Global Gene Expression Profiling of Chemically Induced Rat Mammary Gland Carcinomas and Adenomas

Abstract

Chemical carcinogens induce both benign and malignant mammary gland tumors in female Sprague-Dawley rats. To identify gene expression profiles associated with malignancy, cDNA microarray analysis was used to compare gene expression profiles in rat mammary gland carcinomas, adenomas, and normal mammary gland. Tumors were induced with various chemical carcinogens including 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP), 2-amino-3,8-dimethylimidazo[4,5- f]quinoxaline (MeIQx), 7-12-dimethylbenz[a]anthracene (DMBA), N-nitrosomethylurea (NMU), and 4-aminobiphenyl. The global gene expression profiles in carcinomas and adenomas were distinguishable by hierarchical clustering and multi-dimensional scaling analyses. Permutation analysis revealed 110 clones statistically differentially expressed between benign and malignant tumors (p < 0.0005). Carcinomas showed relatively high expression of several genes associated with mammary epithelial cell growth and proliferation (e.g., cyclin D1, PDGFα) and relatively low expression of differentiation marker genes (e.g., β-casein, whey acidic protein, transferrin). Other categories of genes showing differential expression between carcinomas and adenomas were associated with protein homeostasis, cytoskeleton, extracellular matrix, and cell metabolism (fatty acid metabolism, oxidative phosphorylation, and glycolysis). Major gene families implicated in malignancy by over-expression in carcinomas included the annexins (annexin A1 and A4) and Stat family of transcription factors (Stat3 and Stat5a). The elevated expression of the prolactin receptor in carcinomas concomitant with several components of the mitogenic prolactin signaling pathway implicated prolactin/prolactin receptor/Stat5a/cyclin D1 in rat mammary gland malignancy.

Introduction

Chemically induced rat mammary gland cancer provides an experimental system for better understanding the molecular alterations and pathogenesis associated with human breast cancer (Russo et al., 1990; Russo and Russo, 1996). Rat mammary gland carcinomas mimic human breast cancer in histopathology, epithelial cell of origin, and hormone dependency for tumorigenesis (Russo et al., 1990). In addition, a growing number of recent studies support similarities between human and rat mammary gland cancers in molecular alterations, for example, in over-expression of cyclin D1, ErbB2, PDGFα and Stat5a (Wang et al., 2001; Shan et al., 2002, 2004; Qiu et al., 2003).

One interesting aspect of the rat mammary gland cancer model is that chemical carcinogens induce benign tumors as well as carcinomas (Russo et al., 1990; Russo and Russo, 1996). As is found in the human breast, benign tumors in rats are largely tubular adenomas and fibroadenomas. In the chemical carcinogen-treated rat model, benign tumors are believed to arise from alveolar buds and lobules. In contrast, carcinomas emerge from epithelial cells residing in less differentiated structures such as terminal end buds and terminal ducts that progressively form intraductal proliferations, ductal carcinoma in situ and invasive carcinoma (Russo et al., 1990, 1996; Russo and Russo, 2000, Thompson et al., 2000). Although various types of lesions can be found in the rat mammary gland, the predominant histologic types of carcinoma in chemical carcinogen treated rats are papillary and cribriform carcinomas (Russo and Russo, 2000; Thompson et al., 2000; Costa et al., 2002).

In previous microarray studies we and others observed that 2-amino-1-methyl-6-phenylimidazo[4,5-b]pyridine (PhIP)-and 7-12-dimethylbenz[a]anthracene (DMBA)-induced carcinomas harbored common molecular alterations as well as etiology-specific alterations (Kuramoto et al., 2002; Shan et al., 2002). We hypothesized that the common gene expression alterations may represent major molecular pathways of malignant transformation in the rat mammary gland. Recently this laboratory examined the etiologically distinct gene expression profiles in rat mammary gland carcinomas induced by several different chemical carcinogens including PhIP, DMBA, 2-amino-3,8-dimethylimidazo[4,5-f]quinoxaline (MeIQx), N-nitrosomethylurea (NMU), and 4-aminobiphenyl (4ABP) (Shan et al., 2005). However, the common molecular alterations were not previously examined. Herein microarray analysis was used to determine the common molecular alterations in rat mammary gland carcinomas induced by five different carcinogens. Furthermore, these alterations were compared to gene expression profiles in adenomas to discern the possible molecular pathways associated specifically with malignancy. These studies were undertaken to provide insight into the molecular signaling pathways deregulated in rat mammary gland carcinogenesis.

Methods

Carcinogen-Induced Rat Mammary Gland Tumors

Archival samples of chemically-induced female Sprague-Dawley rat mammary gland carcinomas and adenomas were used that had been stored at −80°C (Ghoshal et al., 1994; Snyderwine et al., 1998; Shan et al., 2002, 2005). Briefly, PhIP, MeIQx, and 4ABP were administered to adolescent rats at 75 mg/kg, p.o., once per day for 10 days over a 12-day period. DMBA and NMU were administered as a single dose at 75 mg/kg, p.o., and 50 mg/kg, i.p., respectively, to 50-day old rats. The 34 carcinomas were induced by PhIP (N = 8), DMBA (N = 8), 2-amino-3,8-dimethylimidazo[4,5- f]quinoxaline (MeIQx) (N = 4), N-nitrosomethylurea (NMU) (N = 9), and 4-aminobiphenyl (4ABP) (N = 5). These carcinomas were histologically classified in our previous study and were cribriform and papillary carcinomas of various histologic grades (Shan et al., 2005). Nine adenomas were obtained from PhIP-treated rats including fibroadenomas (N = 4), tubular adenomas (N = 4) and a lactating adenoma (N = 1). Normal mammary gland (glands 4–6) was obtained from control rats (N = 12) that were vehicle, age and diet matched. Control tissue had been carefully excised to exclude excess fatty tissue, and the lymph nodes removed prior to snap freezing in liquid nitrogen and storage at −80°C.

cDNA Microarray and Hybridization

Total RNA was isolated using TRIzol extraction reagent (Invitrogen, Rockville, MD) and purified using RNease MinElute Cleanup Kit (Qiagen, Valencia, CA). The mouse cDNA microarray, printed by the National Cancer Institute (NCI), National Institutes of Health (NIH), contained 9984 cDNA clones 〈http://nciarray.nci.nih.gov〉. The target preparation and hybridization protocol used were the same as those described on the web site 〈http://nciarray.nci.nih.gov/reference/index.shtml〉. Thirty micrograms of total RNA were used to synthesize cDNA that was labeled by Cy^™ 3 (Cy3) and Cy^™ 5 (Cy5) mono-reactive dyes (Amersham Biosciences, Buckinghamshire, England) using Fairplay Microarray Labeling Kit (Stratagene, La Jolla, CA). The labeled Cy3 and Cy5 cDNA probes were purified with a MinElute PCR Purification Kit (QIAGEN, Valencia, CA). Hybridization was carried out overnight at 42°C. The arrays were scanned on an Axon GenePix 4000 scanner equipped with GenePix Pro 3.0 software (Axon, Union City, CA). For data normalization, interpretation and visualization, the image and raw data were deposited into the NCI microarray database system supported by the Center for Information Technology of NIH 〈http://nciarray.nci.nih.gov〉. Pooled mammary gland RNA from 12 control rats was used as a common reference in all microarray experiments. RNA from control mammary gland was isolated from whole gland that was archival and stored at −80°C. The tumor and control tissue were therefore handled in the same manner. It is notable that adipose tissue is the major stromal contaminant of control virgin rat mammary gland, constituting 70% of the gland by weight (Wrenn et al., 1965). Fat was effectively removed from the control sample during RNA isolation by phenol-chloroform extraction.

Microarray Data Analysis

Statistical analysis of the microarray data was performed using BRB-ArrayTools software (version 3.2), an integrated package for the visualization and statistical analysis of cDNA microarray gene expression data (Richard Simon and Amy Peng Lam, Biometric Research Branch, NCI, 〈http://linus.nci.nih.gov/~brb/〉. The procedures for median normalization and clone spot selection were the same as previously described (Shan et al., 2005). The clone spots were excluded from analysis if any of the following was observed: the spot diameter was less than 10 μm; both red (Cy5) and green (Cy3) intensities of the spot were below 100; the truncate intensity ratio was greater than 64; if less than 20% of the data showed more than a 1.5-fold change in either direction from spot intensity median value across all arrays. The criteria for spot inclusion guaranteed a high quality of spots for analysis and decreased the potential influence of clones with little or no expression among the RNA samples.

Unsupervised hierarchical clustering was performed to analyze the global gene expression profiles of all tumors. Supervised hierarchical clustering and multidimensional scaling (MDS) were performed for the comparative analysis of benign and malignant tumors. Both agglomerative hierarchical clustering and MDS were carried out by centered correlation and average linkage. Univariate permutation analysis (F-test) was used to identify genes that were differentially expressed between benign and malignant tumors. The false discovery rate was controlled by use of multivariate permutation test (confidence level of false discovery rate assessment, 95%; maximum permitted number of false-positive genes, 10). The number of permutations for the univariate and multivariate tests was 2000.

Semiquantitative RT-PCR

One microgram of total RNA was used to synthesize the first strand cDNA in a final volume of 20 μl using SuperScript First-Strand Synthesis Kit (Invitrogen). cDNA (0.5 μl) was used to separately amplify a segment of β-casein, κ-casein, PDGFα, prolactin receptor long form (PR-LF), signal transducer and activator of transcription 3 (Stat3), signal transducer and activator of transcription 5a (Stat5a), cyclin D1 and retinoblastoma binding protein 4 (Rbp4, also known as RbAp48), annexin A1 and activated leukocyte cell adhesion molecule (alcam) separately using the following specific primers: β-casein-F: aggatgcattcactgtgtc, β-casein-R: aggtcttgaacaggcatac; κ-casein-F: agaactgactccgtgtgaag, κ-casein-R: accactgactctgtggtag; PDGFα-F: agtcagatccacagcatc, PDGFα-R: tcctgacatactccact; PR-LF-F: gcaagccagaccatggatac, PR-LF-R: gcagtaatggccatgttgaa; Stat3-F: atcaagcagttcctgcagag, Stat3-R: tctgaacagatccacgatc; Stat5a-F: tgcagaagaaggcagaacac, Stat5a-R: agcttcttcagctcgctct; cyclin D1-F: tgcttaagactgaggagacctg, cyclin D1-R: tggagaggaagtgttcgatga; Rbp4-F: agtggcttccagatgtga, Rbp4-R: agactcatggagcagatg; annexin A1-F: tgaagggacttggaacag, annexin A1-R: agttccagcacccttcatg; alcam-F: tcggatggtacactgtcaac, alcam-R: acaactgagtacctggatc. GAPDH was amplified as the internal control using primers F-accacagtccatgccatcac and R-tccaccaccctgttgctgta.

There was little variation in the intensity of GAPDH across the samples. The PCR was carried out for 25–30 cycles with each cycle consisting of a denaturing step for 45 sec at 94°C, an annealing step for 45 sec at 56–60°C and a polymerization step for 45 sec at 72°C. The PCR product (10 μl) was separated on 2% agarose gel containing ethidium bromide (EtBr) and photographed under UV light. The band intensity of each sample was quantified using Kodak ID image analysis software (Eastman Kodak Co. Rochester, NY). The expression level of a specific gene was presented as the ratio of the specific gene band intensity relative to its corresponding GAPDH band intensity. The relative expression of a gene was calculated as a fold change in comparison to the average level in normal control samples. Statistical differences were assessed by the Student’s t-test (Sigma Stat 2.0, Jandel Corporation, San Rafael, CA).

Results

To determine the common molecular alterations among various chemically-induced rat mammary gland carcinomas, we first compared the gene expression profiles in carcinomas to the normal rat mammary gland by microarray analysis. Eighty-six known genes were found abnormally expressed in carcinomas based on the criteria that the change in expression was at least 2-fold different from the control mammary gland and detected in at least 55% of carcinoma samples (Table 1). Many under-expressed genes were milk protein-associated genes including β-casein, κ-casein, whey acidic protein, and transferrin. The lower expression of these genes is consistent with an undifferentiated status of carcinomas relative to the normal gland. The majority of known genes identified by microarray analysis were over-expressed in carcinomas. A large group of over-expressed genes were involved in cell growth and proliferation including the prolactin receptor, PDGFα, cyclin D1, Stat5a, RAN and cell division cycle 42 homolog. Among the genes showing elevated expression in carcinomas, many were protein synthesis/degradation related-genes such as ribosomal proteins and proteasomes. Twelve over-expressed genes as well as 3 under-expressed genes were associated with changes in cytoskeleton and extracellular matrix. A large set of cellular metabolism-related genes were further found over-expressed as well as 5 found under-expressed. Several of the over-expressed genes in this group were associated with fatty acid metabolism including fatty acid coenzyme A ligase (long chain 2), fatty acid binding protein 4, thyroid hormone responsive SPOT14 homolog, stearoyl-coenzyme A desaturase 2 and uncoupling protein 2 (mitochondrial).

We next compared the gene expression profiles in malignant tumors to benign tumors to further discern the genes associated with malignancy. Global differences in gene expression were clearly observed between malignant and benign tumors (Figure 1A). Based on 3353 genes that passed the criterion for selection, unsupervised hierarchical clustering revealed 3 groups of tumors. There were 2 groups that contained only malignant tumors, and 1 group that contained 7 carcinomas as well as all 9 benign tumors. Within this latter group there were further subgroups of tubular adenomas and fibroadenomas that were more tightly correlated. Although the benign tumors were all from PhIP treated rats, the 7 carcinomas that clustered with the adenomas were induced by DMBA, NMU, and PhIP (only one carcinoma). Clustering appeared to be largely influenced by histopathologic features in common between adenomas and carcinomas. For example, the PhIP carcinoma that subclustered with the fibroadenomas (at an approximate correlation of 0.2) also showed fibrosis. The 6 additional carcinomas that subclustered with the tubular adenomas were papillary, and 5 of these 6 carcinomas were well differentiated. Thus it appeared that certain histopathological features were reflected in the gene expression profiles of both adenomas and carcinomas. Tubular adenomas and fibroadenomas were further distinguishable by cluster analysis, and 92 genes were found to be statistically differentially expressed between these groups of tumors (p < 0.005, supplemental Table 1).

By univariate and multivariate permutation analysis, 110 clones were statistically differentially expressed between carcinomas and adenomas (p < 0.0005). From these differentially expressed genes, it was possible to completely resolve carcinomas and adenomas by supervised agglomerative hierarchical clustering and MDS (Figures 1B and 1C). Adenomas comprised a unique cluster with sets of genes under-expressed and over-expressed relative to carcinomas (Figure 1B). Known genes showing differential expression between adenomas and carcinomas at p < 0.0005 are shown in Table 2. (Supplemental Table 2 provides a more comprehensive list of 196 genes found statistically differentially expressed between carcinomas and adenomas at p < 0.001.)

Microarray data from selected genes of interest are shown in Table 3. Many genes associated with cell growth and proliferation were more highly expressed in carcinomas than in adenomas including the prolactin receptor, PDGF receptor family ligands PDGFα and Kit ligand, Stat family genes, and cyclin D1. Also up-regulated in carcinomas were the annexins A1 and A4, ubiquitous phospholipid and calcium binding proteins involved in a several cellular processes (Gerke and Moss, 2002). Differentiation-associated genes in the casein family showed statistically lower relative expression in carcinomas than in adenomas. Furthermore, multiple genes associated with cytoskeleton, cell adhesion and extracellular matrix were statistically differentially expressed between carcinomas and adenomas, with adenomas showing higher relative expression of several extracellular matrix genes. A large category of differentially expressed genes encoded ribosomal proteins, proteasomes, heat shock proteins, proprotein convertase and protein modification enzymes that were related to the protein quality control system (protein homeostasis) (Table 2). The majority of the differentially expressed cell metabolism-related genes were associated with oxidative phosphorylation and glycolytic energy production.

The differential expression of 10 genes (β-casein, κcasein, PDGFα, PR-LF, stat3, stat5a, cyclin D1, Rbp4, annexin A1 and alcam) was further verified by semiquantitative RT-PCR (Figure 2). Many of these genes were selected for RT-PCR analysis because they were found to be differentially expressed in carcinomas relative to normal mammary gland as well as statistically differentially expressed between carcinomas and adenomas by microarray analysis (compare Tables 1 and 2). Additional genes were selected that fell into similar gene families (e.g., casein and Stat families) or were known to regulate the Stat genes (e.g., prolactin receptor). In accordance with the microarray data, all 10 genes examined by RT-PCR showed a statistically significant difference between carcinomas and adenomas (Student’s t-test, p < 0.05, N = 16–34 carcinomas, 9 adenomas).

Discussion

Microarray analysis was used to examine the profile of gene expression across a panel of rat mammary gland carcinomas induced by various chemical carcinogens. In comparison to normal mammary gland, specific gene expression alterations were found in a majority of carcinomas irrespective of the etiologic agent. These common molecular alterations included many genes associated with cell growth and differentiation, protein homeostasis, cell signaling, cytoskeleton and extracellular matrix and cell metabolism. These studies extend previous studies that examined the genes differentially expressed in common between DMBA- and PhIP-induced carcinomas (Shan et al., 2002).

To further determine if gene expression alterations were associated specifically with malignancy, the rat mammary gland carcinomas were compared to adenomas by cluster analysis. Carcinomas and adenomas showed unique profiles of gene expression that were distinguishable by cluster and MDS analyses. Carcinomas showed relatively high expression of multiple genes associated with cell growth and proliferation, such as cyclin D1 and PDGFα, and markedly low expression of genes associated with mammary gland differentiation, such as the milk protein genes β-casein and whey acidic protein. Interestingly, some of the genes with altered expression in carcinomas also showed altered expression in adenomas, but the magnitude of the change was different. For example, the expression of cyclin D1, although elevated in carcinomas and adenomas (relative to control mammary gland), showed a 2-fold higher over-expression in carcinomas than adenomas (Figure 2, Table 3). The expression of β-casein and κ-casein was low in both carcinomas and adenomas relative to normal mammary gland, but was significantly lower in carcinomas than adenomas.

There were several classes of genes that stood out as distinctly differentially expressed between adenomas and carcinomas. One group of growth-related genes showing higher over-expression in carcinomas was the annexin family genes including annexin A1 and annexin A4. The annexin calcium-and phospholipid-binding proteins associate with the components of the cytoskeleton and mediate interactions between the cell and extracellular matrix. Studies support that annexins play a role in mitogenic signal transduction and cell transformation (Gerke and Moss, 2002; Wang et al., 2004). Annexin I has been found to be elevated in human as well as rat mammary adenocarcinoma (Pencil and Toth, 1998). Annexin A1 was first discovered as a substrate of the EGF receptor and is well-recognized to mediate the inflammatory response and effects of glucocorticoids (Gerke and Moss, 2002; Wang et al., 2004). Interestingly, annexin I also forms a heterotetramer with calcyclin (S100 calcium binding protein A6), another gene associated with proliferation and differentiation (Young et al., 1996; Hirata and Hirata 2002; Wang et al., 2004), that we found concomitantly up-regulated in rat mammary gland carcinomas. Calcyclin was also observed previously to be overexpressed in NMU-induced carcinomas (Young et al., 1996).

Another family of genes associated with rat mammary gland carcinomas was the Stat family of transcription factors. The Stat family of proteins are latent cytoplasmic transcription factors that convey signals from cytokine and growth-factor receptors to the nucleus (Yu and Jove, 2004). Both Stat3 and Stat5a showed over-expression specifically in carcinomas. Recently our laboratory reported an increased nuclear localization of Stat5a in PhIP- and DMBA-induced rat mammary gland carcinomas, suggesting that Stat5a is phosphorylated and transcriptionally active in the majority of carcinomas (Shan et al., 2004). Adenomas, however, show a statistically lower frequency of Stat5a nuclear staining than carcinomas (data not shown) suggesting that Stat5a activation and transcription of Stat5a regulated genes is lower in benign tumors. The findings from the current study are consistent with the conclusion that Stat5a plays a role in malignant transformation of the rat mammary gland, however, further studies are required to confirm this notion.

There is growing evidence to support the involvement of Stat3 and Stat5a in a variety of human malignancies including breast cancers (Watson, 2001; Bromberg, 2002; Yu and Jove, 2004). Stat3 and Stat5a participate in cell growth and survival and are a point of convergence for numerous oncogenic signaling pathways. Stat3 and Stat5a have been shown to directly or indirectly upregulate the expression of genes required for uncontrolled proliferation including cyclin D1, a gene overexpressed in rat mammary gland carcinomas (Wang et al., 2001; Brockman et al., 2002; Qiu et al., 2003). Interestingly, another overexpressed gene in rat mammary gland carcinomas was the prolactin receptor. Prolactin via the prolactin receptor is recognized to regulate growth and differentiation of the mammary gland via several pathways including the Jak2/Stat5a pathway (Vonderhaar, 1998; Clevenger et al., 2003). Prolactin is a well-recognized mitogen in breast cancer cells, and there is substantial evidence from the literature to support the central importance of prolactin and its receptor in rat mammary gland carcinogenesis (Welsch and Nagasawa, 1977; Vonderhaar, 1998). The up-regulation of the prolactin receptor concomitant with Stat5a in carcinomas implicates the prolactin/prolactin receptor/Jak2/Stat5a pathway in malignancy.

In addition to growth and differentiation genes, other broad categories of genes differentially expressed between carcinomas and adenomas were involved in protein homeostasis, cell signaling, cytoskeleton and extracellular matrix, and cellular metabolism. One notably large group of genes differentially expressed between the two tumor types were cytoskeleton and extracellular matrix genes. Adenomas were characterized by a relatively high expression of many of the genes in this category including sparc (osteonectin), smoc2, laminin alpha 2, procollagen (type VI, alpha 2) and the proteoglycans osteoglycin and lumican (Supplemental Tables 2 and 3). A previous study has reported that sparc is down-regulated in carcinoma in situ induced by DMBA (Yan et al., 2004). Alterations in the extracellular matrix and cytoskeleton are likely to influence hormonally regulated growth and differentiation and may also potentially contribute to the low expression of milk protein genes in carcinomas despite the elevated expression of the prolactin receptor and Stat5a (Clevenger et al., 2003; Zoubiane et al., 2004). Altered expression of multiple extracellular matrix genes was observed in carcinomas. Galectin 7, shown previously to be over-expressed in rat mammary gland cancers and considered to be a target gene for p53 (Lu et al., 1997; Shan et al., 2002; Wang et al., 2004), and alcam, a cell adhesion gene linked to tumor invasion (King et al., 2004) showed higher expression in carcinomas than in adenomas.

In summary, microarray analysis of common molecular alterations in rat mammary gland carcinomas induced by various carcinogens and further comparison between these carcinomas with adenomas suggested possible pathways associated with malignant transformation in the mammary gland. Abnormal expression of multiple genes and pathways appear to coordinately establish the malignant nature of rat mammary gland tumors. Further studies to delineate the complement of gene expression alterations required for malignancy in the animal model may help to provide strategies for the prevention and treatment of human breast cancer.