Cross-Species Comparison of Human and Mouse Intestinal Polyps Reveals Conserved Mechanisms in Adenomatous Polyposis Coli (APC)-Driven Tumorigenesis

Abstract

Expression profiling is a well established tool for the genome-wide analysis of human cancers. However, the high sensitivity of this approach combined with the well known cellular and molecular heterogeneity of cancer often result in extremely complex expression signatures that are difficult to interpret functionally. The majority of sporadic colorectal cancers are triggered by mutations in the adenomatous polyposis coli (APC) tumor suppressor gene, leading to the constitutive activation of the Wnt/β-catenin signaling pathway and formation of adenomas. Despite this common genetic basis, colorectal cancers are very heterogeneous in their degree of differentiation, growth rate, and malignancy potential. Here, we applied a cross-species comparison of expression profiles of intestinal polyps derived from hereditary colorectal cancer patients carrying APC germline mutations and from mice carrying a targeted inactivating mutation in the mouse homologue Apc. This comparative approach resulted in the establishment of a conserved signature of 166 genes that were differentially expressed between adenomas and normal intestinal mucosa in both species. Functional analyses of the conserved genes revealed a general increase in cell proliferation and the activation of the Wnt/β-catenin signaling pathway. Moreover, the conserved signature was able to resolve expression profiles from hereditary polyposis patients carrying APC germline mutations from those with bi-allelic inactivation of the MYH gene, supporting the usefulness of such comparisons to discriminate among patients with distinct genetic defects.

Colorectal cancer (CRC) is one of the major causes of morbidity and mortality among Western societies. Although the vast majority of CRC cases are sporadic, a considerable fraction has been attributed to hereditary and familial factors.1 Hereditary CRC syndromes have served as unique models to elucidate the molecular and cellular mechanisms underlying colorectal tumor initiation and progression to malignancy, as the same genes mutated in the germline of affected individuals are also known to play rate-limiting roles in the majority of the sporadic cases.2 This is certainly the case of the adenomatous polyposis coli (APC) tumor suppressor gene, known to be mutated in the germline of individuals affected by familial adenomatous polyposis (FAP)3,4,5,6 and in the majority of the sporadic CRC cases.7,8,9 In fact, loss of APC function appears to play a rate-limiting and initiating role in the adenoma-carcinoma sequence.10 Among the multiple functional domains characterized in its coding sequence, APC’s ability to bind and down-regulate β-catenin, the main signaling molecule in the canonical Wnt pathway, is generally regarded as its main tumor-suppressing activity.10 Loss of APC function or oncogenic activation of β-catenin, as observed in the vast majority of the sporadic CRC cases, leads to intracellular accumulation of β-catenin and its translocation to the nucleus, where it binds to transcription factors of the T cell factor/lymphoid enhancer-binding factor (TCF/LEF) family and modulates transcription of a broad spectrum of downstream target genes.11,12 The vast majority (70 to 90%) of FAP patients have been shown to carry germline APC mutations.13 More recently, biallelic mutations in the base excision repair gene MYH were found in a subset of polyposis families with attenuated clinical presentation and an autosomal recessive inheritance pattern, often referred to as MAP (MYH-associated polyposis).14

In view of its initiating role in intestinal cancer, several preclinical models carrying germline mutations in the endogenous mouse Apc tumor supressor gene have been generated, and their phenotype has been characterized.15 The predisposition of these mouse models to multiple intestinal adenomas closely resembles the FAP phenotype at the molecular, cellular, and phenotypic level.15 One exception to the latter is represented by the proximal localization and distribution of adenomas along the gastrointestinal (GI) tract of Apc-mutant mouse models when compared with the colorectal clustering of polyps among FAP patients.15

Expression profiling by cDNA and oligonucleotide microarrays represents a powerful tool for genome-wide transcriptional analysis. Several studies in the literature have reported on the comparison of expression profiles from colorectal tumors and normal intestinal mucosa in an attempt to identify differentially expressed genes, predict, whenever feasible, clinical outcome, and elucidate the molecular and cellular mechanisms underlying colorectal tumorigenesis.16 However, the different lists of genes differentially expressed in CRC are often very extensive and only partially overlapping among independent studies, possibly reflecting differences in the methodologies and in cohorts used.16 To pinpoint conserved and functionally relevant genes differentially expressed between normal and malignant tissues, cross-species comparison have been successfully applied by comparing expression signatures of hepatocellular carcinoma, and prostate and lung cancer derived from human patients and mouse models.17,18,19,20

Here, we report the cross-species comparison of expression profiles of intestinal adenomas from FAP patients with established APC germline mutations and from Apc^+/1638N, a mouse model for familial polyposis previously developed in our laboratory and characterized by the development of an average of five tumors in the upper GI tract, together with other extra-intestinal manifestations characteristic of FAP patients, such as epidermal cysts and desmoids.21,22 A total of 166 genes were found to be highly conserved between the two species and are likely to play important roles in the cellular and molecular mechanisms underlying adenoma formation in the gastrointestinal tract. Among these, several Wnt downstream target genes are included, as also expected from the selection of APC-mutant mouse and human adenomas. Notably, the conserved APC signature also made it possible to distinguish FAP tumors from MAP ones in an unsupervised fashion.

Materials and Methods

Patients and Tumor Samples

Colorectal adenomatous polyps were obtained from a total of 13 patients from the Department of Surgery, Heinrich Heine University (Dusseldorf, Germany); the Institute of Medical Genetics, Cardiff University (Cardiff, UK); the Department of Pathology, Leiden University Medical Center (Leiden, The Netherlands); and the Department of Surgery, Erasmus Medical Center (Rotterdam, The Netherlands). Of the 13 polyposis patients, 8 carried truncating APC mutations, whereas 5 were found to carry biallelic MYH mutations. Detailed characterization of the polyposis patients carrying biallelic MYH or monoallelic APC germline mutations and of the corresponding tumor samples used in the present study has been reported elsewhere.23 Normal epithelial mucosa from 3 healthy individuals was also collected (control samples NC1 to NC3). All tissue samples were snap-frozen, embedded in OCT medium, and stored at −80°C. Detailed sample processing procedures were as previously described.24 All of the analyzed adenomatous lesions were matched for histology (low-grade dysplasia) and anatomical location (left-sided colon or rectum). Two to six polyps were analyzed for each individual patient.

Mouse Strains and Material

All wild-type and Apc^+/1638N mice used in this study were inbred C57BL6/J, maintained under SPF conditions and fed ad libitum. Duodenal adenomas and normal mucosa samples were collected from 9-month-old males, briefly washed in PBS, and snap-frozen in OCT medium. Hematoxylin and eosin (H&E) staining of all tissues was performed to determine histology and tumor content.

Laser Capture Microdissection and RNA Isolation

Sample preparation and laser capture microdissection (LCM) were performed as previously described using a PALM MicroBeam microscope system (P.A.L.M. Microlaser Technologies, Bernereid, Germany).23 In short, 10-μm cryosections were mounted on microscope slides with a polyethylene naphthalate membrane and stained by H&E to allow histological identification of the desired normal and tumor cells. Approximately 1000 to 2000 cells, corresponding to 600,000 (human samples)- and 1,200,000 (mouse specimens)-μm² areas, were microdissected.

RNA was isolated from the LCM samples by Mini RNeasy columns (QIAGEN, Valencia, CA), according to the manufacturer’s instructions, including a DNase step on the column. Quality of the isolated RNA was checked with RNA 6000 Pico LabChip kit (Agilent Technologies, Palo Alto, CA).

Expression Profiling and Data Analysis

Human Adenoma Samples

Each RNA sample that passed the quality controls was linearly amplified with two rounds of amplification using the MessageAmp kit (Ambion, Huntingdon, UK), according to the manufacturer’s protocol. Quality and quantity of each amplified RNA sample was again evaluated by Nano Lab-on-Chip (Agilent Technologies). One-μg aliquots of target and reference amplified RNA were labeled with Cy5-dUTP and Cy3-dUTP, respectively (Amersham Biosciences, Amersham, UK) by reverse transcription using the CyScribe First Strand cDNA Labeling kit (Amersham Biosciences). Each labeling reaction was further purified with the CyScribe GFX purification kit (Amersham Biosciences). Subsequently, both labeled cDNAs were hybridized on a human 18K cDNA microarray encompassing 19,200 spots (representing 18,432 independent cDNAs) produced and obtained from The Netherlands Cancer Institute Microarray Facility (Amsterdam, The Netherlands). The cDNAs spotted in this array platform were PCR-amplified from a clone set purchased from Research Genetics (Huntsville, AL). Hybridization and washing procedures were performed according to the manufacturer’s (The Netherlands Cancer Institute Microarray Facility) protocol. Sixteen-bit fluorescent images from the expression arrays were acquired with an Agilent DNA Microarray Scanner (Agilent Technologies), and the resulting TIFF images were analyzed with the software GenePix Pro 4.0 (Axon Instruments, Union City, CA). For each array, a GenePix results file (.gpr) with the extracted Cy3 and Cy5 spot and background raw intensities was generated.

Expression data analysis was performed with a set of functions implemented in R (http://www.R-project.org/25). Briefly, .gpr files from the array platforms were directly loaded into the R environment using the marray package to extract the background-corrected Cy3 and Cy5 median raw intensity per spot. Intensity data from both platforms were normalized with the variance stabilization and normalization function implemented in the vsn package.26

To find genes that could better characterize the histological and mutational status of the analyzed samples, we have used “mixed-effects” regression models. Briefly, we fitted the following model: Y(i) = α + β∗histology + γ∗mutation + δ∗patient + error(i), where Y represents the log-ratio of the expression value for clone i; histology and mutation are categorical variables represented as stages (normal or adenoma) and (APC or MYH), respectively, whereas α represents the baseline expression level of clone i. A patient effect must be included in the model to correctly handle multiple samples derived from the same patient. Although both histology and mutation are assumed to have fixed effects, the patient effect is assumed to be random, as the patients included in this study represent the heterogeneous (outbred) population of familial CRC patients. This model was fitted to the data using the MAANOVA package.27 Moderated F-test statistics, which take advantage of the large number of genes being analyzed simultaneously to yield more reliable variance estimates, were extracted corresponding to histology and mutation effects. Their P values were subsequently corrected for multiple testing using Benjamini and Hochberg’s false discovery rate (FDR) method.28

Mouse Adenoma Samples

RNA samples were submitted to a double round of amplification according to the Small Sample Labeling Protocol vII (Affymetrix, Inc, Santa Clara, CA). Quality of synthesized cRNA was checked using the RNA 6000 Nano LabChip kit (Agilent Technologies). Labeled cRNA products were hybridized to mouse arrays MOE430A (Affymetrix, Inc) according to the manufacturer’s instructions. Data analysis was performed using R Statistical Computing software v2.4.125 complemented with BioConductor Packages affy,29 limma,30,31 and vsn.26 Cel files were uploaded and summarized using the affy package and normalized with vsn at the probe level. Using an empirical-Bayesian linear model (implemented in the package limma), we identified genes that were differentially expressed, with multiple testing correction performed using Benjamini and Hochberg’s FDR step-up method.28 Hierarchical clustering (Euclidean similarity measure) was performed on vsn-normalized data for all probe identifications (IDs) using Spotfire DecisionSite 9.0. (http:www.spotfire.com).

To test and/or confirm that a set of genes yielded a differential expression signature between groups of samples, the globaltest32 of vsn-normalized data was performed using R Statistical Computing software v2.4.125 complemented with the BioConductor Package globaltest.32

Functional Annotation Analysis

GenBank accession numbers or Affymetrix probe set IDs were assigned to biological process using the GO (Gene Ontology) chart feature offered by the Database for Annotation Visualization and Integrated Discovery (DAVID) 2006 (http://david.abcc.ncifcrf.gov/home.jsp). Individual genes from selected expression signatures were also placed in the context of their molecular and functional interactions by using Ingenuity Pathways Analysis (IPA) tools according to the manufacturer’s instructions (Ingenuity Systems, Redwood City, CA).

Data Integration

The cross-species comparison was performed using the Sequence Retrieval System (SRS).33 Both sets were first annotated using Unigene; next, the SRS system was used to pair the two species’ Unigene entries based on Homologene. The MOE430A array includes a total of 22,626 probes, whereas the human cDNA array encompasses 19,200 probes. Of the 22,626 mouse probes in the mouse Affymetrix platform, the SRS system retrieved a total of 12,083 homologous genes encompassed by the human array. Due to probe multiplicity in both platforms, the total overlap between the platforms consists of 18,369 entries.

From the two original data sets (mouse adenomas versus normal mucosa and human adenomas versus normal mucosa), probe sets with the same direction of differential expression were selected, adding to a total of 9495 probes. Further selection was applied based on the statistical significance according to the following thresholds: mouse data FDR <5% and human data FDR <0.5% (n = 234 probes). To exclude the possibility that the observed overlap resulted by sheer chance, we have performed a χ² test with the selected probes and reached a highly significant P value (P = 0.007).

Immunohistochemistry

Immunohistochemistry of mouse and human normal and adenomatous intestinal sections was performed according to standard protocols. The following antibodies were used and optimized for human and mouse tissues: MacMarcks (Calbiochem cat.442708; EMD Biosciences, San Diego, CA), CyclinA (cat.GTX27956; Genetex, San Antonio, TX), AnnexinI (cat.71-3400; Zymed, South San Francisco, CA). Signal detection of these antibodies was obtained by the Rabbit Envision+ System-HRP (cat.K4011; DakoCytomation, Carpinteria, CA). CD44 (cat.553131; BD Pharmingen, San Diego, CA) was detected using a secondary antibody goat anti-rat, HRP labeled.

Results

The overall rationale and strategy of the present study was to attempt the comparative expression profiling of intestinal adenomas derived from FAP patients carrying APC germline mutations23 and from the Apc^+/1638N mouse model.21,22 In both cases, we aimed at using tumor samples in which the initiating and rate-limiting mutation event is represented by loss of APC tumor-suppressing function to gain insight into conserved molecular and cellular pathways relevant for intestinal tumor initiation. Furthermore, we explored the ability of the conserved gene signature to discriminate between adenomas from polyposis patients carrying germline APC mutations from those with different genetic defects.

Expression Profiling Analysis of Familial Adenomatous Polyps

Colorectal adenomatous polyps (n = 42) have been obtained from a total of eight unrelated FAP patients carrying previously identified germline APC mutations.23 To obtain expression signatures exclusively derived from parenchymal cells and avoid the confounding effects of infiltrating and adjacent stromal cells, dysplastic tumor cells were collected by laser capture microdissection (LCM). Control LCM samples were obtained from the intestinal mucosa of three individuals with no CRC history. RNA was extracted from the microdissected tumor and normal specimens and subsequently used for expression profiling by hybridization to human 18K cDNA arrays generated at The Netherlands Cancer Institute Microarray Facility (see Materials and Methods). The expression profiling data have been deposited at the National Center for Biotechnology Information Gene Expression Omnibus (http://www.ncbi.nlm.nih.gov/geo) and is accessible through GEO Series accession number GSE9689.

Two-dimensional hierarchical clustering was first applied to all 19,200 array probes to generate an overview, without any prior filtering, of the global gene expression differences among all samples (Figure 1a). Notably, the three normal colon mucosa specimens (NC1-3) do not cluster separately from the adenoma samples. Also, several samples from the same individual are often observed to cluster together, indicative of a patient effect. In view of the latter, a statistical approach using a mixed-effects regression model34 was applied to all samples to determine whether specific patterns of gene expression could be associated with the adenomatous polyps. The linear mixed-effects model was fitted considering histology (adenoma versus normal mucosa) as having a fixed effect and patient as having a random effect. This procedure not only allowed us to calculate P values for all genes, but also to control the variance component associated with random patient-specific genetic variation, ie, the variability in gene expression related to the genetic background of each individual patient. With an FDR set to 0.5%, a total of 1859 probes appeared to be differentially expressed between normal colonic epithelium and adenomas (see Supplemental Tables S1 and S2 at http://ajp.amjpathol.org). The relatively high number of differentially expressed genes even under highly stringent conditions clearly illustrates the strong effect of histology on global gene expression. This is also further illustrated by the empirical cumulative P values distribution function, which is clearly distinct from what would be expected if no effect was detectable (Figure 1b). Of the 1859 probes, 839 (45%) and 1020 (55%) were, respectively, up- and down-regulated in tumor cells when compared with normal mucosa (see Supplemental Tables S1 and S2 at http://ajp.amjpathol.org).

Figure 1.

Unsupervised hierarchical cluster analysis of expression profiles from human and mouse intestinal polyps, without preliminary gene selection. Up- and down-regulated probes are depicted in red and green, respectively. a: Unsupervised hierarchical clustering analysis of 42 colorectal adenomatous polyps (obtained from eight unrelated FAP patients with previously identified germline APC mutations23) and 3 control normal mucosa samples (labeled as NC1-3) obtained from individuals with no history of CRC. b: Distribution of P values (left plot) relative to the comparison of patient-derived colorectal adenomas versus normal mucosa samples, sorted in ascending order (blue line). The dashed (black) line represents what would be expected if no effect was detectable. In the right plot, FDR-adjusted sorted P values are shown. The dashed line represents the FDR threshold used in our study to select the differentially expressed genes in the human set that led to the selection of 1859 probes. c: Unsupervised hierarchical clustering analysis of duodenal adenomas (n = 3, labeled T1–T3) and normal mucosa (n = 2, N1–N2) samples obtained from inbred C57BL6/J Apc^+/1638N and Apc^+/+ mice, respectively. d: Distribution of P values (left plot) relative to the comparison of mouse duodenal adenomas versus normal tissue samples, sorted in ascending order (blue line). The dashed (black) line represents what would be expected if no effect was detectable. In the right plot, FDR-adjusted sorted P values are shown. The dashed line represents the FDR threshold used in our study to select the differentially expressed genes in the mouse set that led to the selection of 4137 probes.

To gain insight into the biological relevance of the newly generated list of genes differentially expressed in adenomatous polyps, we used the GO-based bioinformatics tool DAVID35 (see Materials and Methods). An overview of the functional categories represented by the genes differentially expressed among APC-mutant adenomatous polyps when compared with normal colonic mucosa reveals a very broad spectrum of biological processes, ranging from different aspects of cellular metabolism to apoptosis, cell migration, and immune response (data not shown). The broadness of the transcriptional profiles of the colorectal polyps when compared with normal mucosa is likely to reflect the heterogeneity of these benign tumors even at this very early stage of the adenoma-carcinoma sequence.

Expression Profiling Analysis of Apc^+/1638N Mouse Intestinal Adenomas

Duodenal adenomas and normal mucosal samples from age- and sex-matched C57BL6/J Apc^+/1638N mice (n = 3) and Apc^+/+ controls (n = 2) were collected and snap-frozen as for the above human polyps. Histological analysis of these lesions confirmed their benign adenomatous nature (not shown). Also in these cases, dysplastic epithelial cells were collected by LCM. Control expression signatures were obtained from wild-type C57BL6/J epithelial cells microdissected from the same anatomical location. Total RNA was extracted from normal and tumor samples and hybridized to Affymetrix MOE430A arrays. The corresponding data have been deposited in National Center for Biotechnology Information’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE9580. Unsupervised two-dimensional hierarchical clustering was able to discriminate and correctly cluster tumor from normal samples (Figure 1c). Empirical-Bayesian linear regression analysis allowed the identification of statistically significant differences between normal and tumor samples.30,31 Notwithstanding the admittedly limited sample size, a strong gene expression signature of the tumor samples is detected as illustrated by the empirical cumulative distribution function of the P values, which is clearly distinct from what would be expected if no effect was detectable (Figure 1d). Such a strong histology-specific gene expression signature, despite the small sample size, may result from the use of inbred animals in which genetic background is identical among unrelated tumor-bearing mice. An FDR threshold of 5% resulted in the identification of as many as 4137 probes differentially regulated between normal and tumor tissue, and 2163 (52%) and 1974 (48%) up- and down-regulated, respectively (see Supplemental Tables S3 and S4 at http://ajp.amjpathol.org). As for the human gene list, annotation analysis of the mouse genes by DAVID revealed a very broad and partially overlapping spectrum of cellular functions (data not shown).

Cross-Species Comparison

As shown above, expression profiling analysis of human and mouse intestinal adenomas and their normal tissue counterparts resulted in very extensive lists of differentially expressed genes even when stringent parameters were used. We postulated that the cross-species comparison of the genes differentially expressed between the two independent data sets would limit bystander and adaptation effects and help in narrowing down the list to conserved transcripts more likely to play relevant roles in the tumorigenic process. As the two profiling data sets were generated with different microarray platforms (cDNA and oligonucleotide arrays for the human and mouse tumors, respectively), the comparative analysis was performed exclusively on the probes present in both platforms (see Materials and Methods).

Conserved probes were selected by applying the following inclusion criteria: FDR <5% for the mouse set (n = 4137 probes) and FDR <0.5% for the human set (n = 1859 probes). Further filtering was performed to eliminate probes with discordant transcriptional directions (eg, up- vs. down-regulated probes) between the two species. Following this procedure, a total of 234 probes representative of 166 genes were selected, 100 and 66 of which were up- and down-regulated, respectively (Tables 1 and 2). In those cases in which a gene is represented by more than one probe in the array platform, the probe ID associated with the lowest FDR value was selected.

Table 1.

The Cross-Species Conserved 166-Gene Signature: Up-Regulated Genes

Mouse			Human
Probe ID	Unigene*	Gene symbol	Unigene†	Gene symbol	Gene description
1448213_at	Mm.248360	Anxa1	Hs.494173	ANXA1	Annexin A1
1424460_s_at	Mm.284649	Aytl2	Hs.368853	AYTL2	Acyltransferase like 2
1424278_a_at	Mm.8552	Birc5	Hs.514527	BIRC5	Baculoviral IAP repeat-containing 5 (survivin)
1416815_s_at	Mm.927	Bub3	Hs.418533	BUB3	BUB3 budding uninhibited by benzimidazoles 3 homolog (yeast)
1455356_at	Mm.36834	Camsap1	Hs.522493	CAMSAP1	Calmodulin-regulated spectrin-associated protein 1
1416884_at	Mm.280968	Cbx3	Hs.381189	CBX3	Chromobox homolog 3 (HP1 gamma homolog, Drosophila)
1425616_a_at	Mm.36697	Ccdc23	Hs.113919	CCDC23	Coiled-coil domain containing 23
1427031_s_at	Mm.24035	Ccdc52	Hs.477144	CCDC52	Coiled-coil domain containing 52
1417911_at	Mm.4189	Ccna2	Hs.58974	CCNA2	Cyclin A2
1417419_at	Mm.273049	Ccnd1	Hs.523852	CCND1	Cyclin D1
1438560_x_at	Mm.296985	Cct4	Hs.421509	CCT4	Chaperonin containing TCP1, subunit 4 (delta)
1417258_at	Mm.282158	Cct5	Hs.1600	CCT5	Chaperonin containing TCP1, subunit 5 (epsilon)
1423760_at	Mm.423621	Cd44	Hs.502328	CD44	CD44 molecule (Indian blood group)
1452242_at	Mm.9916	Cep55	Hs.14559	CEP55	Centrosomal protein 55 kDa
1417457_at	Mm.222228	Cks2	Hs.83758	CKS2	CDC28 protein kinase regulatory subunit 2
1449300_at	Mm.200327	Cttnbp2 nl	Hs.485899	CTTNBP2NL	CTTNBP2 N-terminal like
1454268_a_at	Mm.271671	Cyba	Hs.513803	CYBA	Cytochrome b-245, alpha polypeptide
1419275_at	Mm.148693	Dazap1	Hs.222510	DAZAP1	DAZ associated protein 1
1424198_at	Mm.68971	Dlg5	Hs.500245	DLG5	Discs, large homolog 5 (Drosophila)
1435122_x_at	Mm.128580	Dnmt1	Hs.202672	DNMT1	DNA (cytosine-5-)-methyltransferase 1
1452052_s_at	Mm.27695	Eif3s1	Hs.404056	EIF3S1	Eukaryotic translation initiation factor 3, subunit 1 alpha
1426674_at	Mm.21671	Eif3s9	Hs.371001	EIF3S9	Eukaryotic translation initiation factor 3, subunit 9 eta
1448797_at	Mm.4454	Elk3	Hs.591015	ELK3	ETS-domain protein (SRF accessory protein 2)
1437211_x_at	Mm.427018	Elovl5	Hs.520189	ELOVL5	ELOVL family member 5, elongation of long chain fatty acids (FEN1/Elo2, SUR4/Elo3-like, yeast)
1420965_a_at	Mm.241073	Enc1	Hs.104925	ENC1	Ectodermal-neural cortex (with BTB-like domain)
1451550_at	Mm.6972	Ephb3	Hs.2913	EPHB3	EPH receptor B3
1417301_at	Mm.4769	Fzd6	Hs.591863	FZD6	Frizzled homolog 6 (Drosophila)
1419595_a_at	Mm.20461	Ggh	Hs.78619	GGH	Gamma-glutamyl hydrolase (conjugase, folylpolygammaglutamyl hydrolase)
1419205_x_at	Mm.46029	Gpatc4	Hs.193832	GPATC4	G patch domain containing 4
1433736_at	Mm.248353	Hcfc1	Hs.83634	HCFC1	Host cell factor C1 (VP16-accessory protein)
1423051_at	Mm.426956	Hnrpu	Hs.166463	HNRPU	Heterogeneous nuclear ribonucleoprotein U (scaffold attachment factor A)
1426705_s_at	Mm.21118	Iars	Hs.445403	IARS	Isoleucine-tRNA synthetase
1422546_at	Mm.440026	Ilf3	Hs.465885	ILF3	Interleukin enhancer binding factor 3
1456097_a_at	Mm.257094	Itgb3bp	Hs.166539	ITGB3BP	Integrin beta 3 binding protein (beta3-endonexin)
1421344_a_at	Mm.100253	Jub	Hs.655832	JUB	ajuba homolog (Xenopus laevis)
1452118_at	Mm.102761	2600005C20Rik	Hs.129621	KIAA0179	KIAA0179
1427080_at	Mm.29068	2610036D13Rik	Hs.370118	KIAA0406	KIAA0406
1448169_at	Mm.22479	Krt18	Hs.406013	KRT18	Keratin 18
1416621_at	Mm.285453	Llgl1	Hs.513983	LLGL1	Lethal giant larvae homolog 1 (Drosophila)
1434210_s_at	Mm.245210	Lrig1	Hs.518055	LRIG1	Leucine-rich repeats and immunoglobulin-like domains 1
1417511_at	Mm.28560	Lyar	Hs.425427	LYAR	Hypothetical protein FLJ20425
1439426_x_at	Mm.177539	Lzp-s	Hs.524579	LYZ	Lysozyme (renal amyloidosis)
1455941_s_at	Mm.325746	Map2k5	Hs.114198	MAP2K5	Mitogen-activated protein kinase kinase 5
1437226_x_at	Mm.424974	Marcksl1	Hs.75061	MARCKSL1	MARCKS-like 1
1439081_at	Mm.122725	Mgea5	Hs.500842	MGEA5	Meningioma expressed antigen 5 (hyaluronidase)
1424001_at	Mm.280311	Mki67ip	Hs.367842	MKI67IP	MKI67 (FHA domain) interacting nucleolar phosphoprotein
1449478_at	Mm.4825	Mmp7	Hs.2256	MMP7	Matrix metallopeptidase 7 (matrilysin, uterine)
1455129_at	Mm.130883	Mtdh	Hs.377155	MTDH	Metadherin
1419254_at	Mm.443	Mthfd2	Hs.469030	MTHFD2	Methylenetetrahydrofolate dehydrogenase (NADP+ dependent) 2, methenyltetrahydrofolate cyclohydrolase
1452778_x_at	Mm.290407	Nap1l1	Hs.524599	NAP1L1	Nucleosome assembly protein 1-like 1
1423046_s_at	Mm.290027	Ncbp2	Hs.591671	NCBP2	Nuclear cap binding protein subunit 2
1455035_s_at	Mm.29363	Nol5a	Hs.376064	NOL5A	Nucleolar protein 5A (56 kDa with KKE/D repeat)
1416606_s_at	Mm.28203	Nola2	Hs.27222	NOLA2	Nucleolar protein family A, member 2 (H/ACA small nucleolar RNPs)
1449140_at	Mm.276504	Nudcd2	Hs.140443	NUDCD2	NudC domain containing 2
1428277_at	Mm.387724	Otud6b	Hs.30532	OTUD6B	OTU domain containing 6B
1435368_a_at	Mm.277779	Parp1	Hs.177766	PARP1	poly-(ADP-ribose) polymerase family, member 1
1452620_at	Mm.29856	Pck2	Hs.75812	PCK2	Phosphoenolpyruvate carboxykinase 2 (mitochondrial)
1426838_at	Mm.37562	Pold3	Hs.82502	POLD3	Polymerase (DNA-directed), delta 3, accessory subunit
1427094_at	Mm.9199	Pole2	Hs.162777	POLE2	Polymerase (DNA directed), epsilon 2 (p59 subunit)
1449648_s_at	Mm.3458	Rpo1–1	Hs.584839	POLR1C	polymerase (RNA) I polypeptide C
1433552_a_at	Mm.273217	Polr2b	Hs.602757	POLR2B	polymerase (RNA) II (DNA directed) polypeptide B
1436505_at	Mm.11815	Ppig	Hs.470544	PPIG	Peptidylprolyl isomerase G (cyclophilin G)
1428265_at	Mm.7726	Ppp2r1b	Hs.546276	PPP2R1B	protein phosphatase 2 (formerly 2A), regulatory subunit A, beta isoform
1423775_s_at	Mm.227274	Prc1	Hs.567385	PRC1	protein regulator of cytokinesis 1
1452032_at	Mm.30039	Prkar1a	Hs.280342	PRKAR1A	Protein kinase, cAMP-dependent, regulatory, type I, alpha (tissue-specific extinguisher 1)
1451576_at	Mm.71	Prkdc	Hs.491682	PRKDC	Protein kinase, DNA-activated, catalytic polypeptide
1435859_x_at	Mm.2462	Psmc2	Hs.437366	PSMC2	Proteasome (prosome, macropain) 26S subunit, ATPase, 2
1426631_at	Mm.58660	Pus7	Hs.520619	PUS7	Pseudouridylate synthase 7 homolog (S. cerevisiae)
1448899_s_at	Mm.204634	Rad51ap1	Hs.591046	RAD51AP1	RAD51-associated protein 1
1423700_at	Mm.12553	Rfc3	Hs.115474	RFC3	Replication factor C (activator 1) 3
1456375_x_at	Mm.314056	Trim27	Hs.440382	RFP	Tripartite motif-containing 27
1439403_x_at	Mm.435574	Rnf12	Hs.653288	RNF12	Ring finger protein 12
1437309_a_at	Mm.180734	Rpa1	Hs.461925	RPA1	Replication protein A1
1453362_x_at	Mm.16775	Rps24	Hs.356794	RPS24	Ribosomal protein S24
1416276_a_at	Mm.66	Rps4x	Hs.446628	RPS4X	Ribosomal protein S4, X-linked
1416120_at	Mm.99	Rrm2	Hs.226390	RRM2	Ribonucleotide reductase M2 polypeptide
1422864_at	Mm.4081	Runx1	Hs.149261	RUNX1	Runt-related transcription factor 1 (acute myeloid leukemia 1; aml1 oncogene)
1420824_at	Mm.33903	Sema4 d	Hs.655281	SEMA4D	Sema domain, immunoglobulin domain (Ig), transmembrane domain (TM) and short cytoplasmic domain, (semaphorin) 4D
1434972_x_at	Mm.391719	Sfrs1	Hs.68714	SFRS1	Splicing factor, arginine/serine-rich 1 (splicing factor 2, alternate splicing factor)
1417623_at	Mm.399997	Slc12a2	Hs.162585	SLC12A2	Solute carrier family 12 (sodium/potassium/chloride transporters), member 2
1418326_at	Mm.27943	Slc7a5	Hs.513797	SLC7A5	Solute carrier family 7 (cationic amino acid transporter, y+ system), member 5
1422771_at	Mm.325757	Smad6	Hs.153863	SMAD6	SMAD family member 6
1424206_at	Mm.246803	Smarca5	Hs.589489	SMARCA5	SWI/SNF-related, matrix-associated, actin-dependent regulator of chromatin, subfamily a, member 5
1452422_a_at	Mm.1323	Snrpb2	Hs.280378	SNRPB2	Small nuclear ribonucleoprotein polypeptide B
1419156_at	Mm.240627	Sox4	Hs.643910	SOX4	SRY (sex determining region Y)-box 4
1433502_s_at	Mm.220843	Tsr1	Hs.388170	SRR	TSR1, 20S rRNA accumulation, homolog (S. cerevisiae)
1415849_s_at	Mm.378957	Stmn1	Hs.693592	STMN1	Stathmin 1/oncoprotein 18
1450743_s_at	Mm.260545	Syncrip	Hs.571177	SYNCRIP	Synaptotagmin binding, cytoplasmic RNA interacting protein
1423601_s_at	Mm.2215	Tcof1	Hs.519672	TCOF1	Treacher Collins-Franceschetti syndrome 1
1416358_at	Mm.34564	0610009O03Rik	Hs.632581	TETRAN	Tetracycline transporter-like protein (TETRAN)
1434317_s_at	Mm.272025	Tex10	Hs.494648	TEX10	Testis expressed sequence 10
1426397_at	Mm.172346	Tgfbr2	Hs.82028	TGFBR2	Transforming growth factor, beta receptor II
1424641_a_at	Mm.219648	Thoc1	Hs.654460	THOC1	THO complex 1
1427318_s_at	Mm.34674	Fer1l3	Hs.500572	FER1L3	Fer-1-like 3, myoferlin (C. elegans)
1449041_a_at	Mm.27063	Trip6	Hs.534360	TRIP6	Thyroid hormone receptor interactor 6
1437906_x_at	Mm.19169	Txnl1	Hs.114412	TXNL1	Thioredoxin-like 1
1422842_at	Mm.3065	Xrn2	Hs.255932	XRN2	5′-3′ exoribonuclease 2
1448363_at	Mm.221992	Yap1	Hs.503692	YAP1	Yes-associated protein 1
1427208_at	Mm.289103	Zfp451	Hs.485628	ZNF451	Zinc finger protein 451
1416757_at	Mm.335237	Zwilch	Hs.21331	ZWILCH	Zwilch, kinetochore associated, homolog (Drosophila)

For simplicity, the gene description is only given for the human entry. 

^*Unigene build 163; 

^†Unigene build 202. 

Table 2.

The Cross-Species Conserved 166-Gene Signature: Down-Regulated Genes

Mouse			Human
Probe ID	Unigene*	Gene	Unigene†	Gene	Description
1424600_at	Mm.213898	Abp1	Hs.647097	ABP1	Amiloride binding protein 1 [amine oxidase (copper-containing)]
1427034_at	Mm.754	Ace	Hs.298469	ACE	Angiotensin I-converting enzyme (peptidyldipeptidase A) 1
1418553_at	Mm.170461	Arhgef18	Hs.465761	ARHGEF18	Rho/rac guanine nucleotide exchange factor (GEF) 18
1459924_at	Mm.340818	Atp6v0a1	Hs.463074	ATP6V0A1	ATPase, H+ transporting, lysosomal V0 subunit a1
1416582_a_at	Mm.4387	Bad	Hs.370254	BAD	BCL2-antagonist of cell death
1423635_at	Mm.103205	Bmp2	Hs.73853	BMP2	Bone morphogenetic protein 2
1456616_a_at	Mm.726	Bsg	Hs.591382	BSG	BSG: Basigin (Ok blood group)
1424226_at	Mm.218590	9030617O03Rik	Hs.309849	C14orf159	Chromosome 14 open reading frame 159
1427944_at	Mm.150568	C1qdc1	Hs.234355	C1QDC1	C1q domain containing 1
1449248_at	Mm.177761	Clcn2	Hs.436847	CLCN2	Chloride channel 2
1416565_at	Mm.400	Cox6b1	Hs.431668	COX6B1	Cytochrome c oxidase subunit Vib polypeptide 1 (ubiquitous)
1420617_at	Mm.339792	Cpeb4	Hs.127126	CPEB4	Cytoplasmic polyadenylation element binding protein 4
1415677_at	Mm.21623	Dhrs1	Hs.348350	DHRS1	Dehydrogenase/reductase (SDR family) member 1
1416697_at	Mm.1151	Dpp4	Hs.368912	DPP4	Dipeptidylpeptidase 4 (CD26, adenosine deaminase complexing protein 2)
1450314_at	Mm.140332	Dqx1	Hs.191705	DQX1	DEAQ box polypeptide 1 (RNA-dependent ATPase)
1421136_at	Mm.9478	Edn3	Hs.1408	EDN3	Endothelin 3
1423005_a_at	Mm.264215	Espn	Hs.147953	ESPN	Espin
1421969_a_at	Mm.256025	Faah	Hs.528334	FAAH	Fatty acid amide hydrolase
1452117_a_at	Mm.170905	Fyb	Hs.370503	FYB	FYN binding protein (FYB-120/130)
1436889_at	Mm.338713	Gabra1	Hs.175934	GABRA1	γ -Aminobutyric acid (GABA) A receptor, alpha 1
1423236_at	Mm.30249	Galnt1	Hs.514806	GALNT1	UDP-N-acetyl-α -d-galactosamine:polypeptide N-acetylgalactosaminyltransferase 1 (GalNAc-T1)
1418863_at	Mm.247669	Gata4	Hs.243987	GATA4	GATA binding protein 4
1429076_a_at	Mm.283495	Gdpd2	Hs.438712	GDPD2	Glycerophosphodiester phosphodiesterase domain containing 2
1449144_at	Mm.260925	Gna11	Hs.654784	GNA11	Guanine nucleotide binding protein (G protein), alpha 11 (Gq class)
1419371_s_at	Mm.195451	Gosr2	Hs.463278	GOSR2	Golgi SNAP receptor complex member 2
1416416_x_at	Mm.37199	Gstm1	Hs.75652	GSTM5	Glutathione S-transferase M5
1425343_at	Mm.41506	Hdhd3	Hs.7739	HDHD3	Haloacid dehalogenase-like hydrolase domain containing 3
1422527_at	Mm.16373	H2-DMa	Hs.351279	HLA-DMA	HLA-DMA: Major histocompatibility complex, class II, DM alpha
1419455_at	Mm.4154	Il10rb	Hs.418291	IL10RB	Interleukin 10 receptor, beta
1418265_s_at	Mm.1149	Irf2	Hs.374097	IRF2	Interferon regulatory factor 2
1433775_at	Mm.331907	C77080	Hs.591502	KIAA1522	KIAA1522
1425547_a_at	Mm.279599	Klc4	Hs.408062	KLC4	Kinesin light chain 4
1451322_at	Mm.28108	Cmbl	Hs.192586	CMBL	Carboxymethylenebutenolidase homolog (Pseudomonas)
1425780_a_at	Mm.241387	Tmem167	Hs.355606	TMEM167	Transmembrane protein 167
1425704_at	Mm.192213	BC022224	Hs.462859	MGC4172	Short-chain dehydrogenase/reductase
1425930_a_at	Mm.628	Mlx	Hs.383019	MLX	MAX-like protein X
1450376_at	Mm.2154	Mxi1	Hs.501023	MXI1	MAX interactor 1
1425230_at	Mm.31686	Nags	Hs.8876	NAGS	N-Acetylglutamate synthase
1448331_at	Mm.29683	Ndufb7	Hs.532853	NDUFB7	NADH dehydrogenase (ubiquinone) 1 beta subcomplex, 7
1415821_at	Mm.15125	Nptn	Hs.187866	NPTN	Neuroplastin
1451274_at	Mm.276348	Ogdh	Hs.488181	OGDH	Oxoglutarate (α- ketoglutarate) dehydrogenase (lipoamide)
1417677_at	Mm.32744	Opn3	Hs.534399	OPN3	Opsin 3
1449330_at	Mm.29872	Pdzd3	Hs.374726	PDZD3	PDZ domain containing 3
1435872_at	Mm.328931	Pim1	Hs.81170	PIM1	Pim-1 oncogene
1425542_a_at	Mm.240396	Ppp2r5c	Hs.368264	PPP2R5C	Protein phosphatase 2, regulatory subunit B′, gamma isoform
1422847_a_at	Mm.2314	Prkcd	Hs.155342	PRKCD	Protein kinase C, delta
1424456_at	Mm.4341	Pvrl2	Hs.655455	PVRL2	Poliovirus receptor-related 2 (herpesvirus entry mediator B)
1430527_a_at	Mm.261818	Rnf167	Hs.7158	RNF167	Ring finger protein 167
1448704_s_at	Mm.22362	H47	Hs.32148	SELS	Selenoprotein S
1448299_at	Mm.246670	Slc1a1	Hs.444915	SLC1A1	Solute carrier family 1 (neuronal/epithelial high affinity glutamate transporter, system Xag), member 1
1424441_at	Mm.330113	Slc27a4	Hs.656699	SLC27A4	Solute carrier family 27 (fatty acid transporter), member 4
1433595_at	Mm.281800	Slc35d1	Hs.213642	SLC35D1	Solute carrier family 35 (UDP-glucuronic acid/UDP-N-acetylgalactosamine dual transporter), member D1
1421225_a_at	Mm.41044	Slc4a4	Hs.5462	SLC4A4	Solute carrier family 4, sodium bicarbonate cotransporter, member 4
1448783_at	Mm.45874	Slc7a9	Hs.408567	SLC7A9	Solute carrier family 7 (cationic amino acid transporter, y+ system), member 9
1436797_a_at	Mm.300594	Surf4	Hs.512465	SURF4	Surfeit 4
1428095_a_at	Mm.33869	Tmem24	Hs.587176	TMEM24	Transmembrane protein 24
1417895_a_at	Mm.25295	Tmem54	Hs.534521	TMEM54	Transmembrane protein 54
1434553_at	Mm.26088	Tmem56	Hs.483512	TMEM56	Transmembrane protein 56
1420412_at	Mm.1062	Tnfsf10	Hs.478275	TNFSF10	Tumor necrosis factor (ligand) superfamily, member 10
1428327_at	Mm.305318	Trak1	Hs.535711	TRAK1	Trafficking protein, kinesin binding 1
1448737_at	Mm.18590	Tspan7	Hs.441664	TSPAN7	Tetraspanin 7
1448782_at	Mm.291015	Txndc11	Hs.313847	TXNDC11	Thioredoxin domain containing 11
1435110_at	Mm.290433	Unc5b	Hs.585457	UNC5B	Unc-5 homolog B
1426399_at	Mm.26515	Vwa1	Hs.449009	VWA1	Von Willebrand factor A domain containing 1
1436953_at	Mm.223504	Wipf1	Hs.654521	WASPIP	WAS/WASL interacting protein family, member 1
1416545_at	Mm.240076	Zdhhc7	Hs.592065	ZDHHC7	Zinc finger, DHHC-type containing 7

For simplicity, the gene description is only given for the human entry. 

^*Unigene build 163; 

^†Unigene build 202. 

Bioinformatic analysis of the 166 genes was performed by assigning them to functional groups based on their GO classification in addition to other information from the scientific literature (Table 3). Overall, up-regulation of genes with functions related to cell division was observed: DNA replication and repair, cell cycle regulation, and the maintenance of genomic integrity. Also, the transcriptional and translational machinery was up-regulated when compared with normal tissues.

Table 3.

GO Annotations of the Cross-Species Conserved Genes

Cellular function	Direction	Genes
Cell cycle	Up	CCND1, CCNA2, CKS2*, CEP55
DNA replication	Up	RRM2, RFC3, PUS7*, POLE2, RPA1, NAP1L1
DNA repair	Up	XRN2*, PUS7*, POLD3, PARP1, RAD51AP1
Apoptosis	Up	DLG5, BIRC5
	Down	BAD, IRF2*, TNFSF10*, UNC5B, PIM1
RNA and protein biogenesis, processing and transport	Up	RPS4X, NOLA2, CCT5, ILF3, NCBP2, TCOF1, MKI67IP, THOC1, EIF3S9, EIF3S1, POLR2B, SYNCRIP, IARS, SNRPB2, RPS24, NOLA5A, HNRPU, XRN2*, POLR1C, SFRS1, PPIG, CCT4
	Down	ZDHHC7, GOSR2, GALNT1, TRAK1
TGFβ	Up	TGFBR2, SMAD6
	Down	BMP2
Chromatin remodelization	Up	CBX3, SMARCA5, DNMT1
Cytoskeleton organization	Up	LLGL1, KRT18, CTTNBP2NL
	Down	ARHGEF18, KLC4, WASPIP*, ESPN*
Genome integrity (mitotic	Up	STMN1, ZWILCH, BUB3, CKS2*, PRC1, BIRC5*, PARP1*, PRKDC
checkpoint and telomere maintenance)	Down	ESPN*
Cell adhesion and migration	Up	SEMA4D, JUB, CD44, TRIP6, DLG5, MAP2K5
	Down	NPTN, PVRL2, TSPAN7, WASPIP*
Transcription factors	Up	SOX4, RUNX1, YAP1, MTDH, ITGB3BP, ZFP451, RFP, RNF12
	Down	MLX, MXI1, GATA4, IRF2
G protein signaling	Up	PRKAR1A
	Down	OPN3, PRKCD, GNA11
Immune response	Up	ANXA1
	Down	IRF2*, TNFSF10*, IL10RB, HLA-DMA, BSG, SELS
Chromatin remodelization	Up	CBX3, SMARCA5, DNMT1
Metabolism	Up	SLC7A5, MTHFD2, GGH, AYTL2, MGEA5, PCK2
	Down	DHRS1, GSTM1, DGAT1, SLC27A4, NAGS, HDHD3, GDPD2, SLC7A9, OGDH
Proteolysis	Up	ENC1, PSMC2, MMP7
	Down	DPP4, ACE, RNF167

Functional annotations were retrieved from GO and the scientific literature. Genes belonging to more than one functional category are marked with*. Gene symbols refer to the human annotation. 

To map the conserved genes to existing signaling and cellular pathways, we used the web-based software application IPA (Ingenuity Systems). As expected from our selection of human and mouse intestinal tumors arising from APC mutations, IPA revealed several differentially expressed genes encoding for members of the Wnt signal transduction pathway (Figure 2). Of the other pathways included in the IPA database, only the extracellular signal-regulated kinase/mitogen-activated protein kinase signaling network encompassed more than two differentially expressed genes in the conserved signature, namely PKA, PKC, PP1/PP2A, and ELK3 (not shown).

Figure 2.

IPA of the genes encompassed by the conserved 166 signatures and belonging to the Wnt signal transduction pathway. The canonical Wnt pathway from the IPA database was slightly modified to accommodate additional Wnt target genes.11,12 The signaling network is represented graphically as nodes (symbols representing genes) and lines/arrows (biological relationship between the genes according to the legend). Red and green gene symbols denote up- and down-regulated genes, respectively. White symbols denote genes not differentially expressed in the conserved signature.

Immunohistochemical Validation of Conserved Targets

To validate the results of our comparative cross-species expression analysis, we have performed immunohistochemistry (IHC) on mouse and human intestinal tissues with antibodies directed against proteins encoded by differentially expressed genes from the list reported in Table 1. Enhanced expression of the cell surface glycoprotein CD44 is an early event in the adenoma-carcinoma sequence both in mouse and human,36,37 and is thought to result from direct CD44 transcriptional up-regulation by Wnt/β-catenin signaling.37 Accordingly, CD44 was found to be conserved in our cross-species analysis and was used as an internal positive control for the IHC analysis (Figure 3, m–p).

Figure 3.

Immunohistochemistry validation analysis of cross-species conserved genes. Human (colorectal polyps and normal mucosa from FAP patients carrying germline APC mutations) and mouse (duodenal adenomas and normal mucosa from inbred C57BL6/J Apc^+/1638N and Apc^+/+ mice) tissue sections were analyzed with specific antibodies (see Materials and Methods) for expression of the following proteins: ANXA1 (a–d), CCNA2 (e–h), MARCKSL1 (i–l), and CD44 (m–p).

The annexin A1 (ANXA1) gene, up-regulated in both human and mouse APC/Apc-mutant adenomas (Table 1), encodes for annexin A1, an anti-inflammatory protein induced by glucocorticoids and overexpressed in colitis in both human and rat.38,39,40 Annexin A1 IHC analysis reveals a distinct perinuclear localization in normal intestinal mucosa, possibly in association with the endoplasmic reticulum (Figure 3, a and b). In Apc^+/1638N and FAP intestinal tumors, cytoplasmic accumulation of annexin A1 is observed concomitantly with loss of the perinuclear localization (Figure 3, c and d). In distinct tumor areas, nuclear localization was also observed, possibly suggestive of mitogenic stimulation as previously reported.41 Also, annexin A1 expression appears not to be confined to the intestinal epithelium, but also to extend to the stromal compartment42 (Figure 3, a–d).

Cyclin A2 (CCNA2) is a ubiquitously expressed regulator of cell cycle progression known to promote G₁/S and G₂/M transitions.43 In normal mouse (upper GI) and human (colon) intestinal mucosa, CCNA2 IHC analysis shows nuclear expression mainly restricted to the crypts (Figure 3, e–h). CCNA2 up-regulation in both mouse and human intestinal adenomas (Table 1) is reflected by an increase in the relative number of cells with nuclear CCNA2 staining spread throughout the tumors. The latter is indicative of enhanced cell proliferation of APC-mutant tumor cells, as was also confirmed by the GO analysis of the conserved gene list (Table 3).

To date, the function of Marks-like protein 1 (MARCKSL1) is not fully elucidated, although evidence in the literature indicates that it might be involved in the regulation of intracellular Ca²⁺ levels under the control of protein kinase C.44 Up-regulation of this protein has been previously reported in prostate cancer,45 and is here found in both FAP and Apc^+/1638N adenomas (Table 1). Similar to what was observed for CCCNA2, MARCKSL1 expression is limited to a specific subset of cells within the normal human (colon) and mouse (upper GI) intestinal crypts. Likewise, a pronounced increase in cytoplasmic expression is observed in the vast majority of tumor cells (Figure 3, i–l). Overall, the above IHC results validate the cross-species expression profiling data for genes belonging to distinct GO and functional categories.

The Conserved Cross-Species Signature As a Tool to Differentiate Hereditary Polyposis Syndromes

Apart from its implications for the understanding of the molecular and cellular mechanisms underlying APC-driven colorectal tumorigenesis, the conserved cross-species signature may represent a useful tool to discriminate among adenomas from hereditary patients with different genetic syndromes, namely APC- and MYH-associated polyposis. To this aim, an additional 14 adenomas have been obtained from five unrelated patients with pathologically confirmed polyposis of the colon and carrying biallelic MYH germline mutations.23 As for the APC-mutant polyps, RNA was extracted from the microdissected MAP adenomas and subsequently used for expression profiling. First, unsupervised hierarchical clustering was applied to all 56 profiles (from both the APC- and MYH-mutant patients) without any prior filtering to generate an overview of the global gene expression differences among all samples. Overall, we could not observe any clear association with mutation status (data not shown). The mixed-effects regression model34 was again applied, this time fitted to consider mutation (APC and MYH germline mutation carriers) as having a fixed effect and patient as having a random effect. With an FDR set to 0.5%, we were able to select 49 genes differentially expressed between FAP and MAP adenomas (Table 4). To investigate further whether the 49-gene signature as a whole can predict the underlying APC or MYH gene defect, we applied the previously described globaltest for the analysis of microarray data.32 This test assesses whether the global expression pattern of a group of genes is significantly related to any given parameter. It should be noted that, when applying the globaltest, the patient effect cannot be regarded as random and therefore be controlled by its inclusion in the model as a confounder, as it would also represent the genotype effect (all samples from a patient belong to the same genotype). We circumvented this problem by first selecting at random one sample at a time from each patient and then applying the globaltest. After repeating this process for 1000 random combinations of patients’ samples, 57% of the computed P values were found to be below the 0.05 threshold, whereas the maximum value is close to 0.5 (Figure 4a). Next, we repeated this procedure with the conserved 166-gene signature. In sharp contrast with the previous result, 99.4% of the computed P values are below the 0.05 threshold with a maximum of 0.06 (Figure 4b). Accordingly, two-dimensional hierarchical clustering analysis of the expression profiles obtained from all of the FAP and MAP polyps with the 49- and 166-gene signatures confirms that the latter is considerably more discriminative than the former in resolving tumors from carriers of APC germline mutations from those derived from MAP patients (Figure 4, c and d).

Table 4.

The 49-Gene Signature Based on Statistically Significant Differences (FDR = 0.5%) Between Expression Profiles of FAP- and MAP-Derived Adenomatous Polyps, After Implementation of the Mixed-Effect Regression Model34 Fitted Considering Mutation (APC vs. MYH) as Having Fixed Effect and Patient as Having a Random Effect

GenBank	Gene symbol	Gene description
H41285	GDPD2	Glycerophosphodiester phosphodiesterase domain containing 2
T46878	EIF3S1	Eukaryotic translation initiation factor 3, subunit 1 alpha
AA479795	ISG20	Interferon-stimulated exonuclease gene
AA151214	G3BP2	Ras-GTPase activating protein SH3 domain-binding protein 2
H28091	PMP22	Peripheral myelin protein 22
N59330	NUP35	Nucleoporin
H19333	LOC285550	Hypothetical protein LOC285550
AA449688	FLJ32065	Hypothetical protein FLJ32065
N/A	N/A	—
AA977417	AA977417	—
N50636	RAP1GDS1	RAP1, GTP-GDP dissociation stimulator 1
T61866	IPO7	Importin 7
AA453435	LTV1	LTV1 homolog (S. cerevisiae)
N91962	EEF1E1	Eukaryotic translation elongation factor 1 epsilon 1
H77636	CD68	CD68 antigen
AI308916	PRSS3	Protease, serine, 3 (mesotrypsin)
AA478589	APOE	Apolipoprotein E
AA459401	KLK10	Kallikrein 10
AA625765	DDA1	DDA1
AA205665	SET	SET translocation (myeloid leukemia-associated)
AA707453	FLJ43855	Similar to sodium- and chloride-dependent creatine transporter
AA464147	CARS	Cysteinyl-tRNA synthetase
AA456630	ARHGEF18	Rho/rac guanine nucleotide exchange factor (GEF) 18
N20475	CTSD	Similar to RIKEN cDNA 6330512M04 gene (mouse)
N31935	ANGPTL1	Angiopoietin-like 1
R77512	PCDH1	Protocadherin 1 (cadherin-like 1)
N31492	FMO4	Topoisomerase (DNA) I pseudogene 1
N45236	KIAA0114	KIAA0114 gene product
H60549	CD59	CD59 antigen, complement regulatory protein
AA907626	KIF26B	Kinesin family member 26B
AA917374	TIMP2	TIMP metallopeptidase inhibitor 2
AA983530	VNN1	Vanin 1
N90109	NCL	U23 small nucleolar RNA
H15431	POLR2D	Polymerase (RNA) II (DNA directed) polypeptide D
H52673	BAK1	BCL2-antagonist/killer 1
T72259	CYP2A6	Cytochrome P450, family 2, subfamily A, polypeptide 6
AA455910	F2R	Coagulation factor II (thrombin) receptor
AA775840	C9orf123	Chromosome 9 open reading frame 123
AA464566	LRP1	Low density lipoprotein-related protein 1
AA489640	IFIT1	Interferon-induced protein with tetratricopeptide repeats 1
AA962541	LOC286167	Hypothetical protein LOC286167
AA464421	PCGF2	Polycomb group ring finger 2
H25761 AI668603	—	—
AA447748	DLD	Dihydrolipoamide dehydrogenase
AA278755	CEP27	Centrosomal protein
AI261833	SLC7A9	Solute carrier family 7 (cationic amino acid transporter, y+ system) member 9
H72802	ESPN	Espin
AA608713	C1QDC1	C1q domain containing 1
AA047465	SLC6A8	Solute carrier family 6 (neurotransmitter transporter, creatine), member 8

Figure 4.

Analysis of the cross-species conserved signature as a tool to separate hereditary polyposis syndromes due to APC (FAP) or MYH (MAP) germline mutations. The globaltest32 was performed with the 49 (a)- and 166 (b)-gene signature and graphically represented by box plots of the P values generated after 1000 iterations in which only one random sample from each patient was used at a time. Box plots were generated using the standard settings present in R2.4.1. The filled blue boxes encompass the range of P values representative of 50% of the data points, whereas the central line represents the median. two-dimensional hierarchical clustering analysis was performed with the 49 (c)- and 166 (d)-gene signature, respectively, on the expression profiles obtained from all 56 colorectal adenomas (42 from APC and 14 from MYH gene mutation carriers). The colored bar above the heat map represents the mutation status of the corresponding polyp samples: red, polyps from patients carrying germline APC mutations; blue, polyps from patients carrying bi-allelic germline MYH mutations.

Discussion

Expression profiling by oligonucleotide and cDNA microarray platforms has rapidly become a commonly used tool for the qualitative and quantitative evaluation of the genome-wide transcriptional activity of human cancers. However, the outcomes of expression profiling of cancers are often very complex as they reflect the heterogeneity of cell types and biological activities present within the neoplastic mass, thus making their functional interpretation a difficult task. This is certainly the case for the expression (and genomic) profiles obtained to date from colorectal cancers. Although several studies have been published in the scientific literature, the degree of overlap between independent data sets is limited, possibly also as a consequence of differences in patient cohorts and methodologies used.16 Cross-species comparison of cancer profiling data represents a valuable approach to i) decrease the complexity of omics signatures, ii) pinpoint conserved target genes more likely to play rate-limiting functional roles in tumor initiation and progression to malignancy, and iii) accelerate the development of tailor-made anticancer therapies.46,47

Notwithstanding the above-mentioned heterogeneity, colorectal cancer represents, at least from a genetic perspective, a relatively homogeneous disease as the vast majority of the sporadic cases is known to be triggered by somatic mutations at the APC or CTNNB1 (β-catenin) genes, leading to the constitutive activation of the canonical Wnt signaling pathway.10 These mutations are known to initiate the formation of aberrant crypt foci and adenomatous polyps, the earliest benign precursors of the adenoma-carcinoma sequence. Also, germline APC mutations underlie FAP, an autosomal dominant predisposition to the development of multiple adenomatous polyps throughout the colon-rectum.48 The availability of a unique collection of adenomatous polyps obtained from FAP patients carrying germline APC mutations and from a mouse model, Apc^+/1638N, carrying a targeted mutation in the endogenous Apc gene allowed us to perform the cross-species computational comparison of their gene expression profiles and derive a conserved 166-gene signature. It should be noted that whereas FAP patients mainly develop polyps in the colon-rectum, Apc mouse models are characterized by adenomas clustering in the upper gastrointestinal tract, mainly in the duodenum. This anatomical difference between the mouse and human adenomas used for the cross-species comparison may exert a confounding effect in our computational analysis as duodenum and colon-rectum represent distinct organs. However, it may also confer an additional advantage to our approach as tissue-specific differences between the two GI tracts are likely to be filtered out, thus retaining only those conserved differentially expressed genes more likely to play functional roles in intestinal tumor formation, regardless of anatomical sub-location. The same holds true for our methodological approach: different microarray platforms were used to derive the human (cDNA arrays) and mouse (oligonucleotide arrays) gene profiles. In both cases, laser-guided microdissection was used to enrich in tumor cells without the confounding effects of contaminating stromal cells. Overall, our IHC analysis of a subset of proteins encoded by the conserved genes confirmed their differential expression between normal tissue and adenomas in both species (Figure 3), thus validating our methodological strategy.

The significance thresholds used to generate the differentially expressed lists of genes for the human (approximately 10% of the represented genes, with FDR = 0.5%) and mouse (approximately 18% of the represented genes, with FDR = 5%) studies are admittedly arbitrary. They were chosen using two generic criteria: i) the gene lists would be representative of the differential signature without encompassing an excessive percentage of false positives, and ii) the resulting conserved list of differentially expressed genes would be sufficiently large to enable pathway analysis. The presence in our cross-species signature of several genes known to be differentially expressed in sporadic colorectal cancers16 also represents an indirect confirmation of the general validity of our computational approach.

GO-based functional analysis of the 166 conserved genes reveals a general increase in cell division as shown by the up-regulation of genes related to DNA replication and repair, cell cycle regulation, and the maintenance of genomic integrity (Table 3). Notably, exclusively up-regulated genes were encompassed within these categories, indicative of the increased proliferation rate of tumor cells when compared with normal ones. Genes belonging to the transcriptional and translational machinery were also up-regulated when compared with normal tissues. These included genes involved in ribosome biogenesis, mRNA synthesis and maturation, and protein synthesis and folding (Table 3).

As expected from our selection of adenomas from APC-mutant patients and mouse models, several members of the Wnt signal transduction pathway are included among the conserved 166 differentially expressed genes (Figure 2). These included the Frizzled receptor homolog FZD6, the protein phosphatase type 2A (PP2A), the HMG box transcription factor SOX4,49 and several Wnt downstream transcriptional targets, namely, the matrix metallopeptidase matrilysin (MMP7),36,37,50 CD44,36 ENC1 (ectodermal-neural cortex 1)51, ephrin receptor B3 (EPHB3),52 cyclin D1 (CCND1),53,54 and the apoptosis inhibitor survivin (BIRC5)55,56 (Figure 2). However, AXIN2, a well known downstream Wnt target gene, is not included in this list simply because its probe is not encompassed by the human cDNA array. Other known Wnt target genes such as EPHB2, SOX9, and MYC were excluded because of the high stringency of the statistical thresholds used or to their absence in one of the platforms. Recently, an “intestinal Wnt/TCF4 signature” was obtained by integrating expression profiling data from CRC cell lines engineered with an inducible block of Wnt signaling and from sporadic human adenomas and carcinomas.11 Comparison of this 208-gene Wnt/TCF gene signature with our cross-species conserved list revealed 10 common entries, 4 of which belong to the Wnt/β-catenin signaling pathway (CD44, ENC1, EPHB3, and SOX4). The latter is not surprising in view of the different computational approaches and tumor cohorts used. Moreover, the use of CRC cell lines with dominant negative TCF4 constructs does not necessarily mimic the initial and rate-limiting loss of APC function characteristic of the mouse and human adenomas used in our cross-species analysis. Of more interest is the comparison with the study by Kaiser et al57 in which a cross-species comparison was performed among human and mouse intestinal tumors together with mouse embryonic stages of intestinal development. As depicted in Supplemental Table S5 (see http://ajp.amjpathol.org), the overlap between the two studies is high, with 46 of 166 differentially expressed genes shared between the data sets. Notably, the overlap is considerably higher with genes showing similar behavior in intestinal tumorigenesis and embryonic development.

Apoptosis inhibition in the adenomas, as suggested by BIRC5 up-regulation, is also strengthened by the conserved down-regulation of the BAD gene, encoding for a potent pro-apoptotic protein. BAD forms heterodimers with BCL2 and BCL-XL, thus repressing their anti-apoptotic function.58,59

Two members of the TGF-β signaling pathways are up-regulated among the cross-species conserved genes, namely SMAD6 and TGFBR2. The TGF-β ligand mediates its effects through the transmembrane type I (TGFBR1) and type II receptor subunits (TGFBR2), and in the cytoplasm through stimulatory and inhibitory SMADs. The up-regulation of the TGFBR2 gene encoding for the type II receptor is remarkable in view of its frequent mutational inactivation in a substantial proportion of sporadic colon cancers.60 The SMAD6 gene encodes for an inhibitory SMAD protein that becomes up-regulated as the result of a negative feedback loop. SMAD6 is thought to represent a key component in the integration of signals from different pathways and was shown to exert BMP inhibitory activity.61 Down-regulation of the bone morphogenetic BMP2 gene apparently confirms the inhibition of this TGF-β-related pathway. Although its role in tumorigenesis is yet unclear, SMAD6 up-regulation has been reported in other tumor types.62 Overall, the conserved gene signature is indicative of the activation of TGF-β and inhibition of BMP signaling at early stages of intestinal tumorigenesis. However, this observation needs to be validated by additional expression and reporter assays.

Among the many genes encompassed by the cross-species conserved signature, the up-regulation of ANXA1 is of interest in view of its phospholipase A2 (PPA2) inhibitory activity, an enzyme involved in the synthesis of prostaglandins during inflammation.42 Antibodies against annexin A1 have been found in patients with inflammatory bowel disorders.38 Also, its up-regulation was shown to occur in mitogenically stimulated cells in a PKC phosphorylation-dependent fashion, accompanied by its translocation from the cytoplasm to the nucleus.41 Notably, changes in ANXA1 subcellular localization were also observed in our IHC validation analysis (Figure 3).

Apart from its implications for the understanding and elucidation of the molecular and cellular mechanisms underlying APC-driven intestinal tumor formation, the cross-species conserved gene signature may also represent a useful tool to discriminate among hereditary polyposis patients with distinct genetic defects. Expression profiling analysis of the additional set of 14 colorectal adenomas obtained from patients carrying bi-allelic mutations at the MYH gene showed a high degree of similarity with the APC profiles. This could be explained by previous observations, according to which the APC gene is a preferential target for somatic mutations in colorectal adenomas from carriers of bi-allelic MYH germline mutations.63 The observed high degree of similarity between expression profiles from FAP and MAP polyps could then be explained provided that the somatic APC mutation does represent the initiating event in MYH-associated polyp formation. Alternatively, human adenoma profiles may be similar notwithstanding the initiating genetic defect, as indicated by our own most recent results with the expression analysis of three polyposis patients of unknown genetic basis (and no germline mutations found after sequencing of the MYH and APC genes). Also in these cases, the resulting profiles were virtually indistinguishable from those derived from MYH- and APC-mutant polyps (data not shown).

Nevertheless, by applying an FDR threshold of 0.5%, we could generate a 49-gene signature based on differences between MYH- and APC-mutant human polyps. Yet, both globaltest and two-dimensional hierarchical clustering analyses showed that the conserved 166-signature clusters more accurately the expression profile data from FAP and MAP patients than does the 49-gene signature (Figure 4).

In conclusion, cross-species comparison of expression profiles of intestinal adenomas obtained from hereditary polyposis patients and mouse models carrying germline APC mutations resulted in a signature of 166 differentially expressed genes. Functional annotation of the conserved genes indicates an overall increase in cell division and the up-regulation of the Wnt/β-catenin signaling pathway. These main cellular and molecular changes are accompanied by a plethora of gene-specific changes yet to be tested by functional assays to determine their relative contribution to intestinal tumor formation. Additional validation on independent polyp cohorts and further fine-tuning of the conserved gene signature are needed toward the development of an expression-based assay to classify hereditary polyposis syndromes.