The RXR-selective retinoid used in this study bexarotene (LGD1069, Targretin) was obtained from Ligand Pharmaceutical, Inc (San Diego, CA).

Female MMTV-erbB2 mice 8, (obtained from The Jackson Lab., Bar Harbor ME) and C3(1)/SV40 T-antigen strain mice 9 (obtained from The National Cancer Institute, Frederick, MD) were housed in the institutional animal facilities. Balb/c p53-null mammary epithelium transplanted into the cleared mammary fat pads of three-week old mice p53 wt Balb/c mice 10 were initiated and maintained at BMC. Each group included age-matched untreated controls and bexarotene-treated mice. All mice were treated 6 days/week during 2 months starting at 8 weeks of age with bexarotene suspended in purified sesame oil (Croda, Inc., Mill Hall, PA). The retinoid was administered by gastric gavage using a 20-gauge gavage needle in a volume of 0.1-ml containing vehicle 100 mg/kg of bexarotene. Virgin animals were used to avoid confounding effects of hormonal surges during pregnancy. All animal research was conducted in AAALAC accredited facilities, following international guidelines and all research was approved by the corresponding Institutional bioethics committees.

The six mouse SAGE libraries were generated following standard procedures as described previously 11. Briefly total RNA was extracted from frozen samples using TRIzol (Invitrogen, Carlsbad, CA, USA). SAGE library construction was performed with the I-SAGE kit (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's protocol and introducing only minor modifications. The anchoring enzyme was NlaIII and the tagging enzyme used was BsmFI. Concatemerized ditags were cloned into pZERO-1 and sequenced with an ABI 3700 DNA Analyzer (Applied Biosystems, Foster City, CA, USA). To decrease the chances potential artifacts due to sample heterogeneity, each control or bexarotene treatment SAGE library represents a pool of three mammary epithelial samples from three age-matched separate mice. For the studies on the p53-null mammary cancer model we used mammary epithelial enriched preparations as previously described 12, for the MMTV-erbB2 and C3(1)/SV40 T-antigen models we used total mammary gland preparations. SAGE libraries were generated at an approximate resolution of 60,000 SAGE tags per library.

SAGE tag extraction from sequencing files was performed by using the SAGE2000 software version 4.0 (a kind gift of Dr. Kenneth Kinzler, John Hopkins University, Baltimore, MD). SAGE data management, tag to gene matching, as well as additional gene annotations and links to publicly available resources such as Gene Ontology (GO), UniGene, and Entrez gene ID, were performed using a suite of web-based SAGE library tools developed by us. In our analyses we only considered tags with single tag-to gene reliable matches. To compare the control (vehicle) vs. bexarotene treatment SAGE libraries in each transgenic mice model, we utilized the Audic and Claverie's significance test 13. Statistical analysis and scatter plot visualization of SAGE libraries were done with the Discovery Space 4 software (Genome Science Centre, BC Cancer Agency, Canada, Vancouver) http://www.bcgsc.ca/platform/bioinfo/software/ds.

The main strategy of this analysis was to identify commonly deregulated genes by bexarotene treatment among the different mammary cancer models tested (Figure 1). Differentially expressed genes were compiled into one Excel spreadsheet pivot Table for comparison of overlapping data between p53-null, MMTV-erbB2 and C3(1)/SV40 T-antigen transgenic mouse mammary models. Any combination of two lists was compared for matching gene-identity. The number and identity of genes commonly affected in two models (e.g. MMTV-erbB2 vs. p53-null) was determined. We used the normal approximation to the binomial distribution as previously described 14 to calculate whether the number of matching genes derived from each pairwise comparison was of statistical significance (p < 0.05). To enable illustration of the commonly deregulated genes between mammary cancer models, we used the TIGR MultiExperiment Viewer (MeV 3.0) software. This tool was used for average clustering of SAGE based on the fold change of tag counts for each transcript comparing bexarotene treatment to control (vehicle) in each transgenic mice mammary model. For automated functional annotation and classification of genes of interest based on Gene Ontology (GO) terms, we used the EASE 15 available at the Database for Annotation, Visualization and Integrated Discovery (DAVID) 16. All of the raw SAGE data reported as additional files in this article are publicly available and also can be viewed at http://sciencepark.mdanderson.org/labs/ggeg/SAGE_Proj_14.htm.

In order to identify the molecular pathways that are mainly affected by the rexinoid bexarotene, we look for protein/gene interaction networks in the common core of modulated genes. The protein-protein interaction network associating genes of the three transgenic mouse mammary models was generated using the database STRING ('Search Tool for the Retrieval of Interacting Genes/Proteins') http://string.embl.de/17. The database STRING aims to collect, predict and unify most type of protein-protein associations, including direct and indirect associations. STRING runs a set of prediction algorithms, and transfers known interactions from model organisms to other species based on predicted orthology of the respective proteins 18. In order to identify each gene in the database, we used both mouse gene name and Entrez gene ID in the 'protein-mode' application. The analysis input options were 'co-occurrence', 'co-expression', 'experiments', 'databases', and 'text mining' data at high confidence level of predicted human orthology groups. Pathways are discriminated by different colors based on up-modulated (red node) or down-modulated (green node) transcripts in order to indicate protein-protein networks modulated by bexarotene across mammary cancer models.

A common observation in cancer gene expression profiling is the systematic up-regulation of ribosomal genes among the most abundant transcripts in human and mouse mammary carcinomas compared with normal tissues 2021222324. The up-regulation of ribosomal genes was significantly correlated with variation in the cell doubling time in vitro, supporting the notion that these genes are up-regulated in relation to the increase of cell proliferation rate or growth rate during malignant transformation. Interestingly and in an opposite manner, bexarotene treatment cause in 'normal' mammary gland the down-regulated expression of more than 10 genes related to protein biosynthesis including numerous ribosomal proteins (Rpl19, Rpl37, Rps4x, Rps8, Rps24, Rps27, Rps29), Eef1b2 (Eukaryotic translation elongation factor 1 beta 2), Eif2s3x (Eukaryotic translation initiation factor 2), Fau (Finkel-Biskis-Reilly murine sarcoma virus) and Tpt1 (tumor protein, translationally-controlled 1). The inhibition of mRNA synthesis for genes encoding ribosomal proteins has been suggested as a mechanism that could reprogram the cancer cell to recover some of its normal functions in a tumor reversion process 25.

Tpt1 (also known as Tctp) encodes a GDP dissociation inhibitor protein of the translation elongation factor eEF1A 26. The human TPT1 gene is overexpressed in cancerous cell lines compared with cell lines derived from normal tissues. Tuynder et al. (2002) demonstrated that the expression levels of TPT1 were strongly down-regulated at the mRNA and protein levels during tumor reversion/suppression. MCF7 and T47D cell lines transfected with Tpt1 siRNA showed a more organized ductal-like structures similar to those generated by down-regulation of β1 integrin 25. Here we observed that bexarotene significantly dowregulated expression of Tpt1 in mammary epithelium.

More than 50 years ago, Warburg proposed that malignant phenotype might be caused by a decrease in mitochondrial energy metabolism paralleled by increased glycolytic flux 27. Increasing evidence is in line with this hypothesis suggesting a close link between metabolic and genetic changes observed during malignant growth 2829. Recently it has been demonstrated that impaired bioenergetic function of mitochondria is a hallmark of carcinogenesis in breast, gastric, lung and oesophageal cancer 3031. Moreover, Schulz et al. (2006) showed that induction of mitochondrial oxidative metabolism efficiently suppresses malignant growth in vitro and in vivo. Interestingly, we identified a systematic up-regulation of transcripts related to oxidative phosphorylation induced by bexarotene treatment in mammary gland (Figure 4). The transcripts commonly up-regulated by bexarotene treatment in at least two of the models were Atp5b (ATP synthase F1 complex beta subunit), Atp5e (ATP synthase F1 complex epsilon subunit), Cyc1 (Cytochrome c-1), Cycs (Cytochrome c somatic), Cox5b (Cytochrome c oxidase, subunit Vb), Cox7b (Cytochrome c oxidase subunit VIIb), Cox8a (Cytochrome c oxidase, subunit VIIIa), Ndufa1 (NADH dehydrogenase 1 alpha subcomplex), Ndufc1 (NADH dehydrogenase 1), Ndufb8 (NADH dehydrogenase 1 beta subcomplex 8), Ndufa11 (NADH dehydrogenase 1 alpha subcomplex 11) and Uqcrh (Ubiquinol-cytochrome c reductase hinge protein) (see Additional files 1 and 2). Consistent with a significant increase of oxidative phosphorylation enzymes, we observed that Atpif1 gene (ATPase inhibitory factor 1) was significantly down-regulated by bexarotene treatment in the MMTV-erbB2 and C3(1)/SV40 T-antigen transgenic mice models. In this sense, Isidoro et al. (2005) showed that down-regulation of ATPase β-F1 per se allowed the identification of a subgroup of breast cancer patients with significant worse prognosis. Finally, is important to note that mitochondrial oxidative phosphorylation is required for efficient execution of apoptosis. Cells which are unable to carry on oxidative phosphorylation have a resistant apoptotic phenotype 32. Overall, these findings suggest the oxidative phosphorylation induction (prevention impaired bioenergetic function) as a novel mechanism of bexarotene's chemopreventive effects.

Lipid metabolism and the intracellular transport of bioactive species is a critical component in the process by which these molecules continuously stimulate proliferation through interactions with nuclear receptors. Bexarotene treatment of MMTV-ErbB2 and C3(1)/SV40 transgenic mammary gland up-regulated various genes related with lipid/fatty acid metabolism (Figure 4) such as: Fabp3 (Fatty acid binding protein 3), Fabp4 (Fatty acid binding protein 4), Dgat2 (Diacylglycerol O-acyltransferase 2), Dbi (Diazepam binding inhibitor), Hadhb (Hydroxyacyl-Coenzyme A dehydrogenase), Acca2 (Acetyl-Coenzyme A acyltransferase 2), and Cpt2 (Carnitine palmitoytransferase 2). Interestingly, a family of cytoplasmic proteins known as FABPs mediates transport and utilization of lipids, and different FABP types have been implicated in control of cell proliferation and cancer progression. Recently, FABP1 and FABP2 were shown to be up-regulated in breast cancer cell lines while FABP3 and FABP4 were down-regulated in breast cancer cells 33. Moreover, FABP4 is a marker protein for differentiated mammary gland that is expressed only in normal lactating cells and not in tumor mammary cells. Transfection of cDNA clone of FABP4 into MCF7 cells results in growth inhibition and lower tumorgenicity in nude mice 34.

Rexinoid bexarotene down-regulated several genes related with cell cycle/proliferation in mammary gland from the different transgenic mice models (Figure 2A; see Additional files 1 and 2).

Among this functional group we find Npm1 (also known as Nucleophosmin/B23) protein that belongs to a nuclear chaperone family of phosphoproteins that take part in various cellular processes such as cell proliferation and transformation 35. Human NPM1 is overexpressed in various tumors types, and it has been proposed as a marker for gastric, colon, ovarian and prostate carcinomas 35. NPM1 overexpression promotes cell survival in several cell types through the inhibition of distinct pro-apoptotic pathways 36. We detected a systematic down-regulation of Npm1 gene in mammary gland from Bexarotene treated mice on the three models studied (average fold change = -2.6; p < 0.01). Interestingly, proteomic analyses identified NPM1 as a protein associated with acquired estrogen-independence in human breast cancer cells 37. Moreover, down-regulation of NPM1 mRNA delay cell-cycle progression and the entry into mitosis 38, whereas NPM1 overexpression decreases the sensitivity of human leukaemia cells to retinoic-acid-induced differentiation and apoptosis 3940.

Another gene in this category includes Stmn1 which encodes an 18 kDa cytosolic phosphoprotein (also known as Stathmin 1 or Oncoprotein 18) that is regulated during cell cycle by transcriptional and posttranscriptional mechanisms. STMN1 overexpression has been demonstrated at mRNA and protein levels in a significant proportion of human breast carcinomas (about 30%) 41. Moreover, STMN1 overexpression was correlated with the loss of ERα and with histological grade III breast carcinomas. STMN1 has been suggested as a key regulator of the cell division through its influence on microtubule dynamics. We identified a statistical significant decrease of Stmn1 expression (average fold change = -5.4; p < 0.05) caused by bexarotene treatment in mammary gland from MMTV-erbB2 and C3(1)/SV40 T-antigen mice. Interestingly, we previously demonstrated that mouse Stmn1 and human homologue STMN1 genes are overexpressed in invasive breast carcinomas by northern and real time RT-PCR analyses 24.

Numerous studies have shown that down-regulation of p27^Kip1, an inhibitor of cyclin-dependent kinase, is associated with poor prognosis in many cancers such as: breast, colorectal, prostate, and lung carcinomas. We previously detected the overexpression of CDC28 protein kinase regulatory subunit 1B (Cks1b) in human and mouse mammary tumors 24. Interestingly, rexinoid bexarotene strongly down-regulated Cks1b expression in the MMTV-erbB2 model (Fold change = -10.0; p = 0.006) (see Additional file 1). Human CKS1B functions as an important adaptor of SCF Skp2 ubiquitin ligase and facilitates SCF Skp2 targeting of the cell proliferation inhibitor p27 (Kip1) for ubiquitination and subsequent degradation 42. It was also suggested that CKS1B may be involved in p21 degradation in a similar fashion 43. Overexpression of CKS1B has been observed associated to poorly differentiated tumors (histological grade III) and with the loss of ER/PR status [Slotky et al., 2005]. In addition, CKS1B overexpression was strongly and independently associated with poor overall survival in human breast cancer 44.

On the other hand, bexarotene treatment up-modulated two apoptotis related genes (Cidec and Cycs) in 'normal' mouse mammary gland from two of the models (Cidec) and in all three models (Cycs) (Figure 3). Cidec (also known as Fsp27) encode a novel family member of the cell-death-inducing DFF45-like-effectors (CIDEs) 45. Although, its well known that DFF45 is a subunit of the DNA fragmentation factor that is cleaved by caspase-3 during apoptosis 46. The molecular mechanism by which Cidec induces apoptosis remains to be elucidated.

During their metastatic conversion, epithelial cells acquire the ability to invade the surrounding tissues and later disseminate to secondary organs mostly via lymphatic vessels. Epithelial cell adhesions, including intercellular (junctional) and cell-extracellular matrix adhesions, are critical to the maintenance of structural integrity, polarity, and cell-cell communication. We detected a significant decrease in Cldn3 (Claudin 3) (Average fold change = -6), Glycam1 (Glycosylation dependent cell adhesion molecule 1) (Average fold change = -7), Pscdbp (Pleckstrin homogy Sec7 binding protein) (Average fold change = -6) gene expression modulated by bexarotene treatment among transgenic mice models. The Claudin genes (Cldn) encode a family of proteins important in epithelial cell tight junction, which are critical to the maintenance of cell polarity and permeability 4748. Most Claudin genes appear with decreased expression in cancer however CLDN3 and CLDN4 genes have been found frequently up-regulated in ovarian, breast, prostate and pancreatic cancers 49505152. Recently, has been suggested that Claudins may be envolved in survival and invasion of cancer cells 48. We detected down-regulation of Cldn3 gene in mammary gland from bexarotene treated mice in the MMTV-erbB2 and C3(1)/SV40 models. The role of Gycam1 and Pscdbp genes in breast cancer progression remains unknowns.