the distribution of the complementary probabilities Pr(DEV|NC) (Fig. 1, left curve). If the cancer and noncancer gene sets are of roughly equal size and the difference between the two empirical distributions (Fig. 1) is statistically significant, one would conclude that developmental gene expression programs are recapitulated at a higher rate in cancer than in the normal tissue. Note that our definitions do not say that more cancer genes are of developmental origin than non cancer genes, nor that more genes overexpressed in cancer are of developmental significance than not.

View larger version (20K): [in this window] [in a new window] [Download PPT slide]   Figure 1. Diagrammatic representation of recapitulation of developmental genes in cancer and of the principal question posed in this study. The likelihood that a gene overexpressed in colon cancer was highly expressed during development of the gut is cast as the conditional probability Pr(DEV|CAN). For equal sizes (n) of cancer (CAN) and noncancer (NC) gene sets, we empirically approximated the underlying Pr(DEV|CAN) (representing the ratio x/n) on multiple cancer gene sets and Pr(DEV|NC) in the normal tissue counterpart. This analysis returns two independent distributions of likelihood values. If the difference between the two empirical distributions is statistically significant, it would indicate that the rate of recapitulation of a developmental program is greater in cancer than in the corresponding noncancerous tissue.

View larger version (28K): [in this window] [in a new window] [Download PPT slide]   Figure 2. Reactivation of developmentally down-regulated genes in human colon cancer. A, cluster distribution of a representative fraction of 254 transcripts that show higher (P < 0.01) expression in the developing mouse intestine on embryonic days E12 and/or E13 relative to E15 and constitute the set denoted DEVINT. These mRNAs were selected to investigate expression in human tumors. >, <, and (stable) are the relationships in relative frequency of SAGE tags between developmental stages. Clusters were generated using a k-model-based algorithm to separate groups according to their temporal variation in expression (35). B, Venn diagram representation of the conditional probability distributions given by Eqs. A and B for the three categories DEVINT, CANINT, and NCINT, illustrating derivation of the two pertinent ratios, x/n1 and y/n2. C, table of approximations for the conditionals on the eight CANINT (two SAGE and six microarray data sets) and NCINT (normal colon expression) groups. In each category, column 1 lists approximations based on strict statistical criteria (P < 0.01, 254 transcripts) for developmental down-regulation and column 2 lists approximations based on relaxed criteria (P < 0.025, 451 genes) for defining DEVINT. Despite the adversity that follows from increased noise when the statistical variables are modified, the results change little, implying that the different distributions on Eqs. A and B represent true underlying biological differences.

To analyze these expression data sets, we adopted the following conditional statistics formulations:

(A)

(B)

Eq. A is the conditional probability that any gene x, given that it is overexpressed in colon cancer, is also repressed after E12 or E13 in mouse gut development. The second conditional is the distribution on the probability of any gene x, given that it is not overexpressed in intestinal cancer, being repressed in gut development. It provides the control necessary to accept or reject the hypothesis that the two conditionals (Eqs. A and B) represent different underlying distributions (Fig. 1).

We computed the conditional probabilities in Eqs. A and B using the eight pairs of CAN_INT and NC_INT gene expression data sets (Table 1) and determined that the distribution on the likelihood of any gene being developmentally regulated in the intestine, given that it is overexpressed in colon cancer, falls between 7.7% and 18.8% (x/n₁ in Fig. 2B and C). However, if a gene is not overexpressed in colon tumors, its likelihood of being repressed during gut development (y/n₂ in Fig. 2B and C) is at most 6.7% and typically much lower. An unpaired, nonparametric (Mann-Whitney) t test on these data reveals the difference in distribution between the two conditional probabilities to be statistically significant (P < 0.0006). Indeed, the estimated probability of a given gene being silenced during intestine development is, on average, three to four times greater if it is overexpressed in colon cancer than if it is not.

To reduce spurious correlations resulting from data biases, we perturbed our calculations by using a larger developmental gene collection, DEV_INT set 2, with relaxed statistical criteria. This expanded set contains 451 transcripts with lower intestine expression at E15 compared with earlier stages (P < 0.025) and the potential addition of noise creates adversity for the hypothesis under investigation. When DEV_INT set 2 was used to approximate the conditional probabilities given by Eqs. A and B, absolute percentage values differed slightly but the general trends and statistical significance of the comparisons remained similar (Fig. 2C). This similarity in the face of adversity implies that the different distributions on Eqs. A and B represent true underlying biological differences. The results thus suggest a measurable bias for developmentally repressed genes to be reactivated in tumors of the same origin and they support a conclusion we recently reported on a smaller, experimental scale (10). Notably, gene expression in human colorectal cancer mimics only a portion of the transcriptional program that is active during intestine organogenesis.

Assessment of tissue specificity. The apparent reactivation of embryonic gene expression programs is not necessarily specific to the tissue of tumor origin. Gene overexpression in tumors could represent two possibilities. In one case, genes silenced during gut development could reactivate selectively in intestinal tumors, reflecting reversal of tissue-specific epigenetic regulatory mechanisms. Alternatively, reactivation of developmental genes in cancer might represent a non–tissue-specific process, where genes repressed during gut development are appropriated for malignant behaviors in diverse cell types. These scenarios differ both in their implications and in probable underlying mechanisms. Tissue-specific gene reactivation implies that oncogenesis represents some degree of developmental reversal, whereas non–tissue-specific reactivation points simply to shared molecular features between development and cancer. Of note, these implications apply only to the consideration of groups of genes; overexpression of any single gene in fetal and tumor tissues may reflect either coincidence or a cellular process that is common to both processes.

To assess the tissue specificity of developmental gene reexpression, we applied a third statistical formulation and evaluated gene expression data sets from NIC:

(C)

This conditional is the distribution on the probability that a gene x, given that it is activated in some nonintestinal cancer, is also repressed after E12 or E13 in mouse gut development. It assesses the specificity with which gut developmental gene expression is recapitulated in colon cancer, and we proposed that the conditionals given by Eqs. A and C form different distributions. The conditional probabilities were computed by taking the ratio of the number of genes found in the intersection of the appropriate gene sets, as shown in Fig. 3. To compute values for breast cancer, for example, we take the ratio y/n (Fig. 3A), where y is the number of genes in the intersection of the DEV_INT and breast cancer gene sets, and n is the number of genes in the breast NIC set with corresponding entries in fetal gut SAGE libraries (Table 2).

View larger version (24K): [in this window] [in a new window] [Download PPT slide]   Figure 3. Tissue specificity of developmental gene reactivation in cancer. A, Venn diagram representation of the conditional probability distributions given by Eq. C. B, table of approximations for the conditional probability (Eq. C) estimated on gene expression data from five different types of human nonintestinal cancer (NIC). Columns 1 and 2 list, respectively, the approximations, n/y, based on strict (P < 0.01, DEVINT set of 254 transcripts) and relaxed (P < 0.025, DEVINT set of 451 transcripts) statistical variables for developmental down-regulation.

Tumor expression of differentiation genes. Next, we examined the degree to which colon cancers differ from normal epithelium in expression of transcripts with the opposite pattern (i.e., increased levels after the developmental villus transition); such genes represent independent controls for the results we have obtained thus far. To define a set of differentiation genes that appear late in mammalian intestine development, we considered all transcripts with significantly higher (P < 0.01 and more than eight tags) SAGE representation at E15 than at E12 and/or E13 (Fig. 4A). One hundred seventy-seven transcripts satisfied this requirement and constitute the set we denote as DIF_INT. To arrive at statistical estimates of the rate at which such late-rising genes are overexpressed in colon cancer, we formulated the following conditional probabilities:

(D)

(E)

Eq. D is the conditional probability that a gene x, given that it is overexpressed in colon cancer, was activated late in fetal gut development. The second conditional is the distribution on the probability of a gene x, given that it is not overexpressed in intestinal cancer, displaying such a developmental profile.

View larger version (29K): [in this window] [in a new window] [Download PPT slide]   Figure 4. Expression of differentiation genes in human colon cancer. A, cluster representation, generated as described for Fig. 2A, of selected transcripts that show higher (P < 0.01) expression in the developing mouse gut at E15 than before; the full set of 177 genes, derived from SAGE profiling, is designated DIFINT. B, table of approximations on the expression of differentiation genes in cancer, given by Eqs. D and E and derived from the eight CANINT, eight NCINT, and DIFINT gene sets.

Oncofetal markers. Our analysis of recapitulation of developmental genes in cancer is limited by at least two factors: the size of currently available gene expression data sets in mouse gut development and human colon cancer and the difficulties in assigning homology between pairs of human and mouse genes. Nevertheless, the statistical basis of our analysis predicts that future studies, which might draw on more comprehensive transcriptional profiles, would yield similar trends. Meanwhile, our current results highlight some developmentally regulated genes that are reactivated in neoplasia (Table 3). Genes with these expression characteristics point to cellular functions that may be common to cancer and developing tissues. The overlap in developmental and malignant gene expression encompasses several factors associated with cellular stress response (heat shock proteins 90 and 1ß, stress-induced phosphoprotein 1, and chaperonin-containing complexes), protein synthesis (ribosomal proteins L39, L3, and S26 and eukaryotic translation elongation factor 1B2), and cell cycle regulation (CDK4 and the Cdc47 homologue). Insulin-like growth factor II (IGF-II), which is known to enhance tumor growth (21), is also highly expressed in the fetal and cancerous gut. Other genes with similar expression include the paracrine growth factor midkine and an intracellular regulator of multiple Ran protein functions, RanBP1. These findings likely reflect the diversity of processes that mark both embryonic development and malignancy and reveal malignant reactivation of developmental genes with a range of cellular functions.

	Discussion

Top Abstract Introduction Materials and Methods Results Discussion References

The idea that cancers share properties with developing embryos has been discussed by many authors (22, 23) and first gained currency following the embryologic studies of Waddington and Needham in the 1930s, when malignant behaviors were considered in the light of tissue organizers, morphogenetic fields, and cellular hierarchies (24, 25). Recent progress in linking epithelial tumors in general, and colon cancer in particular, to cell signaling pathways that regulate gut development and homeostasis extend these ideas significantly (2). Although many other conceptual and experimental advances have highlighted commonalities between development and cancer, the parallels have for the most part been explored at the level of single or small groups of genes and pathways. Here, we report a systematic and quantitative approach to delineate the degree of overlapping gene expression in colon cancer and development of the mammalian intestine.

We applied computational and statistical strategies to relate expression profiles from human colon cancer and the developing mouse gut. The process applied to generate the target data sets ensures that each human gene considered can be mapped with confidence to a probable murine homologue. Our results reveal that 8% to 19% of genes overexpressed in intestinal tumors had previously shown their highest expression concomitant with fetal villus morphogenesis. The frequency at which such genes are expressed in various nonintestinal tumors or the likelihood of genes not overexpressed in colon cancer having an origin in the gut developmental expression program are both much lower. For each of these trends, the results were similar when we analyzed developmental gene sets defined by different statistical criteria, which suggests that the correlations reflect the true underlying biology. Our analysis thus yields a systematic estimate of the global extent to which developmental gene expression may be recapitulated in a solid tumor and suggests that this process displays a high degree of tissue specificity.

Transcripts expressed in human cancers from various sites have been profiled much more extensively than stages in development of individual organs. This difference limits the degree to which our methods may immediately be extended to explore overlaps in gene expression between cancers and development of other tissues. However, our observations do make a testable prediction: strong correlation between activation of groups of genes in particular tumors and the prior silencing of those genes during fetal patterning of the same tissue. Notably, our definitions do not lead to the claim that there are more cancer transcripts than noncancer transcripts of developmental origin; many genes reduced late in development are expressed to varying degrees in the normal adult tissue. Nor do we imply that most transcripts overexpressed in cancer belong to the developmental program. There are undoubtedly many modes of aberrant gene activation in tumors, of which developmental gene reactivation is only one.

Tissue differentiation during development is largely under epigenetic control. It follows that there may be two classes of epigenetic modification pertinent to this discussion. One category is represented in genes that are never accessible to the transcription machinery in a particular cell type and accordingly never expressed therein. The second type of modification occurs in genes after their transient embryonic expression and such changes may be especially amenable to reversal in malignancy. Considerable experimental evidence indicates that epigenetic alteration, including DNA methylation, is important in tumor initiation and progression (26, 27); investigation has traditionally emphasized inactivation of tumor suppressor genes, usually by promoter hypermethylation (28). In contrast, reactivation of developmentally regulated genes may result from the aberrant hypomethylation observed in many tumor types, including colon cancer (29), and likely contributes to malignant behaviors.

Historically, oncofetal proteins have been identified individually. Our computational analysis of malignant and developmental expression profiles represents a strategy to identify new candidates and reveals tumor reactivation of fetal genes with a wide range of cellular functions. A few gene classes are prominent, including factors associated with cellular stress response, cell cycle regulation, and protein synthesis. Although increased expression of ribosomal protein genes is recognized as a feature of human cancers (30, 31), the underlying significance is uncertain. Conversely, monoallelic loss-of-function mutations in gene loci for ribosomal proteins were reported to a high and unexpected degree in tumors in zebrafish (32). These observations support the idea that dysregulated ribosome biogenesis, whether by gain or loss of gene functions, may be tumorigenic (33). We find that developmental transitions are accompanied by significant modulation in expression of selected genes involved in assembly of the 40S and 60S ribosomal subunits (http://genome.dfci.harvard.edu/GutSAGE). Thus, certain aspects of the ribosome or, alternatively, nonribosomal functions of the genes in question, may be common to developmental and malignant cellular processes. IGF-II, which is known to enhance tumor growth and suppress apoptosis (21), is also highly expressed in the fetal and cancerous gut. Loss of genomic imprinting, with resulting abnormal activation of the normally silent maternal Igf2 allele in all cells, is strongly associated with a risk of developing colon cancer (34). Besides their value in cancer diagnosis and surveillance, oncofetal factors could also be good therapeutic targets. If they are expressed on the tumor cell surface or make a material contribution to the malignant phenotype, then specific antibodies or drugs may be developed in anticipation of limited toxicity as the target is absent from the normal adult tissue.

	Acknowledgments

 
Grant support: Robert Black Charitable Foundation and NIH grant R01DK61139.

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 Maina Lepourcelet and Li Cai for their contributions toward early phases of this work and members of the Shivdasani laboratory for helpful discussions.

	Footnotes

 
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).

Received 3/ 1/05. Revised 7/ 7/05. Accepted 7/22/05.

	References

Top Abstract Introduction Materials and Methods Results Discussion References

 

1. Lum L, Beachy PA. The Hedgehog response network: sensors, switches, and routers. Science 2004;304:1755–9.[Abstract/Free Full Text]
2. Sancho E, Batlle E, Clevers H. Signaling pathways in intestinal development and cancer. Annu Rev Cell Dev Biol 2004;20:695–723.[CrossRef][Medline]
3. Vogelstein B, Kinzler KW. Cancer genes and the pathways they control. Nat Med 2004;10:789–99.[CrossRef][Medline]
4. Garrett PE, Kurtz SR. Clinical utility of oncofetal proteins and hormones as tumor markers. Med Clin North Am 1986;70:1295–306.[Medline]
5. Duffy MJ. Evidence for the clinical use of tumour markers. Ann Clin Biochem 2004;41:370–7.[CrossRef][Medline]
6. Kho AT, Zhao Q, Cai Z, et al. Conserved mechanisms across development and tumorigenesis revealed by a mouse development perspective of human cancers. Genes Dev 2004;18:629–40.[Abstract/Free Full Text]
7. Stappenbeck TS, Wong MH, Saam JR, Mysorekar IU, Gordon JI. Notes from some crypt watchers: regulation of renewal in the mouse intestinal epithelium. Curr Opin Cell Biol 1998;10:702–9.[CrossRef][Medline]
8. Potten CS, Booth C, Pritchard DM. The intestinal epithelial stem cell: the mucosal governor. Int J Exp Pathol 1997;78:219–43.[CrossRef][Medline]
9. Velculescu VE, Zhang L, Vogelstein B, Kinzler KW. Serial analysis of gene expression. Science 1995;270:484–7.[Abstract/Free Full Text]
10. Lepourcelet M, Tou L, Cai L, et al. Insights into developmental mechanisms and cancers in the mammalian intestine derived from serial analysis of gene expression and study of the hepatoma-derived growth factor (HDGF). Development 2005;132:415–27.[Abstract/Free Full Text]
11. Rhodes DR, Yu J, Shanker K, et al. ONCOMINE: a cancer microarray database and integrated data-mining platform. Neoplasia 2004;6:1–6.[Medline]
12. Zhang L, Zhou W, Velculescu VE, et al. Gene expression profiles in normal and cancer cells. Science 1997;276:1268–72.[Abstract/Free Full Text]
13. Alon U, Barkai N, Notterman DA, et al. Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc Natl Acad Sci U S A 1999;96:6745–50.[Abstract/Free Full Text]
14. Notterman DA, Alon U, Sierk AJ, Levine AJ. Transcriptional gene expression profiles of colorectal adenoma, adenocarcinoma, and normal tissue examined by oligonucleotide arrays. Cancer Res 2001;61:3124–30.[Abstract/Free Full Text]
15. Ramaswamy S, Tamayo P, Rifkin R, et al. Multiclass cancer diagnosis using tumor gene expression signatures. Proc Natl Acad Sci U S A 2001;98:15149–54.[Abstract/Free Full Text]
16. Lal A, Lash AE, Altschul SF, et al. A public database for gene expression in human cancers. Cancer Res 1999;59:5403–7.[Abstract/Free Full Text]
17. Ma XJ, Salunga R, Tuggle JT, et al. Gene expression profiles of human breast cancer progression. Proc Natl Acad Sci U S A 2003;100:5974–9.[Abstract/Free Full Text]
18. Chen X, Cheung ST, So S, et al. Gene expression patterns in human liver cancers. Mol Biol Cell 2002;13:1929–39.[Abstract/Free Full Text]
19. Singh D, Febbo PG, Ross K, et al. Gene expression correlates of clinical prostate cancer behavior. Cancer Cell 2002;1:203–9.[CrossRef][Medline]
20. De Vos J, Thykjaer T, Tarte K, et al. Comparison of gene expression profiling between malignant and normal plasma cells with oligonucleotide arrays. Oncogene 2002;21:6848–57.[CrossRef][Medline]
21. Furstenberger G, Senn HJ. Insulin-like growth factors and cancer. Lancet Oncol 2002;3:298–302.[CrossRef][Medline]
22. Uriel J. Fetal characteristics of cancer. 3rd ed. In: Becker FF, editor. Cancer, a comprehensive treatise. New York: Plenum Press; 1975.
23. Rubin H. Cancer as a dynamic developmental disorder. Cancer Res 1985;45:2935–42.[Free Full Text]
24. Needham J. New advances in the chemistry and biology of organized growth. Proc R Soc Lond B Biol Sci 1936;29:1577–626.
25. Waddington CH. Cancer and the theory of organizers. Nature 1935;135:606–8.
26. Gaudet F, Hodgson JG, Eden A, et al. Induction of tumors in mice by genomic hypomethylation. Science 2003;300:489–92.[Abstract/Free Full Text]
27. Mihich E, Jaenisch R. Sixteenth Annual Pezcoller Symposium: stem cells and epigenesis in cancer. Cancer Res 2004;64:8474–7.[Abstract/Free Full Text]
28. Jones PA, Baylin SB. The fundamental role of epigenetic events in cancer. Nat Rev Genet 2002;3:415–28.[Medline]
29. Goelz SE, Vogelstein B, Hamilton SR, Feinberg AP. Hypomethylation of DNA from benign and malignant human colon neoplasms. Science 1985;228:187–90.[Abstract/Free Full Text]
30. Loging WT, Reisman D. Elevated expression of ribosomal protein genes L37, RPP-1, and S2 in the presence of mutant p53. Cancer Epidemiol Biomarkers Prev 1999;8:1011–6.[Abstract/Free Full Text]
31. Kondoh N, Shuda M, Tanaka K, Wakatsuki T, Hada A, Yamamoto M. Enhanced expression of S8, L12, L23a, L27 and L30 ribosomal protein mRNAs in human hepatocellular carcinoma. Anticancer Res 2001;21:2429–33.[Medline]
32. Amsterdam A, Sadler KC, Lai K, et al. Many ribosomal protein genes are cancer genes in zebrafish. PLoS Biol 2004;2:690–8.
33. Ruggero D, Pandolfi PP. Does the ribosome translate cancer? Nat Rev Cancer 2003;3:179–92.[CrossRef][Medline]
34. Cui H, Cruz-Correa M, Giardiello FM, et al. Loss of IGF2 imprinting: a potential marker of colorectal cancer risk. Science 2003;299:1753–5.[Abstract/Free Full Text]
35. Cai L, Huang H, Blackshaw S, Liu JS, Cepko C, Wong WH. Clustering analysis of SAGE data using a Poisson approach. Genome Biol 2004;5:R51.[CrossRef][Medline]

This Article Abstract Full Text (PDF) Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via Google Scholar Google Scholar Articles by Hu, M. Articles by Shivdasani, R. A. Search for Related Content PubMed PubMed Citation Articles by Hu, M. Articles by Shivdasani, R. A.