Concordant Effects of Aromatase Inhibitors on Gene Expression in ER⁺ Rat and Human Mammary Cancers and Modulation of the Proteins Coded by These Genes

The preponderance of invasive breast cancers in women are estrogen receptor–positive (ER⁺). Approximately 35 years ago, agents were developed that antagonized the estrogen receptor; for example, tamoxifen (1). Hormonal therapy can also be accomplished by inhibiting the production of estrogens; specifically, inhibition of the cytochrome P450–mediated enzyme aromatase (CYP 19; ref. 2). Anastrozole and letrozole, two highly specific low K_i competitive inhibitors, have proven highly effective in both therapy (inhibiting recurrence) and prevention (inhibition of cancer occurrence in the contralateral breast) in various adjuvant trials (3, 4). More recently, a primary prevention trial of the aromatase inhibitor exemustane has proven highly effective (5). Vorozole (R83848) is a high-affinity competitive inhibitor of aromatase and shows strong activity in early clinical trials in ER⁺ breast cancers (6, 7).

Chemically induced models of ER⁺ mammary cancer in rats were developed several decades ago (8, 9). The resulting cancers were ER⁺, near diploid, and by array analysis were similar to well-differentiated ER⁺ breast cancer in women (10). Our laboratory and others showed that vorozole was highly effective both in the prevention and therapy of ER⁺ mammary cancers in animal models (11, 12). Subsequently, we have done a variety of studies with this agent; examining its effects on pharmacodynamic markers such as estrogen and estradiol levels and expression of insulin like growth factor-I (IGF-I). Changes in these biomarkers in the rat were similar to the responses achieved with aromatase inhibitors clinically (13). In addition, we showed that vorozole significantly decreased proliferation in the cancers (14). This had similarly been observed in ER⁺ breast cancer in women in a neoadjuvant setting (15). This study was undertaken in significant part to validate the methylnitrosourea (MNU)-induced ER⁺ breast cancer model as compared with human data. We performed global gene expression analysis on mammary cancers induced by MNU and exposed to either vehicle or vorozole treatment for 5 days. The major objectives of this study were to: (i) identify differentially expressed genes (DEG) and related biologic pathways that may be relevant to the mechanism of response to vorozole in ER⁺ mammary cancers, (ii) examine whether the gene expression changes in the rat mammary cancer model significantly overlapped the changes in gene expression observed in certain published neoadjuvant studies with aromatase inhibitors in humans, (iii) compare gene changes obtained in animals with in vitro results of estrogen withdrawal, (iv) compare results obtained in (i) and (ii) with a large set of samples taken from an independent neoadjuvant trial with anastrozole, and (v) determine whether certain number of the changes in expression of proliferation-related genes could be confirmed at the protein levels by immunohistochemistry (IHC). Protein expression was examined both in vorozole-treated tumors and vorozole-treated normal mammary epithelium.

Vorozole (R-83842) was supplied by Johnson & Johnson Pharmaceuticals. The purchase of rats and their treatment regimens were identical to our previously published methods (14). The carcinogen MNU was injected intravenously (75 mg/kg BW) via the jugular vein when the rats were 50 days of age. When an animal developed a tumor of approximately 100 to 150 mm², the rat was given vorozole at 1.25 mg/kg BW/day by gavage for 5 days [vehicle was ethanol/polyethylene glycol 400 (10:90, v/v)]. At termination of the study, the animals were sacrificed and the cancers were removed. Treatment with vorozole for 5 days did not significantly decrease tumor volume relative to initial volume although treatment for a period of 2 weeks will cause significant tumor regression (data not shown). Portions of the tumors were fixed for histopathology and IHC; the remaining portion was frozen in liquid nitrogen and stored at −85°C for subsequent molecular assays.

Total RNA from tumor tissues were isolated by Trizol (Invitrogen), and purified using the RNeasy Mini Kit and RNase-free DNase Set (QIAGEN). One microgram of each RNA sample was processed using Affymetrix GeneChip Whole Transcript Sense Target Labeling Assay. GeneChip WT cDNA Synthesis Kit, cDNA Amplification Kit, and Terminal Labeling Kit (Affymetrix, Inc.) were used for target preparation. Eight microgram of cRNA were input into the second-cycle cDNA reaction. Hybridization cocktails containing 3 to 4 μg of fragmented, end-labeled cDNA were prepared and applied to GeneChip Rat Exon 1.0 ST arrays. Hybridization was performed for 16 hours using the MES_EukGE-WS2v5_450-DEV fluidics wash and stain script (precommercial FS450_0001 script). Arrays were scanned using the Affymetrix GCS 3000 7G and GeneChip Operating Software v. 1.3 to produce CEL intensity files (16).

The two-sample Student t test was used to identify DEGs and simultaneously relative fold changes were calculated. The DEGs were defined as genes with P < 0.05 and fold change > 1.5 between the groups. The DEGs were analyzed for GOTERM and pathway enrichment by the program Database for Annotation, Visualization and Integrated Discovery (DAVID) (17). The gene expression data were input into gene set enrichment analysis (GSEA). GSEA is a computational method that determines whether a set of genes shows statistically significant differences in expression between two biologic states. This has proven successful in discovering molecular pathways involved in human diseases (18). Using the Kolmogorov–Smirnov statistic, GSEA assessed the degree of “enrichment” of a set of genes (e.g., a pathway) in the entire range of the strength of associations with the phenotype of interest. It was used to identify a priori defined sets of genes that were differentially expressed (17). We used curated gene sets that contain genes on certain molecular pathways, and Gene Ontology (GO) gene sets that consist of genes annotated by the same GO terms in the Molecular Signature Database (MSigDB; refs. 19, 20). Because of the small sample sizes, GSEA with gene set permutation option was performed. Selected gene sets identified from GSEA were then visualized with MetaCore (21).

The following four datasets of gene expression profiles of human ER⁺ breast cancer treated with aromatase inhibitors (letrozole/anastrozole) were downloaded and compared with the animal results. Miller and colleagues (22) performed a microarray study from paired tumor core biopsies taken before and after 14 days of treatment with the aromatase inhibitor letrozole (2.5 mg/day, orally) in 58 patients with postmenopausal ER⁺ breast cancer. Creighton and colleagues (23) examined 18 patient pairs before and after 14 days of treatment with letrozole. Desmedt and colleagues (GSE 16391) performed a microarray study using 7 patients with relapse and 21 patients with no-relapse ER⁺ breast cancer receiving letrozole treatment. The array data were downloaded from Gene Expression Omnibus (GSE5462, GSE10281, and GSE16391). The raw fluorescence intensity data within CEL files were preprocessed with MAS 5.0 algorithm. Mackay and colleagues (24) studied the molecular response to aromatase inhibitors in postmenopausal patients with primary ER⁺ breast cancer. Anastrozole (1 mg/day, orally) and letrozole (2.5 mg/day, orally) was given for 2 weeks to 18 and 16 patients, respectively. Biopsies were taken before and after treatments. The data were downloaded from ArrayExpress (E-TABM-180).

Mammary tumors or normal mammary glands were excised and drop-fixed in Zamboni's fixative [0.03% picric acid (w/v) and 2% paraformaldehyde (w/v)] for 48 hours at 4°C and then transferred to a 20% sucrose solution with 0.05% sodium azide in PBS for storage. Processing and staining of tumors or normal mammary epithelium were carried out according to a published procedure (25). Whole tumors were cryosectioned into 80 μm sections. Floating sections were incubated first with the primary antibody to detect cyclin A (Ccna2; Rb-Santa Cruz Biotechnology: sc-751), BUBR1 (Gt-Santa Cruz Biotechnology: sc-16195), Cdc2 (Rb-Santa Cruz Biotechnology: sc-747), PTTG (Rb-Santa Cruz Biotechnology: sc-22772), and TPX2 (Rb-Santa Cruz Biotechnology: sc-32863) and then incubated in 1:1,000 anti-goat immunoglobulin G (IgG) conjugated to Cy2, Cy3, or Cy5 (Jackson ImmunoResearch), raised in donkey. Optical sections were captured by laser scanning confocal microscopy (NIKON C1si Confocal Spectral Imaging system, NIKON Instruments Co.)

Final results were compared with an unpublished relatively large dataset of primary ER⁺ breast tumors treated with anastrozole. This larger dataset included data from 112 patients (among these, 81 samples had both baseline and samples following 2 weeks of anastrozole treatment) that were used to confirm the genes identified in vorozole-treated rats and the above published human datasets. Fourteen-gauge core biopsies were obtained from postmenopausal women with stage I to IIIB ER⁺ early breast cancer before and after 2 weeks of anastrozole treatment in a neoadjuvant trial (26).

Good quality RNA extracted from pre- and posttreatment biopsies was analyzed on Illumina HumanWG-6 v2 Expression BeadChips (>48,000 probes). Data processing and normalization were performed using Lumi package in Bioconductor software. Probes were filtered out for removal from downstream analyses if they were not detected in any of the samples (detection P > 1%). Genes that significantly changed expression upon anastrozole treatment were identified using paired class comparison of a set of 81 matched baseline and 2-week posttreatment pairs from which good quality global gene expression data were available.

Changes in expression of various genes were observed between control rat mammary cancers and mammary cancers in rats treated with vorozole (Fig. 1A). These genes included 162 genes downregulated and 180 genes upregulated in tumors of vorozole-treated rats (P < 0.05 and absolute fold change >1.5 between the two groups; Supplementary Table S1). Significant GO level 3 biologic process terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways are listed in Table 1 (P < 0.0001). The downregulated genes in tumors treated with vorozole are highly enriched in cell cycle–related pathways, and the upregulated genes were enriched in cell motility and response to drug (Table 1).

We applied GSEA to the microarray data of control mammary cancers and mammary cancers from rats treated with vorozole. Of 1,892 curated gene sets and 1,454 GO gene sets, the dysregulated genes in mammary cancers from vorozole-treated rats significantly involved 23 gene sets/pathways; all of them were significantly downregulated. There were no gene set significantly enriched in control mammary cancers at FDR < 0.25 (Table 1). Several gene sets related to cell cycle and cancer genesis were significant, including E2F1-related target genes.

In dataset GSE5462, 121 downregulated and 112 upregulated genes (P < 0.05 and fold change > 1.5) were identified when comparing pretreatment and treatment with letrozole at a dose of 2.5 mg/day orally for 10 to 14 days (Supplementary Table S2). The downregulated genes in cancers from patients treated with letrozole were highly enriched in cell cycle–related pathways, and the upregulated genes were enriched in lipid metabolic process and PPAR signaling pathway (Supplementary Table S3). In dataset GSE10281, 609 genes were downregulated and 1,613 genes were upregulated (P < 0.05 and fold change > 1.5) when comparing pretreatment and letrozole-treated postmenopausal patients (Supplementary Table S4). The downregulated genes in tumors treated with letrozole were highly enriched in cell cycle and p53 signaling pathways (Supplementary Table S3). In dataset E-TABM-180, the expression change profiles were similar in samples treated with or without anastrozole and letrozole either separately or jointly. This clearly argues that two different aromatase inhibitors induce similar changes; 733 genes were differentially expressed (P <0.05 and fold change >1.5; 286 genes downregulated and 447 genes upregulated). The significant GOTERM of upregulated genes included regulation of immune response, response of stimulus, cell motility, etc. (Supplementary Table S3). In dataset GSE16391, the Cox regression model identified 902 genes related to relapse-free survival after letrozole treatment (P < 0.01), including 249 genes with negative Z value and 653 genes with positive Z value in Cox proportional hazards survival analysis. A positive Z value indicated a worse prognosis and a negative coefficient indicated a protective effect of the variable with which it is associated. If the relapse-free survival correlated with the gene changes due to drug effect, the gene with negative Z value should be upregulated with the treatment in comparison with treatment and no-treatment groups in the above three human datasets. We have reversed the sign of the change to make the results in line with the other sets of data (Table 2).

There were a total of 90 genes significantly differentially expressed with the same trend in at least three of four human datasets. The significant GO level 3 biologic process terms and KEGG pathways are listed in Table 1. Cell cycle and response to hormone stimulus are common terms in these datasets. Thirty-two of these 90 genes were identified to be significant with the same trend in the animal model (P < 0.05 or fold change >1.5; Table 2). Fos, Arid5b, Dcn, Ptprc, Ms4a6b, and Tf were upregulated after vorozole treatment. The downregulated genes after vorozole treatment included: Top2a, Pbk, Cenpf, Bub1, Mki67, Cdc2, Hmmr, Ect2, Ttk, Ccna2, Prc1, Ccnb1, Bub1b, Gins1, Ns5atp9, Melk, Tyms, Aspm, Aurka, Ccne2, Bard1, E2f8, Dtl, Cep55, Kif20a, and Espl1.

Significantly, DEGs after treatment with an aromatase inhibitor in at least one dataset were uploaded into MetaCore, a systems biology pathway analysis tool. The GeneGo Pathway map for the anaphase-prometaphase complex (APC; which is involved in chromosome condensation in prometaphase) was consistently altered (Fig. 1B). Most of the key genes related to chromosome condensation such as Aurora-A, BUB1, BUB1B, cyclin A, cyclin B, and CDK1 were downregulated after treatment with an aromatase inhibitor. These genes were consistently changed in both the human and animal data.

In addition to comparing results with various in vivo studies, we compared results with an in vitro study in which estrogen was deleted from an ER⁺ human breast cancer cell line (MCF-7). Using a genome-wide analysis of gene expression, various E2F-regulated gene networks were regulated (27). Because we had observed in our initial analysis of the cancers from vorozole-treated rats that a significant number of the modulated genes were related to E2F, we pursued this further. Genes that were E2F-modulated were examined and 13 genes were differentially expressed with vorozole treatment. They were clearly related to cell proliferation and were E2F1 target genes. Seven of the genes contained either ER- or E2F-binding sites when analyzed by Genomica analysis and the data from Caroll and colleagues (ref. 28; Table 3).

Thirty-two genes identified as significantly differentially expressed with the same trend in the animal model and the four published human datasets in an independent set of human samples (in which 81 ER⁺ tumors were treated with anastrozole for 2 weeks in a neoadjuvant setting) were examined (Table 4). These studies showed that in this large independent set of samples all of the proliferation-related genes were significantly decreased. Interestingly, the relative decrease in array expression were relatively small for the majority of the genes (<50% decrease in expression).

The expression of 10 proliferation-related genes were examined, which were determined to be decreased 35% to 65% by arrays when comparing vorozole-treated and control tumors. They were evaluated for their altered expression by quantitative IHC in tumors and normal mammary epithelium. As can be seen in Fig. 2, we observed altered expression of multiple proteins in treated tumors that were modulated at the transcription level based on arrays. In addition, we examined modulation of these proteins in normal rat mammary epithelium and found similar changes. Quantitative data from five of these proteins, cyclin A, BUBR1, Cdc2, PTTG, and TPX-2, which were similarly modulated in normal mammary epithelium (Fig. 2A) and mammary tumor tissues (Fig. 2B), are presented. In the supplementary figure, we show the IHC results with these five proteins in vorozole-treated tumors.

Although our primary objective was to examine genes most directly related to the mechanism of action of vorozole, the effects of the aromatase inhibitors on estrogen-responsive genes not directly related to the mechanism of action of vorozole were also examined. These are genes that have previously been shown to be responsive to estrogen stimulation or withdrawal. These genes and their relative decreases in expression include: (i) TFF1 (PS2) −4.51 (P < 0.001); (ii) TFF-3 −2.1 (P < 0.001); (iii) Greb 1 −6.0 (P < 0.0001); (iv) progesterone receptor −6.1 (P < 0.0001); (v) PDZK1 −2.52 (P < 0.001). When we examined roughly the 1,000 genes whose expression were modulated (P < 0.01) by vorozole, TTF1, Greb-1, and the progesterone receptor were among the 10 genes whose expression was most highly decreased.

As presented in the background, our laboratories have extensively examined the effects of the aromatase inhibitor vorozole in the ER⁺ rat model of breast cancer (11–14). These studies have shown the efficacy of aromatase inhibitor, and are similar to prior studies showing that ovariectomy, various selective estrogen response modifiers (SERMs), and even pregnancy yielded similar results in this model to those observed in humans. It was also shown in this model that similar changes in pharmacodynamic markers (e.g., estrogen levels, serum IGF-I, and effects on proliferation) were comparable with those achieved in humans with aromatase inhibitor treatments (13, 14). Therefore, we feel that MNU-induced mammary cancers are a reasonable model for ER⁺ breast cancer in women. In fact, the major goal of this study was to validate gene changes in this model relative to those observed in humans. A second objective was to determine whether we could observe these same effects in normal rat mammary epithelium. Although there are many similarities between rat mammary and human ER⁺ tumors, there are substantial differences. For example, roughly 50% of these tumors have mutations in HaRas, which are virtually not seen in humans (29). Furthermore, the tumors are near diploid (again less common in humans), and induced tumors arise quickly in young rats. Because this is a reasonable model for ER⁺ human breast cancer and clearly responds to the same class of agents, the present studies determined whether gene changes observed in this model were similar to those observed in humans. The strength of the approach is that the human samples selected were looking at a slightly more homogeneous group of tumors (ER⁺ cancers only) and that the aromatase inhibitors were given as monotherapies. One potential limitation is that we did not use the same agents (anastrozole or letrozole in human vs. vorozole in rats). However, it is assumed that agents of the same class that work by the same mechanism would yield similar set of gene expression changes, as was observed in the study TABM-180 comparing anastrozole and letrozole in humans (24). Rat tumor samples for biomarker studies were treated for a period of 5 days to parallel a presurgical model in humans. We have previously used this method to look at changes in biomarkers in this model and found that at a 5-day time point one does not observe significant regression compared with controls (14, 30).

We observed that vorozole-treated mammary cancers showed substantial differences in gene expression when compared with controls (Fig. 1A). The effects of vorozole on various genes in the APC chromosome condensation in the prometaphase pathway are shown in Fig. 1B. The APC complex is involved in stimulating cells to progress through mitosis by supporting inhibition of CDK1 via the degradation of cyclin B. Genes in this pathway and the associated spindle assembly complex that were modulated in this study with vorozole include: aurora kinase A, CDK1, cyclin B, cyclin A, CDC2, TPX-2, BUB1, and BURB1. In fact, many of these genes are a portion of the genes with altered expression in the lower one third of Fig. 1A. All of these genes were modulated in many of the human aromatase studies (Tables 2 and 4). The decreased expression of genes in this pathway would seem to be a prime candidate for the profound decrease in cell proliferation observed in rat and human mammary cancers treated with aromatase inhibitors (14, 15).

Comparisons were made both with human clinical samples using aromatase inhibitors as well as with gene changes associated with estrogen deprivation in MCF-7 cells in vitro (Table 3; ref. 27). The reason for examining the latter was to determine whether gene changes observed in cell culture might serve as biomarkers and mechanistic leads. We found 13 genes differentially expressed with vorozole treatment in the animal model most clearly related to cell proliferation and which were shown to be modulated by E2F. Of these 13 highly significant proliferation genes seven showed E2F1-binding sites and four contained ER-α–binding sites (Table 3). Interestingly, a number of these genes have proven to be overexpressed in ER-α receptor cells, which have become resistant to SERMs and aromatase inhibitor (31). Thus, the expression of certain of the E2F-related genes become independent of continued estrogen stimulation and may account in major part for the hormone resistance of these cells.

In addition, we were able to confirm the genes that were identified in Tables 2 and 3 in an independent set of human samples treated with anastrozole (Table 4; ref. 26). This was a large study using 81 pairs of samples taken at baseline or after 2 weeks of treatment with anastrozole. One interesting aspect of this latter study is that the magnitude of the decreases even with baseline samples and treated samples were often relatively small. Thus, a 1.5-fold decrease should represent a decrease in gene expression of roughly 33%; and many of the changes are less than that; albeit highly significant statistically.

On the basis of the consistent gene expression decreases in both rat and human studies and their integral role in the mechanism of the efficacy of aromatase inhibitors, we investigated the effects of vorozole on the protein levels using IHC of the proliferation-related genes. We were initially wary of examining proteins for which the genes were reduced only 40% to 55%. We examined 11 genes that were modulated at the gene level by IHC and found eight of 11 of the proteins were significantly modulated in vorozole-treated tumors. In addition, we examined modulation of these same proteins in normal mammary epithelium. In this case, we observed statistical changes in six of 11 proteins, and present the data with the five overlapping proteins. The reason we examined these changes in normal epithelium is that often changes in normal breast epithelium have been used in phase II breast cancer chemoprevention trials (32). In fact, Fabian and colleagues have observed decreases in Ki67 in at-risk breast epithelium treated with letrozole (33). The present molecules would offer additional related biomarkers. Although significant agreement in normal epithelium when using a hormonal agent was observed, it is not known whether this will extrapolate to use with other classes of agents. Nevertheless, based on gene arrays we could determine both relevant pathways and potential candidates that were confirmed by IHC. Whether IHC examination or real-time PCR (RT-PCR) would be a better approach for measuring these changes in limited clinical samples (e.g., biopsy) is a separate and independent question. Obviously, there are additional questions about whether one can observe similar IHC differences in human samples in which the gene expression differs.

Finally, alterations in expression of five genes that were associated with estrogen deprivation were observed. These genes, although estrogen dependent, may not be directly involved in the efficacy of the aromatase inhibitors (34), and include progesterone receptor, TFF1, TFF3, Greb 1, and PDZK1. Although these genes are less likely to be directly associated with the efficacy of the aromatase inhibitors, they are a direct confirmation of altered expression of genes known to be associated with estrogen withdrawal. Therefore, they function as an independent confirmation that the animal model is yielding similar results to those obtained in humans.

The present studies show that a relevant model that yields similar preventive and therapeutic effects to that obtained in humans can yield identical RNA expression changes. Thus, the animal model revealed both relevant pathway changes, but perhaps as importantly, revealed a host of specific genes that were directly relevant to those found in clinical samples and might serve as biomarkers of efficacy. As important, it demonstrates that the animal models yield biomarker changes in the same pathways and individual genes that are observed in similarly treated human cancers. Furthermore, it yielded potential proteins (confirmed by IHC), which were modulated both in mammary cancers and normal mammary epithelium.

No potential conflicts of interest were disclosed.

Conception and design: Y. Lu, M. You, C.J. Grubbs, R.A. Lubet

Development of methodology: Y. Lu

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Lu, W. Wen, A.M. Bode, C.J. Grubbs

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Lu, Z. Ghazoui, P. Liu, P.T. Vedell, R.A. Lubet

Writing, review, and/or revision of the manuscript: Y. Lu, M. You, Z. Ghazoui, P.T. Vedell, W. Wen, A.M. Bode, R.A. Lubet

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Lu, M. You

Study supervision: M. You

This study was supported by National Cancer Institute, contract number HHSN261200433001C and HHSN261200433008C.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.