There exist many mouse models of breast cancer (7, 8, 9, 10, 11). Besides having differences in histopathology and possibly in cell of origin, these models also have distinct gene expression profiles (12, 13, 14, 15, 16). For example, distinct gene signatures have been identified for mammary tumors arising in mice transgenic for Neu, c-Myc, H-Ras, SV40 T antigen, and polyoma middle T antigen (12).

Gene expression profiles are very different in tumors arising in transgenic mice that express Wnt-1 from the mouse mammary tumor virus promoter (MMTV-Wnt-1) as compared with tumors from MMTV-Neu transgenic mice (15). However, spontaneous or germline secondary genetic alterations, such as an activating Ras mutation or deletion of either the Pten or p53 tumor suppressor gene, do not much alter the global expression profiles of mammary tumors induced by the MMTV-Wnt-1 transgene, though a small subset of genes can be identified to be differentially expressed between tumors with those genetic alterations using the permutation t test (15).

Mice bitransgenic for MMTV-Wnt-1 and MMTV-Neu develop mammary tumors much more rapidly than mice harboring either transgene; thus, these two oncogenes collaborate to induce mammary tumors (17). The resulting mammary tumors contain heterogeneous cell types, including epithelial and myoepithelial cells as well as cells that express putative progenitor cell markers (17), similar to tumors arising in MMTV-Wnt-1 transgenic mice (18). These tumors in bitransgenic mice also contain patches that lack myoepithelial cells, a feature usually associated with tumors arising in MMTV-Neu transgenic mice (17). However, it is not known whether the expression profiles of tumors arising in these bitransgenic mice resemble those of tumors arising in MMTV-Wnt-1 or in MMTV-Neu transgenic mice.

In this study, we sought to answer this question and attempted to identify genes that are uniquely expressed in tumors arising in the bitransgenic mice, in hope of elucidating the mechanism of collaboration between Wnt-1 and Neu in mammary tumorigenesis.

Materials and Methods

Animals and Tumors

MMTV-Wnt-1 (FVB/NJ-Tg[Wnt-1] 1 Jev/J) and MMTV-Neu (FVB/N-Tg [MMTVNeu] 202 Mul/J) mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA) and maintained on the FVB background. All mice were housed in a specific pathogen-free facility on a standard diet. Tumors were collected fresh, snap-frozen in liquid nitrogen, and kept at -80 °C until RNA extraction. We chose five animals in each group based on the guidelines set by The Tumor Analysis Best Practices Working Group (19), which require three or four inbred mice per group.

RNA Extraction and Affymetrix GeneChip Array Hybridization

Frozen samples were ground into fine powder in liquid nitrogen, and their RNAs were extracted by Trizol (Lift Technologies) following the manufacturer?s suggestions with an additional step of isopropanol precipitation to reduce potential residual chemical contamination in the Trizol reagent. Double-stranded cDNA was synthesized from 5 μg of total RNA for each sample. This cDNA served as the template for producing biotin-labeled cRNA through an in vitro transcription reaction. The One-Cycle Target Labeling and Control Reagents kit (Part #900493; Affymetrix, Santa Clara, CA, USA) was used for both cDNA synthesis and cRNA biotin-labeling. Fragmented cRNA (6.5 ?g) was hybridized onto the Affymetrix GeneChip^® M430A 2.0 in the Affymetrix Hybridization Oven 640 for 16-19 hours. Prehybridization, hybridization, washing, and staining with streptavidin-phycoerythrin (SAPE) were performed following manufacturer?s recommended protocols. Antibody amplification was employed using anti-streptavidin antibodies conjugated to biotin (Vector Laboratories, Burlingame, CA). After automated staining and washing using Affymetrix GeneChip^® Fluidics Station 450 (Protocol: Midi_euk2v3 450), the arrays were scanned using the Affymetrix GeneChip^® scanner 3000, which was controlled using the GeneChip^® Operating System (GCOS) v1.2. Image analysis and probe level quantification were carried out using GCOS1.2, and the probe level data (CEL file) were saved for each array for subsequent analyses.

Data Normalization and Modeling

Both GCOS and dChip (http://dchip.org) (20, 21) were used for quality assessment. From GCOS analysis, the percent present calls for all the 15 arrays were between 61-69% and the GAPDH ratio was between 0.61-0.77. No obvious variation for these two measurements was observed among these samples. The CEL files for all 15 arrays were then imported into dChip, and normalized using the invariant set method (the baseline was set using the WntT H11 tumor sample). The perfect match only (PM-only) modeling algorithm was used to estimate expression. Expression values of probe sets, after rounding up to one for any value that was less than one, were Log2-transformed for further analyses. dChip analysis showed that all the arrays had percentages of array outliers <5% and percentage of single outliers <0.6%, indicating acceptable quality for all the arrays (19).

Statistical Analyses

Expression patterns among samples were visualized using Hierarchical clustering and multidimensional scaling (MDS), both of which use Pearson correlation as a similarity measure. The clustering analysis was implemented in dChip, with the distances between samples represented on a dendrogram. MDS was developed in Matlab environment (Natick, MA, USA) (22). After filtering out those that had nearly identical expressing values, 7,082 (of 22,690) probe sets were selected for both unsupervised clustering and MDS analyses. To identify genes differentially expressed between samples, we used two-sample t test for comparing two groups of samples and One-way Analysis of Variance (ANOVA) for three groups. QVALUE (implemented in R) (23) was used to determine false discovery rates (FDRs), and the estimation of FDRs is based on the entire list of p values obtained by ANOVA or t test.

Results

Gene expression profiles were determined using Affymetrix GeneChip^® M430A 2.0 arrays for five tumors from mice bitransgenic for Wnt-1 and Neu, five tumors from MMTV-Wnt-1 transgenic mice, and five tumors from MMTV-Neu transgenic mice. From these 15 arrays, probe sets that had similar expression values were filtered out. The remaining 7,082 (of 22,690) most variable probe sets were used for both unsupervised clustering and MDS analyses. Neu-induced tumors were clearly separated from Wnt-1-induced tumors (Figure 1), consistent with our previous report using cDNA arrays (15). Tumors from the bitransgenic mice were separable from Neu-induced tumors; however, to our surprise, they were not separated from tumors from MMTV-Wnt-1 transgenic mice, using both analyses (Figure 1).


Figure 1: Unsupervised MDS (A) and Clustering (B) analyses. These analyses were carried out using the 7082 (of 22,690) probe sets which satisfied the filtering criteria of variation across samples: 0.04 < SD/Mean < 10.00. Samples include tumors from MMTV-Wnt-1 transgenic mice (Red), MMTV-Neu transgenic mice (Blue), and MMTV-Wnt-1/MMTV-Neu bitransgenic mice (Green).

In an attempt to identify potential differentially expressed genes/patterns among these three groups, One-way ANOVA was performed using all 22,690 probe sets. 2537 genes were identified (p≤0.001) (Figure 2). From these 2537 genes, we would expect about 0.001 × 22,690 = 23 genes by chance. To better estimate the false discovery rate, we have used a standard method QVALUE to compute the false positives among the identified genes. We observed many more genes with lower p values than with higher p values; a trend indicates more significant genes than by chance (Figure 3A). More importantly, the number of false positives is very low when compared with the identified genes (significant tests) (Figure 3B). Approximately nine of the 2537 were estimated as false positives by QVALUE (Figure 3B). This data indicated that nearly all of the 2537 genes identified by ANOVA were truly differentially expressed among these three groups of samples.


Figure 2: MDS (A) and clustering (B) analyses on genes identified by One-way ANOVA. One-Way ANOVA identified 2537 genes differentially expressed among tumors from MMTV-Wnt-1 transgenic mice (Red), MMTV-Neu transgenic mice (Blue), and MMTV-Wnt-1/MMTV-Neu bitransgenic mice (Green).

Next we used the 2537 genes for Hierarchical clustering analysis and MDS. We expected these three groups of samples to be separated from each other, since clustering on selected significant genes usually leads to group differentiation. We also hoped to identify genes/patterns specifically regulated in tumors induced jointly by Wnt-1 and Neu. These 2573 genes clearly separated tumors induced by Neu from tumors induced by either Wnt-1 alone or a combination of Wnt-1 and Neu (Figure 2). However, tumors induced jointly by Wnt-1 and Neu could not be differentiated from tumors induced by Wnt-1 alone, by either analysis (Figure 2). These data suggest that these two groups of tumors are very similar in their gene expression profiles.

To ensure that the above ANOVA on the three groups did not mask small but consistent differences between tumors arising in MMTV-Wnt-1/MMTV-Neu bitransgenic mice and in MMTV-Wnt-1 transgenic mice, the two-sample t test was used to uncover genes that might differ between these sets of tumors. Among all the 22,690 probe sets, two genes were identified (p=<0.001). Unlike results from ANOVA analysis (Figure 3A), we did not observe more genes with lower p values than with higher p values with the two sample t test (Figure 3C); more importantly, the estimation of the false discovery rate by QVALUE, using all the p values from this two sample t test, indicated that the same number of genes that would be identified as significant would also be identified by chance (FDR 100%) (Figure 3D). Such a null hypothesis predicted a uniform p value distribution. Unexpectedly, the p value distribution was not uniform across all the genes, with a higher prevalence of very large p values (Figure 3C), suggesting that, for any particular gene, the difference between the means of the two groups is smaller than would be expected by chance, given the within-group variability. Small experimental/biological variations from pairing individual samples between the compared groups during sample collection, cDNA/cRNA synthesis, and array hybridization, may have been the reason for such a non-uniform distribution of p values, though these experiments were designed to minimize introduction of experimental and biological variations between groups. Noticeably, unsupervised clustering analysis clustered the ten tumors into five pairs (Figure 1B), four of which reflected these commonalities (two pairs were littermates of the same age and two pairs were assayed at the same time), while the remaining pair (WntNeuT H10 vs. WntT H4) had no identifiable commonality. (These subtle differences between pairs of samples were noticeable on clustering analysis, apparently, because these two groups of samples were very similar. In contrast, such pairing did not affect the comparison between Neu-induced tumors and tumors arising in mice either bitransgenic for Neu and Wnt-1 or monotransgenic for Wnt-1, presumably due to the large differences in expression profiles among them.) Therefore, we next performed a paired t test for samples between tumors arising in MMTV-Wnt-1/MMTV-Neu bitransgenic mice and in MMTV-Wnt-1 transgenic mice. As expected, only a small number of genes were identified (11 genes, p≤0.001), and the estimated FDR was 100% (data not shown). Importantly, the distribution of p values was nearly uniform across all the genes (Figure 3E), confirming the null hypothesis. Hence, we conclude that gene expression profiles of tumors from MMTV-Wnt-1/MMTV-Neu bitransgenic mice are highly similar to those of tumors from MMTV-Wnt-1 transgenic mice.

Ecommerce Solutions
Online Payment Solutions