Comparison of expression profiles of tumors from several transgenic models has led to the identification of expression signatures for different oncogenic pathways 7. In order to determine whether tumors from MMTV-Wnt-1 transgenic mice also have a distinctive expression profile, we determined the profiles of 23 mammary tumors from MMTV-Wnt-1 transgenic mice and, for comparison, 12 mammary tumors from mice carrying the MMTV-Neu transgene (Tables 1 and 2 provide sample information and a list of all comparisons). Neu (ErbB2/HER2), a proto-oncogene that is amplified in approximately 25% of human breast cancers 17, encodes a member of the epidermal growth factor receptor family of receptor tyrosine kinases 18. It activates signaling pathways different from those activated by Wnt-1, and the two oncogenes can collaborate in mammary tumorigenesis 19.

The expression profiles of these two sets of tumors were clearly separated into two groups by unsupervised average linkage hierarchical clustering analysis (Figure 1), suggesting that the global expression patterns of these two sets of tumors differ significantly. This finding extends previous reports of significant divergence in histopathobiology, cellular composition, and possibly the cell types of origin between these two groups of tumors 82021.

In an effort to identify genes that are specifically dysregulated in tumors induced by MMTV-Wnt-1, we performed a permutation t-test (see Materials and methods, below, for details) on these two groups of array data. In total, 1,296 genes were differentially expressed between MMTV-Wnt-1-induced and MMTV-Neu-induced tumors (P < 0.001; Table 2 and Additional data file 1). Among the 1,296 genes that we found to be differentially expressed between Wnt-1-induced and Neu-induced tumors, 842 genes are represented in the 8.7k chips used in the previous report 7. In that study, 672 genes were found to be differentially expressed among tumors from MMTV-Neu, MMTV-Ha-Ras, MMTV-c-Myc, MMTV-polyoma middle T antigen, C3(1)/simian virus 40 T/t antigen, and Wap-simian virus 40 T/t antigen transgenic mice using the 8.7k chips. Comparing the 842 differentially expressed genes in the present study with the 672 genes from the earlier study, we found that 165 genes were present in both lists (Additional data file 1), including 91 of the 178 genes (51%) reported as the Neu-Ras-polyoma middle T antigen cluster 7. Examples of these 91 genes include Rap1-GTPase activating protein 1, matrix metalloproteinase 15, and CD81 (Additional data file 1).

It should be noted that the MMTV-Wnt-1 transgenic mice had a mixed genetic background that was mostly FVB (>75%), whereas MMTV-Neu transgenic mice were on a pure FVB background. Although this small variation in genetic background between these two groups of mice is unlikely to account for the differences in expression profiles we detected, we cannot exclude the possibility that some of the genes identified by this analysis might be due to variation in genetic background.

A panel of 652 genes were reported to be differentially expressed between MMTV-Neu-induced tumors and normal virgin mammary glands in the study of Desai and coworkers 7 using the 8.7k chips (> two-fold). In the present study comparing 12 tumors from MMTV-Neu transgenic mice and five nontransgenic normal virgin mammary glands using the 15k chips, 1,263 genes were differentially expressed (P < 0.001, more than three-fold; Table 2). Among these 1,263 genes, 626 genes were represented in the 8.7k arrays used by Desai and coworkers. Of these 626 genes, 225 (35%) overlapped with the 652 genes reported to be differentially expressed between MMTV-Neu-induced tumors and normal virgin mammary glands in the study conducted by Desai and colleagues. We consider this to be an acceptable level of reproducibility, considering the multiple differences in the generation of the two data sets (including differences in reference RNAs, array prints, age of the virgin mammary glands, and sample size).

Genes that were more highly expressed (P < 0.001) in MMTV-Wnt-1-induced tumors than in MMTV-Neu-induced tumors include genes reported to be transcriptional targets of Wnt signaling 2223242526 such as cyclin D₁ (2.0-fold), c-Myc (2.0-fold), frizzled 7 (2.1-fold), and Wnt-5a (9.2-fold; Additional data file 1). Wnt-5b, another member of the Wnt family, was also more highly expressed (3.7-fold) in tumors from MMTV-Wnt-1 transgenic mice than in tumors from MMTV-Neu transgenic mice; it remains to be determined whether this Wnt member is also a transcriptional target of Wnt signaling.

Retinoic acid signaling has been reported to synergize with Wnt signaling to induce gene expression 2728. Retinoic acid receptor and Stra6, a gene activated by the addition of retinoids to cultured cells 29, have also been suggested to be targets of Wnt signaling 27. Consistent with these reports, we found higher level of Stra6 (P < 0.001, 9.0-fold) in MMTV-Wnt-1-induced tumors than in MMTV-Neu-induced tumors. In addition, cellular retinol binding protein (RBP)1, a gene related to retinoic acid signaling, was also more highly expressed (P < 0.001, two-fold) in MMTV-Wnt-1-induced tumors than in MMTV-Neu-induced tumors, which is consistent with our recent report that RBP1 is induced by β-catenin 30.

MMTV-Wnt-1-induced tumors contain both epithelial and myoepithelial cells in approximately equal numbers, unlike tumors induced by the MMTV-Neu transgene, which contain only epithelial tumor cells 82131. Consistent with these reports, we observed higher expression levels (P < 0.001) of myoepithelial markers, including calponin 1 (12.5-fold) and calponin 2 (2.5-fold and 4.0-fold for two separate clones), in tumors from MMTV-Wnt-1 transgenic than in tumors from MMTV-Neu transgenic mice (Additional data file 1). Consistent with earlier reports that tumors may arise from mammary progenitor cells in MMTV-Wnt-1 transgenic mice 815, we found that RNA encoding the candidate progenitor cell markers keratin 6 (13-fold), tenascin (3.1-fold), osteoblast specific factor 2 (2.0-fold), insulin-like growth factor binding protein 7 (2.0-fold), and nidogen 1 (1.8-fold) 832 were more abundant (P < 0.001) in tumors from MMTV-Wnt-1 transgenic mice. Using immunoassays, we demonstrated that keratin 6 and nidogen proteins are expressed at higher level in MMTV-Wnt-1-induced tumors than in MMTV-Neu-induced tumors (Figure 1b) 8.

The distinct patterns of genes expressed in MMTV-Wnt-1-induced and MMTV-Neu-induced tumors described in the preceding section suggest that initiating oncogenes strongly influence gene expression in the tumors arising in these two models. We have observed that other genetic events accelerate tumorigenesis in MMTV-Wnt-1 transgenic mice 1433. We next evaluated whether the events that mediate acceleration are reflected in the gene expression patterns.

We recently reported that approximately 50% of mammary tumors in MMTV-Wnt-1 transgenic mice have activating mutations in the Ha-Ras locus 19. Thus, we first considered whether tumors carrying mutant Ha-Ras have an expression profile distinct from that observed in tumors that are wild-type at the Ha-Ras locus. We sequenced Ha-Ras cDNA to seek mutations in 21 out of the 23 MMTV-Wnt-1-induced tumors: 12 tumors carry Ha-Ras mutations and nine have only Ha-Ras wild-type alleles (Figure 1a). Tumors with and without Ha-Ras mutations did not have distinct global expression profiles (Figure 1a). Independent multidimensional scaling (MDS) and hierarchical clustering of these 21 tumors based on expression profiles also did not separate them according to Ha-Ras status (data not shown). Nevertheless, permutation t test identified 40 genes differentially expressed between tumors bearing wild-type Ha-Ras and those carrying a mutant Ha-Ras (Table 2 and Additional data file 2). This is more than expected (P < 0.001) but many fewer than we saw in our earlier comparison between MMTV-Wnt-1-induced and MMTV-Neu-induced tumors. In addition, the average fold difference is much smaller (Additional data file 2) than that in the earlier comparison.

We previously determined that loss of either p53 or Pten accelerates mammary tumorigenesis in MMTV-Wnt-1 transgenic mice 3435. To further investigate the influence of these genetic alterations on expression patterns in MMTV-Wnt-1-induced tumors, we determined the expression profiles of six tumors from MMTV-Wnt-1 transgenic mice that were p53 null and three tumors from MMTV-Wnt-1 transgenic/Pten^+/- mice that had lost the wild-type allele of Pten (i.e. loss of heterozygosity). When genes in the 8.7k array data set from these two groups of tumors and those from 10 tumors from MMTV-Wnt-1 transgenic mice that were otherwise wild-type were subjected to analysis by MDS or unsupervised hierarchical clustering, the three groups of samples could not be separated from each other (Figure 2 and data not shown). These findings suggest that the global expression profiles of MMTV-Wnt-1 tumors carrying different additional genetic alterations cannot be distinguished.

Permutation t test identified 113 genes that were differentially expressed (p < 0.001) between tumors from MMTV-Wnt-1 transgenic mice and those from MMTV-Wnt-1 transgenic/p53^-/- mice (Table 2). Among the 113 genes, 43 were upregulated, and 70 were downregulated in the latter set of tumors (Additional data file 3). Examples of the upregulated genes are cyclin D₂ (3.7-fold), Myb (2.9-fold and 2.8-fold for two separate clones), Bcl11a (1.5-fold), and Pbx3 (1.8-fold average), which promote proliferation or survival. Examples of the downregulated genes are CD59a antigen (two-fold), a potential p53 target 36, the Rb1 tumor suppressor gene, and the Met proto-oncogene. Using similar analyses, we found that 115 genes were differentially expressed (P < 0.001) between tumors from MMTV-Wnt-1 transgenic mice and those from MMTV-Wnt-1/Pten^+/- mice with Pten loss of heterozygosity (Table 2). Forty-five were upregulated, and 70 of them were downregulated in the latter set of tumors (Additional data file 4). Interestingly, among the downregulated genes is tensin (two-fold), a cell adhesion molecule that is related to Pten. Similar to the comparison between Ha-Ras mutant and Ha-Ras wild-type tumors in MMTV-Wnt-1 transgenic mice, the number of genes differentially expressed and the average fold difference were much smaller in the above two comparisons than in the comparison between MMTV-Wnt-1-induced and MMTV-Neu-induced tumors (Additional data files 1, 3, and 4).

Collectively, these findings suggest that tumors from MMTV-Wnt-1 transgenic mice are similar to each other in their global expression profiles, regardless of whether the tumors have additional genetic alterations. It is not known whether the modest differences in RNA levels we identified among these groups of tumors explain the accelerating effects of these alterations on tumorigenesis in MMTV-Wnt-1 transgenic mice. We plan to test some of these changes by expressing cDNAs in mammary glands in Wnt-1 transgenic mice using TVA-mediated somatic gene transfer technology 37.

Hyperplastic lesions are widespread in MMTV-Wnt-1 transgenic mice before the development of mammary tumors 14. To determine whether unique gene expression patterns accompany the evolution from normal mammary cells to hyperplasias and then to tumors, we compared expression profiles among mammary glands of nontransgenic virgin mice, hyperplastic mammary glands, and mammary tumors from MMTV-Wnt-1 transgenic mice. Unsupervised hierarchical clustering analysis and MDS showed that expression profiles from these three groups of tissues were separated from each other (Figure 3a and data not shown). The difference between hyperplastic and normal glands is unlikely to be due to decreased contribution of stromal RNA in the preparation of RNAs from the hyperplastic glands from MMTV-Wnt-1 transgenic mice, because the expression levels of epithelial and myoepithelial marker genes (keratin 19, calponin 1, and calponin 2) were not significantly statistically different between hyperplastic mammary glands from MMTV-Wnt-1 transgenic mice and mammary glands from age-matched nontransgenic virgins.

In total, 584 genes were differentially expressed (P < 0.001, Table 1) between hyperplastic mammary glands from MMTV-Wnt-1 transgenic mice and normal mammary glands from nontransgenic littermates. Among these 584 genes, 121 were more highly expressed in the hyperplastic glands (Additional data file 5), which includes some of the known transcriptional targets of Wnt signaling such as c-Myc (3.6-fold) and frizzled 7 (2.1-fold). This list may therefore provide an important starting point for confirming mammary-specific target genes and for discovering novel in vivo targets of Wnt signaling. In fact, two genes in this list, namely RBP1 (2.9-fold) and tumor-associated calcium signal transducer (3.5-fold), were shown to be upregulated by β-catenin in 293 cells in our recent studies 30.

One of the greatest challenges in identifying specific genes and expression patterns associated with the evolution from hyperplastic glands to tumors is the change in cellular composition. Normal and MMTV-Wnt-1-induced hyerplastic ductal trees are embedded in stroma, but tumors often contain much less stroma. Thus, the differential contribution of RNA from the stromal cells may skew array analysis, which is based on total RNA content. However, stromal cells are mostly large adipocytes whose RNA to mass ratio is small; thus, the relative contribution of RNA from these cells is probably much less than it appears to be from histologic assessment. The average expression level of epithelial and myoepithelial markers (keratin 19, calponin 1, and calponin 2) was 2.6-fold higher in the tumors (which contain very few stromal cells) than in the hyperplastic tissues, suggesting that 38% (1/2.6 = 38%) of the RNA in the hyperplastic tissues might come from the ducts and alveoli. Thus, to eliminate genes that were not truly differentially expressed, we filtered out any genes that were less than three-fold different in our comparison between tumors and hyperplastic glands in Table 2.

Based on the above calculation, approximately 62% of the RNAs from hyerplastic mammary glands might come from adipocyte-rich stroma. Thus, tumor samples, which have very little contribution from adipocytes, may appear to have downregulated the genes that are associated with adipocytes. In order to identify these genes, we compared the expression profiles of a set of three fat samples with those of the 35 mammary tumors from MMTV-Wnt-1 and MMTV-Neu transgenic mice. Expression of 741 genes was at least three-fold or higher (P < 0.001) in fat than in the mammary tumors (Table 2 and Additional data file 6). These include published fat-specific genes (Additional data file 6), such as fat-specific gene 27, lipoprotein lipase, CD36, carbonic anhydrases, and solute carrier family members 7. We note these genes in our table comparing hyperplastic mammary glands with tumors (Additional data file 7).

In total, 1,372 genes were differentially expressed (P < 0.001) between tumors and hyperplastic glands from MMTV-Wnt-1 transgenic mice. Among them, expression levels for 388 differed by at least three-fold (Additional data file 7). This subgroup is likely to contain genes that are important for the evolution of hyperplastic lesions into tumors, including genes that are required for tumor cell proliferation and survival. One such candidate is c-Kit, a proto-oncogene that is frequently overexpressed in cancers and that encodes a receptor that activates both Ras and Akt pathways. The expression of c-Kit was 3.6-fold higher in tumors than in hyperplastic lesions (Additional data file 7), although it was similarly expressed in normal virgin glands and hyperplastic mammary glands from MMTV-Wnt-1 transgenic mice. This is consistent with a recent report that c-Kit is highly expressed in the basal group of human breast tumors compared to other groups 38. Using immunohistochemical staining, c-Kit protein was barely detectable in normal mammary glands from nontransgenic mice and in hyperplastic mammary glands from MMTV-Wnt-1 transgenic mice, but was readily and widely detected in the tumor samples from MMTV-Wnt-1 transgenic mice (Figure 3b).

Some of the genes that were differentially expressed between mammary tumors and hyperplastic glands in MMTV-Wnt-1 transgenic mice may be needed for evolution of tumors induced by both Wnt-1 and other oncogenes. Other genes may be uniquely important for induction of tumors from hyperplastic glands in MMTV-Wnt-1 transgenic mice. For example, certain signaling pathways may need to be activated in hyperplastic cells in MMTV-Wnt-1 transgenic mice before a tumor will form, but they may be optional for tumorigenesis initiated by other oncogenes. To discover genes that might be uniquely important for tumors to develop in hyperplastic glands in MMTV-Wnt-1 transgenic mice, we compared the 388 genes that we found to be differentially expressed between tumors and hyperplastic glands from MMTV-Wnt-1 transgenic mice with the 1,296 genes that we found to be differentially expressed between tumors from MMTV-Wnt-1 transgenic and MMTV-Neu transgenic mice. Fifty-six genes corresponding to 59 cDNA clones in the former group were shared in the latter group (Table 3), suggesting they might be specifically involved in neoplastic progression in MMTV-Wnt-1 transgenic mice. Among these 56 genes, 23 were more highly expressed and 33 were expressed at lower level in MMTV-Wnt-1-induced tumors than in either MMTV-Wnt-1-induced hyperplasia or MMTV-Neu-induced tumors (Table 3). The upregulated genes (P < 0.001) include TNFRSF19 (3.5-fold), NGFR (3.6-fold), apolipoprotein D (4.7-fold), and Wnt5a (4.4-fold), and the downregulated genes (P < 0.001) include BNIP3 (2.5-fold) and caveolin (2-, 3.3-, and 3.3-fold for three separate clones). Of note, apolipoprotein D has been reported to be upregulated in a subset of human breast cancers 39, and caveolin 1, a negative regulator of the Ras-p42/p44 mitogen-activated protein kinase cascade, has been reported to inhibit growth in human breast cancer cells 40.

The mouse 15k slides and 8.7k slides used in this study were arrayed at the National Cancer Institute microarray facility. All slides of each array type were printed in a single batch. The 8.7k slides contain the 8700 Incyte GEM1 clone set, which are mapped to 6,877 Unigene cluster IDs, among which 2,953 are named genes, 2,206 are expressed sequence tags, and 1,628 are Riken cDNAs. The 15k slides contain the 8700 Incyte GEM1 clone set and the mammary 6000 clone set; a total of 1,444 clones do not map to a Unigene cluster ID, whereas the rest of the clones map to 10,062 unique genes as defined by Unigene cluster ID. Among the 10,062 Unigene clusters, 3,750 are named genes, 3,922 are expressed sequence taqs, and 2,390 are Riken cDNAs.

All nontransgenic and MMTV-Neu transgenic animals used in this study were on the FVB background. All MMTV-Wnt-1 transgenic mice 14 were a mixture of FVB (>75%), SJL, and C57BL/6 strains. MMTV-Neu transgenic mice 41 were purchased from Jackson Laboratories (Bar Harbor, ME, USA). This transgenic line carries a rat cDNA encoding the wild-type Neu protein. Fat tissues were collected from intestinal fat in virgin FVB mice. All samples were collected fresh and snap-frozen in liquid nitrogen. RNA was extracted by Trizol (Invitorgen, Carlsbad, CA, USA). Reference RNA was a mixture of ovarian RNA (Ambion, Austin, TX, USA; Cat number 7824) and RNA extracted from tissues of liver, spleen, kidney, thymus, pancreas, lung, and normal lactating mammary gland of FVB mice of 6 months of age. All reference RNA used in this study is from a single preparation.

The cDNA probes were prepared from a total of 35-50 μg reference RNA and 50-75 μg sample RNA from normal, hyperplastic, or tumor tissues, as described 4243. The cDNA from reference RNA was labeled with cyanine 3-dUTP, and that from sample RNA was labeled with cyanine 5-dUTP. Fluorescent images of hybridized microarrays were obtained by using a GenePix 4000 scanner (Axon Instruments, Foster City, CA, USA). Microarray images were analyzed using the ArraySuite software 4445 based on the Scanalytics IPlab platform (Scanalytics, Fairfax, VA, USA). For each cDNA probe location, fluorescence intensity ratio and its associated measurement quality (q) were calculated. The evaluation of measurement quality is based on spot size, signal to noise ratio, background uniformity, and saturation pixel percentage 45. The range of measurement quality is from 1.0 to 0, with higher measurements reflecting better quality. Areas of the array with obvious blemishes were automatically given a low quality value.

Normalized log test to reference ratios and their corresponding quality measurements in each experiment were calculated as described previously 45. A gene was excluded from further analyses (see below for description) if the average quality measurement was under 0.5 across samples in that specific comparison. Approximately 14,000 genes are suitable for analyzing on the 15k chips and 8,000 genes on the 8.7k chips.

Two methods were used to visualize the expression patterns among samples. Both used the Pearson correlation as a similarity measure. In average linkage hierarchical clustering, the distances between samples are represented on a tree called a dendrogram. In MDS, samples with similar expression ratios were placed closer to each other in three dimensional space. Average linkage hierarchical clustering analysis was implemented using the CLUSTER program, and the results were displayed using TREEVIEW 46. MDS was developed in the MATLAB (Natick, MA, USA) environment.

A permutation t-test was used to select genes significantly differentially expressed between any two groups 47. Here, a standard t-statistic was computed between two groups on the log-transformed ratios of each gene. The group labels were randomly permuted and the t-statistic for each gene in the permuted data set was calculated. The process was repeated 10,000 times. A P value was reported for each gene by comparing the observed statistic with the permutation statistics. To control for multiple comparisons, only genes with P values less than 0.001 were considered differentially expressed. The distribution of the significant differences expected by chance and the probability of observing as many or more differentially expressed genes were calculated from the permuted data. This latter probability is the P value reported in the Results and discussion section.

Antibodies used include rabbit IgG directed against nidogen and c-Kit (Santa Cruz, CA, USA). Tissues were fixed in 10% neutral formalin and processed as previously described 34 to obtain paraffin sections of 4 μm in thickness. For immunohistochemistry, the sections were boiled for 15 minutes in 10 mmol/l citrate buffer of pH 6.0 (to unmask antigen epitopes). Endogenous peroxidase activity was inactivated by 10 minute incubation in 3% hydrogen peroxide, and subsequent steps were performed using Vector ABC kits and the Nova-Red substrate (Vector Laboratories, Burlingame, CA, USA) following the manufacturer's recommendations.

For Western blotting, tumors were ground to powder in liquid nitrogen and lysed in the M-PER tissue lysis solution (Pierce, Rockford, IL, USA) with gentle shaking overnight at 4°C. Proteins in resulting supernatant (25 μg protein) were denatured using 2-mercaptoethanol, resolved on 10% polyacrylamide mini-gels containing 10% sodium dodecyl sulfate, and transferred to nitrocellulose membranes. The membranes were then incubated with primary antibodies and peroxidase-conjugated secondary antibodies (Jackson Laboratories) in tris-borate buffer (1 mmol/l Tris and 13.7 mmol/l NaCl, pH 7.6)/0.05% Tween 20/5% nonfat dried milk. Proteins recognized by specific antibodies were visualized using a chemiluminescent substrate (Supersignal; Pierce).

The GEO accession number for the series of array data-sets is GSE2860.