A gene transcription signature associated with hormone independence in a subset of both breast and prostate cancers

In 2006, on the order of 234,000 men and 213,000 women were diagnosed with prostate cancer and breast cancer, respectively, and about 27,000 men and 41,000 women died (American Cancer Society statistics). Steroid hormone receptor signaling has been linked to all stages of prostate and breast carcinogenesis [1, 2]. Initial treatment of clinically localized prostate cancer (PCA) and invasive breast cancer (IBC) usually involves surgical removal of the cancerous tissue or radiation therapy. The clinical use of adjuvant anti-androgen therapy in PCA and of anti-estrogen therapy in IBC has aided greatly in prolonging or preventing disease recurrence, as the majority of these cancers, at least initially, depend upon their associated hormones for growth. However, significant fractions of PCA and IBC either initially present as hormone independent or develop hormone independence over the course of anti-hormone therapy [3, 4]. In the case of IBC, hormone independence correlates closely with expression of the estrogen receptor alpha (ER), with 30–35% of IBC being ER-negative (ER-) and exhibiting no requirement of estrogen for growth [3, 5]. Except for the fraction of ER- IBC that express HER2/neu, no targeted therapy is currently in widespread use for ER- IBC. In the case of advanced PCA, androgen ablation therapy effectively results in tumor regression over the short-term; in most cases, however, the recurrence of highly aggressive and metastatic prostate cancer that is resistant to hormone therapy occurs as a result [1, 4].

Breast and prostate cancers share much in common with each other, in that they both manifest as either hormone dependent or independent. The hypothesis explored in this present study is that the molecular mechanisms of acquirement of hormone independence are similar between IBC and PCA [4]. Global gene expression profiling studies, carried out in breast and prostate cancers separately, indicate that on the order of hundreds or even thousands of genes might be involved in hormone independence in each disease [6–8]. If a select set of genes common to hormone independent breast and prostate cancers could be identified, it might be indicative of a core transcriptional program on which attention could be focused. The main strategy of this study was to look for patterns of enrichment, i.e. to look for a non-random amount of shared overlap between distinct sets of genes associated separately with either breast or prostate cancers. Such a pattern of enrichment may involve only a fraction of the genes from each cancer type and yet may hold biological and clinical significance.

One of the "holy grails" of both breast and prostate cancer research is to determine how these cancers acquire hormone independence and how best to treat them when they do. A number of molecular mechanisms have been postulated to explain resistance to hormone therapy in prostate cancer. One class of resistant prostate tumors continues to rely upon androgen receptor (AR) signaling through a number of means, including: over-expression of the AR gene, through DNA amplification or some other mechanism [14, 23]; "promiscuous" point mutation in AR, allowing the receptor to be activated by steroids other than androgen, including anti-androgens and estrogen [24]; and ligand-independent activation of AR, mediated by oncogenes such as ERBB2 or HRAS [25]. While tumors that fall under the above may often be referred to as "androgen independent" [4], for the purposes of this present study we must draw a distinction between the above class of hormone therapy resistant prostate cancer and a second class, which bypasses AR function completely and does not rely upon androgens for growth. This second class is what appears to be involved in the gene signature of hormone independence uncovered here as being common to both ER- breast cancer and prostate cancer that is completely AI.

The specific alternative signaling pathways that allow prostate tumors to bypass AR have been somewhat elusive, one candidate pathway possibly involving BCL2 [4]. This present study has implicated EGFR signaling has playing an important role in bypassing AR. While transcriptional targets of EGFR were enriched in AI prostate cancer (Figure 4), EGFR mRNA itself did not appear elevated in AI cell lines. Neither was EGFR mRNA consistently elevated in the clinical PCA samples that exhibited an ER-/AI molecular phenotype (Figure 3). However, EGFR protein itself is elevated in AI over AS cell lines [26]. Recent studies have found EGFR protein expression in tumor tissues to be strongly associated with hormone refractory status [27–29]. The EGFR tyrosine kinase inhibitor gefitinib causes cell cycle arrest and initiates apoptosis in primary PCA cultures and in PCA cell lines, including DU145 and PC3 as well as LNCaP [26]. In studies using the AI cell line PC3, suppression of EGF-R signaling reduced the incidence of prostate cancer metastasis in nude mice [30].

This present study indicates that a subset of primary PCA presents as hormone independent, prior to patient treatment with adjuvant anti-androgen therapy (Figure 3). It has been previously thought that a sub-population of androgen-resistant cells might coexist with androgen-dependent cells within the tumor, and that anti-androgen therapy would therefore kill off the dependent cells and leave the resistant cells to thrive [1]. This study lends support to an adjuvant therapy strategy of combining EGFR inhibitors with anti-androgens. One question to consider in designing clinical trials testing this treatment regimen is whether PSA recurrence would be a suitable endpoint, as AI prostate cancer may not express PSA (Figure 3). The ability to identify a hormone independent subset of primary breast cancer using ER as a biomarker has implications for selecting the course of adjuvant treatment [3]. If a subset of primary prostate tumors could be diagnosed in the clinic as being hormone independent, it could warrant more aggressive treatment with alternative therapies to anti-androgens. Individual genes in the list of 81 in the gene signature of hormone independence might be good candidates for prognostic markers in PCA, or several genes in the signature could perhaps be used together.

In other studies, gene expression profiling has been carried out on hormone refractory metastases of prostate cancer [31, 32]. One issue with comparing results derived from those datasets with this present study's gene signature of hormone independence is that, as discussed above, these metastases likely represent several mechanisms of hormone therapy resistance, not simply the use of EGFR-mediated bypass of AR function, as appears to be manifested in the AI prostate cell lines analyzed here. Genes expressed in the EGFR-dependent AI subtype may not be uniformly expressed in all varieties of hormone refractory cancers. The dataset from ref [32] was analyzed here in the context of the datasets used in this study; when considering the genes high in hormone-refractory prostate metastases compared to PCA, no significant overlap of these genes was observed with the genes high in the AI prostate cell lines or the ER- breast tumors (results not shown). A number of expression profiling studies using prostate tumor xenografts acquiring resistance to hormone therapy have been carried out [14, 33–36]; many of these studies appear to represent cancers that develop hypersensitivity to androgen pathway activation, rather than androgen pathway independence. The hormone independent gene signature of this present study did not show coordinate expression in a profile dataset from Chen et al. [14] of hormone therapy-resistant prostate tumors xenografts (results not shown); as these tumor xenografts uniformly up-regulated AR, it could be presumed that these xenografts represented a model of increased sensitivity to androgen levels.

One limitation with this present study is the small number of prostate cancer cell lines for which gene expression profile data was available (five AS and three AI). Profiling studies in breast cancer indicate that there are at least two subtypes of ER- IBC, a HER2+ subtype and a "basal" subtype [6, 7, 12]. Interestingly, recent studies indicate that a subset of ER- breast cancer may rely upon the androgen pathway rather than the estrogen pathway [37, 38], though this does not appear to represent the subset of ER- considered here, as androgen-regulated genes were not enriched in the ER- gene signature of Figure 2 (results not shown). If more prostate cancer cell lines were profiled, it might uncover additional subtypes of AI PCA to the EGFR-dependent subtype uncovered here. The basal subtype of ER- IBC also appears to rely upon EGFR signaling (Figure 4). A number of clinical trials testing the efficacy of EGFR tyrosine kinase inhibitors (TKIs) in breast cancer treatment are underway. Initial phase II studies have suggested that the EGFR TKIs do not have a high efficacy in a heavily pre-treated population of patients with metastatic breast cancer; however, in patients with hormone therapy-resistant ER+ tumors, EGFR inhibition does appear to have a significant therapeutic effect [39]. In at least one small trial, there was evidence of only minimal efficacy in advanced, metastatic ER- breast cancer [39], though one could make the conjecture that anti-EGFR therapy might be more effective in less advanced cancers.

The 81 genes in the signature of hormone independence common to prostate and breast is a much shorter set compared to the hundreds of genes associated separately with either prostate or breast (Figures 1 and 2). One may expand the set of 81 somewhat by using slightly less stringent statistical cutoffs. At the same time, the set of 81 may provide a good starting point for further study. Interestingly, a number of the 51 genes higher in AI prostate and ER- breast were previously associated with the immune response, including ELF4, GBP1, CXCL2, IL6, IL7R, and IL15. IL6 (interleukin 6) in particular – indicated here to be a transcriptional target of the EGFR pathway (Figure 4) – has been shown to promote prostate tumor growth and to play a role in the interaction between epithelial and stromal cells in prostate cancer [40]. One next step in studying these genes would likely be validation of their expression patterns in breast and prostate tissues or cell lines, using some alternative technique from microarrays, such as westerns or quantitative RT-PCR; it is expected, however, that most of the genes in the set of 81 would validate, as their expression patterns were observed in multiple profile datasets generated on different microarray platforms, which in itself could be considered validation [41].

In conclusion, the hope for this study is that it may aid in the development of therapy regimens to target the subset of breast and prostate cancers that up until the present have been the most difficult to treat.

The gene expression profile datasets used in this study were all publicly available. From the 66 PCA profiles described in the study by Yu et al. [15], 60 were available for this present study (collection for this dataset was facilitated by A.M. Chinnaiyan and the Oncomine team). Recent evidence emerged that the MDA-MB-435 cell line was not breast but melanoma [42] and so the MDA-MD-435 profiles in the Bild dataset [11] were removed from the analysis. Gene expression values in each dataset were log-transformed. Gene expression values in the clinical breast tumor and cell line profile datasets were centered on the centroid mean of ER- and ER+. Values in the Zhao prostate cell line profile dataset [8] were centered on the centroid mean of AS and AI. Clinical prostate tumor datasets were transformed to standard deviations from the median. For the androgen dataset from Chen et al. [14], expression values within the AR+ group of samples were transformed to standard deviations from the mean; values within the vector group of samples were separately transformed. Expression values in the Creighton MCF-7 dataset [22] were centered on the mean of the MCF-7/lt-E2 control group.

As the expression profile datasets were generated on different platforms, and as many of the genes represented were measured on multiple probes in any one dataset, a method to select the optimal probe to represent each gene in an unbiased fashion was used when joining multiple datasets. For the Affymetrix array datasets, the probe with the greatest variation across samples represented the gene. For the cDNA microarray datasets (Zhao cell line and Lapointe PCA), the probe with the most unflagged values across samples, followed by the probe with the greatest variation, represented the genes. The Entrez Gene identifier was used in mapping genes across datasets. Two-sample t-tests determined significant differences in gene expression between groups of samples. For the Chen androgen dataset, the Pearson's correlation between gene expression and the log of the concentration of R1881 determined significance of R1881 induction. Expression values were visualized as heat maps using the Cluster [43] and Java TreeView software [44]. Prior to heat map generation, genes were manually sorted using Microsoft Excel in order to highlight gene groups of interest.

The one-sided Fisher's exact test determined significance of overlap between any two distinct sets of genes. Q1–Q2 enrichment analysis [15] was carried out essentially as described in ref [45]. For "stratifying" a set of profiles in a given dataset on the basis of a pre-defined expression pattern (e.g. stratifying the clinical PCA profiles using the ER-status gene signature in Figure 3), each gene involved in the pattern was represented as "1" or "-1" (for up or down, respectively), and the Pearson's correlation coefficient was computed between the pattern and each individual profile (with significance by two-sided t-test).