Breast cancer presents as either estrogen receptor α (ERα) positive or negative, with ERα+ tumors responding to antiestrogen therapy and having a better prognosis. By themselves, mRNA expression signatures of estrogen regulation in ERα+ breast cancer cells do not account for the vast molecular differences observed between ERα+ and ERα− cancers. In ERα− tumors, overexpression of epidermal growth factor receptor (EGFR) or c-erbB-2, leading to increased growth factor signaling, is observed such that mitogen-activated protein (MAP) kinase (MAPK) is significantly hyperactivated compared with ERα+ breast cancer. In ERα+/progesterone receptor–positive, estrogen-dependent MCF-7 breast cancer cells, we stably overexpressed EGFR or constitutively active erbB-2, Raf, or MAP/extracellular signal-regulated kinase kinase, resulting in cell lines exhibiting hyperactivation of MAPK, estrogen-independent growth, and the reversible down-regulation of ERα expression. By global mRNA profiling, we found a “MAPK signature” of ∼400 genes consistently up-regulated or down-regulated in each of the MAPK+ cell lines. In several independent profile data sets of human breast tumors, the in vitro MAPK signature was able to accurately distinguish ER+ from ER− tumors. In addition, our in vitro mRNA profile data revealed distinct mRNA signatures specific to either erbB-2 or EGFR activation. A subset of breast tumor profiles was found to share extensive similarities with either the erbB-2-specific or the EGFR-specific signatures. Our results confirm that increased MAPK activation causes loss of ERα expression and suggest that hyperactivation of MAPK plays a role in the generation of the ERα− phenotype in breast cancer. These MAPK+ cell lines are excellent models for investigating the underlying mechanisms behind the ERα− phenotype. (Cancer Res 2006; 66(7): 3903-11)
- breast cancer
- estrogen receptor
- gene expression profiling
- ERα-negative breast cancer
Breast cancer can present as either estrogen receptor α (ERα) positive or negative. Whereas ERα+ tumors have a better prognosis and respond to antiestrogen therapy ( 1– 4), their ERα− counterparts have a worse prognosis, are resistant to antiestrogens, and frequently present with greatly elevated growth factor receptor expression and/or signaling with resultant p42/44 mitogen-activated protein (MAP) kinase (MAPK) signaling ( 5– 15). Many questions exist as to the biological pathways and alterations involved in determining whether ERα is expressed and how it is regulated during breast cancer evolution. Available evidence suggests that most ERα− breast cancers arise from ERα+ cells that stop expressing the receptor ( 16). Proposed mechanisms for the origin of ERα− breast cancers include that of pressures being exerted on ERα+ cells by estrogen withdrawal ( 17), hypoxia ( 18), or overexpression of epidermal growth factor (EGF) receptor (EGFR) or c-erbB-2 due to hyperactivation of MAPK ( 19), as well as reversible chromosomal alterations, such as methylation of the ERα promoter ( 20).
The hypothesis that we explore here is that aberrant growth factor signaling resulting in hyperactivation of MAPK will induce a gene expression profile reflective of the hyperactive MAPK and that this MAPK profile will be indicative of breast cancer behavior. Using ERα+ MCF-7 cells engineered for constitutive activation and/or overexpression of different components of growth factor signaling pathways, we have developed models to mimic the MAPK hyperactivation seen in ERα− breast cancer ( 19, 21– 23). These models have shown that hyperactive MAPK leads to the loss of ERα expression ( 19). We have also shown that the indirect activation of nuclear factor-κB due to MAPK hyperactivity plays a role in down-regulating ERα expression ( 24).
In this study, we have investigated the effects of hyperactivation of growth factor signaling pathways, both constitutively and acutely, on gene expression profiles in MCF-7 breast cancer cells. We have contrasted gene expression in our hyperactive MAPK cell line models to gene expression in control-transfected MCF-7 (coMCF-7) cells. A MAPK gene expression profile was generated, as well as factor-specific profiles: erbB-2, EGFR, Raf, and MAP/extracellular signal-regulated kinase kinase (MEK). As expected, we find that a subset of the MAPK genes are inversely regulated compared with estrogen-regulated genes, a likely consequence of the ERα down-regulation that has occurred in these cell lines. A small number of genes are similarly affected, suggesting that both estrogen and MAPK action involves some common effectors. In addition to the effects on estrogen-regulated genes, we also found that hyperactivation of MAPK alters the expression of a number of genes involved in various aspects of cell growth, survival, metabolism, etc. To determine the physiologic/clinical relevance of this MAPK profile generated from cell lines engineered using the same parental cell line, this MAPK profile was compared with profiles generated from ERα+ versus ERα− breast tumors. The MAPK profile predicted very well the ERα status of clinical breast tumors from four different clinical studies. In a similar manner, both the erbB-2 and EGFR profiles from the cell lines correlated well with profiles generated from erbB-2 and EGFR-overexpressing breast tumors, respectively. Thus, these MCF-7-based cell lines representing up-regulated growth factor signaling via hyperactivation of MAPK model very closely ERα− breast cancer.
Materials and Methods
Cell lines and cell culture. All cell lines used in the study have been described earlier ( 19, 24). coMCF-7 cells were cultured in the presence of estrogen in antibiotic-free, antimycotic-free, modified IMEM with l-glutamine, without gentamicin sulfate, with phenol red (Gibco/Invitrogen, Carlsbad, CA), and with 10% heat inactivated fetal bovine serum (Valley Biomedical, Winchester, VA). The following cell lines were maintained in the absence of estrogen: coMCF-7/lt-E2 (coMCF-7 adapted long term for estrogen-independent growth), constitutively active Raf-transfected MCF7 [(ca)Raf], constitutively active c-erbB-2 transfected MCF7 [(ca)erbB2], constitutively active MEK-transfected MCF7 [(ca)MEK], and EGFR (ligand-activatable EGFR-transfected MCF7). Cells grown in the absence of estrogen were cultured in antibiotic-free, antimycotic-free, modified IMEM with l-glutamine, without gentamicin sulfate, without phenol red (Gibco/Invitrogen) and 10% heat-inactivated, charcoal-stripped calf serum (Valley Biomedical). For EGF treatment of the EGFR-overexpressing cells, EGFR cells were treated with 10 ng/mL EGF (human recombinant, Upstate, Charlottesville, VA) for 8 hours before harvesting for RNA analysis (EGFR + EGF). For the EGFR + EGF time course to determine protein expression, cells were treated with EGF for 8 hours, EGF was removed, and untreated medium was added for the duration of the time course. Cells were incubated in a 37°C, 5% CO2 forced-air humidified incubator. All cell lines were used at 12 to 17 passages after thaw.
Affymetrix microarray profiling. Triplicate RNA samples were prepared for each cell line using TRIzol reagent (Invitrogen) according to instructions of the manufacturer, and 20 μg total RNA at a minimum concentration of 1.3 μg/μL was used for further processing before microarray analysis. RNA for microarray analysis was labeled and hybridized according the Affymetrix protocol (Affymetrix GeneChip Expression Analysis Technical Manual, Revision 3) by the University of Michigan Comprehensive Cancer Center Affymetrix and cDNA Microarray Core Facility. Gene expression patterns were determined using Affymetrix Genechip U133A Arrays. Arrays were normalized and compared using DNA-Chip Analyzer software ( 25). Array data has been deposited in the public Gene Expression Omnibus database, accession no. GSE3542.
Microarray data analysis. Gene expression values were log-transformed. Breast tumor data sets were transformed to SDs from the mean. For the estradiol data set from Rae et al. ( 26), values within each cell line were transformed to SDs from the cell line mean. Two-sample t tests were done as criteria for determining significant differences in mean gene mRNA levels between groups of samples. Global views of the variation in gene expression among cell lines were obtained using principal components analysis as described in ref. 27. Expression values were visualized as color maps using the Cluster and Java TreeView software ( 28). Gene ontology annotation terms were obtained and searched for enrichment within gene sets as described in ref. 27.
As the estrogen data set from ref. 26 was generated on the same U133A platform as the MAPK profile data set, a mapping between the two data sets was made using the mRNA probe identifier. For the five breast tumor profile data sets, which were generated on different platforms from U133A, a mapping between the tumor data sets and the MAPK data set was made using the common gene name. Where a gene was represented more than once in a tumor data set (e.g., probes for alternatively spliced mRNAs), the probe that showed the most significant difference between ERα+ and ERα− tumors (either direction) was used in the mapping. For the van't Veer ( 29) and Huang ( 30) data sets, profiles for tumors with only moderate protein expression of ERα (van't Veer: >0% and <80%, Huang: + and ++) were removed from the analysis.
When classifying tumor profiles as ERα+ or ERα− using the MAPK mRNA signature ( Fig. 4), an “idealized” MAPK pattern reflecting the direction of the expression differences was constructed (+1 for genes up-regulated in the MAPK signature; −1 for down-regulated genes). The Pearson correlation was used as a measure of similarity (correlation > 0) or dissimilarity (correlation < 0) of each tumor profile with the MAPK pattern. In a similar manner, tumor profiles were compared with (ca)erbB-2-specific and EGFR + EGF–specific patterns ( Fig. 5), with the significance of correlation assessed by t statistic.
Real-time PCR. Triplicate RNA samples from each cell line were used: one from the microarray analysis and two prepared at separate times over the course of several passages of the cells. RNA was DNase-treated using TURBO DNase (RNase-free; Ambion, Austin, TX) following the protocol of the manufacturer. The DNase-treated RNA was then subjected to reverse transcription using TaqMan Reverse Transcription Reagents (Applied Biosystems, Foster City, CA). RNA was denatured at 65°C for 5 minutes and a quick chill on ice before reverse transcription. Reverse transcription reactions were carried out in a final volume of 10 μL with 0.1 μg RNA, 1× reverse transcriptase buffer, 5.5 mmol/L MgCl2, 500 μmol/L of each deoxynucleotide triphosphate, 2.5 μmol/L random hexamers, 0.4 units/μL RNase inhibitor, and 3.125 units/μL MultiScribe Reverse Transcriptase. Minus reverse transcription reactions were also done. Reverse transcription reactions were incubated at 25°C for 10 minutes, 37°C for 60 minutes, 95°C for 5 minutes, and 4°C hold in a thermocycler. Real-time PCR was carried out to determine relative quantification using the comparative CT method (2−ΔΔCT; ref. 31). Plus reverse transcription reactions were run in triplicate. Minus reverse transcription reactions and no template controls were run in duplicate to verify all amplification is due to cDNA. Real-time PCR for ERα and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was done on the iCycler IQ Real-Time PCR Detection System (Bio-Rad, Hercules, CA) in a 25 μL reaction volume of 1× iQ SYBR Green Supermix (Bio-Rad), 5 ng reverse transcription reaction, and 300 nmol/L of each ERα primer or 200 nmol/L of each GAPDH primer. Reactions were cycled for 10 minutes at 95°C, and 40 cycles of 95°C for 15 seconds/60°C for 1 minute. For ERα, primer sequences were as follows: forward, 5′-CCACCAACCAGTGCACCATT; reverse, 5′-GGTCTTTTCGTATCCCACCTTTC ( 32); for GAPDH: forward, 5′-CACCAGGGCTGCTTTTAACTCTGGTA; reverse, 5′-CCTTGACGGTGCCATGGAATTTGC (Bio-Rad Tech Note 2804). The following TaqMan Gene Expression Assays (Applied Biosystems) were used: PDCD4 Hs00377253_m1, TOB1 Hs00271739_s1, ANGPTL4 Hs00211522_m1, ETV5 Hs00231790_m1, GREB1 Hs00536409_m1, CXCL12 Hs00171022_m1, MYB Hs00193527_m1, and GAPDH Hs99999905_m1. Real-time PCR using TaqMan Gene Expression Assays was carried out in a 25 μL reaction volume of 1× TaqMan Universal PCR Master Mix (Applied Biosystems), 5 ng reverse transcription reaction, and 1× TaqMan Gene Expression Assay. Reactions were cycled at 95°C for 10 minutes, and 40 cycles of 95°C for 15 seconds/60°C for 1 minute on the Applied Biosystems 7900HT real-time PCR system. Results were calculated using the comparative CT method (2−ΔΔCT). Results are normalized to GAPDH and relative to the coMCF-7/lt-E2 cells.
Gel electrophoresis and Western blotting. Whole cell protein lysates were prepared in modified gold lysis buffer, as described previously ( 19), from cells grown to ∼80% confluence. Protein concentrations were obtained using the BCA protein assay kit (Pierce). Protein was denatured in Laemmli sample buffer and 5 or 10 μg were loaded onto Tris/glycine PAGEr Gold Precast gels (Cambrex, Rockland, ME) and electrophoresed in 1× Tris/glycine/SDS buffer. Gels were transferred to Hybond-P polyvinylidene difluoride membrane (Amersham, Piscataway, NJ) in 1× Tris/glycine buffer [25 mmol/L Tris, 192 mmol/L glycine, 20% MEOH (pH 8.3); Bio-Rad]. Following transfer, the membranes were blocked in TBS-T + 5% bovine serum albumin [BSA; 10 mmol/L Tris-HCl (pH 8.0), 150 mmol/L NaCl, 0.1% Tween 20, and 5% BSA] and incubated in antibodies diluted in TBS-T + 5% BSA. The following primary antibodies were used: ERM (H-100; ETV5; Santa Cruz Biotechnology, Santa Cruz, CA), Pdcd-4 (C-16; Santa Cruz Biotechnology), and mouse monoclonal antibody [4B1] to TOB (Abcam, Cambridge, MA). The following secondary antibodies were used: donkey anti-goat IgG horseradish peroxidase (HRP; Santa Cruz Biotechnology), ECL antimouse IgG HRP–linked whole antibody (from sheep; Amersham), and ECL antirabbit IgG HRP–linked whole antibody (from donkey; Amersham). Membranes were washed with TBS-T. Chemiluminescent detection was accomplished using SuperSignal West Pico substrate (Pierce, Rockford, IL) following the protocol of the manufacturer. Membranes were stripped using Restore Western blot stripping buffer (Pierce) for 15 minutes at room temperature and reprobed with actin (I-19)-HRP–conjugated antibody (Santa Cruz Biotechnology) to verify even loading.
Widespread changes in the transcriptome of breast cancer cells result from hyperactivation of MAPK pathway initiators EGFR, erbB-2, Raf, or MEK. The MAPK signaling pathway may be initiated by activation of either the EGFR or erbB-2 growth factor receptors from which the signal is channeled via Ras, Raf, and MEK, ultimately resulting in activating phosphorylation of MAPK and gene transcription ( Fig. 1A ). We had previously generated MCF-7 cell lines stably overexpressing constitutively active Raf-1, constitutively active MEK, constitutively active c-erbB-2, or ligand-activatable (i.e., +EGF) EGFR as models of overexpressed growth factor signaling ( 19, 21– 23), as well as control-MCF-7–transfected cells (coMCF-7) and control-vector–transfected cells long-term adapted for estrogen-independent growth (coMCF-7/lt-E2). The four engineered cell lines, called (ca)Raf, (ca)MEK, (ca)erbB-2, and EGFR + EGF, all show hyperactivation of MAPK, estrogen-independent growth, and loss of ERα expression ( 19). To investigate alterations in gene expression in general, and to generate a hyperactive MAPK gene expression profile in breast cancer cells in particular, Affymetrix U133A oligonucleotide arrays (composed of 22,283 probe sets designed to measure human mRNAs) were used to generate mRNA expression profiles of the (ca)erbB-2, (ca)MEK, (ca)Raf, and EGFR + EGF cell lines, as well as of the two control cell lines, coMCF-7 and coMCF-7/lt-E2.
Using principal components analysis of all 22,283 probe sets, a pair of coordinates was determined for each mRNA profile to construct a two-dimensional view that reflects the relative global similarity or dissimilarity of the profiles to each other. On this two-dimensional view, profiles within each experimental group formed distinct clusters from the other groups ( Fig. 1B), which is a good indication both of reproducibility for replicate profiles within each group and of widespread differences in gene expression existing between the groups. The coMCF-7/lt-E2 and coMCF-7 profile groups clustered relatively close together, indicating that numerous similarities in the expression of individual mRNAs exist between these two groups relative to the other groups.
We did a supervised clustering of all 6,779 genes (mRNA transcripts) showing significant up- or down-regulation (P < 0.01) relative to the relevant control vector–transfected cell line, coMCF-7/lt-E2, in at least one of the five other cell lines. The number of genes up-regulated and the similarity of these genes between the MAPK+ cell lines are shown by Venn diagram in Fig. 1C. The expression values for these genes were visualized as a color map ( Fig. 2 ), revealing both expression patterns that were unique to a particular experimental cell line and patterns that were common to two or more cell lines. For each hyperactive MAPK cell line, on the order of 500 genes were differentially expressed primarily in that cell line, showing little or no change relative to control in the other cell lines.
An mRNA expression signature of MAPK activation in breast cancer cells in vitro. Whereas hyperactivation or overexpression of different MAPK pathway initiators (erbB-2, EGFR, Raf, and MEK) was observed to result in different mRNA expression patterns, a number of genes were found to be consistently up-regulated or down-regulated in each of the MAPK+ cell lines ( Fig. 2). We focused on these genes as a set of particular interest, as they would likely represent a common mRNA signature of activation of the MAPK pathway regardless of how the pathway was initiated. Of 3,437 genes (transcripts) up-regulated (P < 0.05) in the (ca)erbB-2 cell line relative to coMCF-7/lt-E2, 3,064 genes up-regulated in (ca)MEK, 1,920 genes up-regulated in EGFR + EGF, and 1,598 genes up-regulated in (ca)Raf, 153 were common to all gene sets, where on the order of 3 might have been in common had the different profile groups no relation to each other ( Fig. 1C). Of the 153 genes, 26 were also up-regulated (P < 0.05) in coMCF-7 relative to MCF-7/lt-E2. Out of 3,375 genes down-regulated (P < 0.05) in the (ca)erbB-2 cell line, 2,965 genes down-regulated in (ca)MEK, 2,474 genes down-regulated in EGFR + EGF, and 2,899 genes down-regulated in (ca)Raf, 316 were in common, 71 of which were also down-regulated (P < 0.05) in coMCF-7. We designated the entire set of 469 up- or down-regulated mRNA transcripts (representing 393 uniquely named genes) as a MAPK mRNA signature for further analysis.
We examined the MAPK signature for significantly enriched (i.e., overrepresented) gene ontology annotation terms (P < 0.05) for functional gene classes of possible interest. Of the 123 uniquely named genes up-regulated in the MAPK signature, five (HLA-B, HLA-C, HLA-E, HLA-F, and HLA-G) were annotated as having MHC class I receptor activity function (with 13 genes in this class being represented on the U133A platform, enrichment P = 1.1e−07), 12 (including protein kinase Cα, transforming growth factor β1–induced antiapoptotic factor 1, GADD45A) had annotated roles in cell death (total 388, P = 0.0005), 13 (including estrogen-related receptor α, IFN regulatory factor 1, FOSL2, JUNB, and RELB) had roles in transcription factor activity, 12 (including E2F4, ELF3, ETV4, PDGFA, and VEGF) had roles in the cell cycle, and 30 had roles in signal transduction. Of the 270 uniquely named genes down-regulated in the MAPK signature, 19 were located in the mitochondrion (total 424, P = 0.002), 7 were components of the ribosome, 15 had roles in protein biosynthesis, and four (including ER1, progesterone receptor, NR4A2, and NR5A2) had roles in steroid hormone receptor activity.
Breast cancer cells with MAPK activation down-regulate ERα and numerous estrogen-inducible transcripts. The up-regulation of MAPK activation in breast cancer is associated with ERα− breast cancer. In addition, it is thought to coincide with increasing pressures of estradiol deprivation ( 17, 33). In confirmation of our previously reported MAPK-induced loss of ERα protein in these hyperactive MAPK cell lines ( 19), we observed transcript levels of ERα gene (ESR1) to be significantly down-regulated (P < 0.05) in all four MAPK cell line profiles. We reasoned that a number of other estrogen-inducible genes might be underexpressed along with ESR1 in cells with up-regulated MAPK activity. We obtained a data set from Rae et al. ( 26) of mRNA expression profiles of three different ERα+, estrogen-dependent breast cancer cell lines grown in steroid-depleted medium and stimulated or not with estradiol. A highly disproportionate number of 50 of the 316 transcripts down-regulated in the MAPK signature were also found to be induced (P < 0.01) by estradiol (enrichment P = 1.0e−07; Fig. 3 ). Similarly, a significant number of 15 transcripts repressed by estradiol were also up-regulated (P < 0.01) in the MAPK signature (enrichment P = 0.009).
Verification of the down-regulation of several estrogen-induced genes has been done by analysis of mRNA expression using real-time PCR. Down-regulation of ERα protein expression had been shown previously ( 19), and here we show the down-regulation of ERα mRNA expression as well ( Fig. 3C). In addition, we confirmed the down-regulation of GREB1 mRNA, SDF-1 (CXCL-12) mRNA, and Myb mRNA expression ( Fig. 3C).
The mRNA expression signature of MAPK activation in vitro is similar to mRNA expression signatures of ERα− breast tumors in vivo. As the MAPK mRNA signature was obtained from in vitro cell line models, we went on to examine the expression patterns of these genes in vivo in public mRNA profile data sets of human breast tumors. When examining a data set of 79 tumors from ref. 29, we found that most of the genes that were up-regulated in the MAPK signature were also overexpressed in ERα− breast tumors relative to ERα+ tumors ( Fig. 4A ). Similarly, most of the genes down-regulated in the MAPK signature were underexpressed in ERα− relative to ERα+ tumors, including ESR1 and several estrogen-inducible genes. The MAPK mRNA signature was found to be so similar to the ERα− breast tumor signature that the in vitro MAPK signature expression pattern could be used to distinguish between ERα− and ERα+ tumor profiles with 87% accuracy. When each tumor profile was classified (using the Pearson correlation) as being similar or dissimilar to an “idealized” MAPK pattern (+1 for genes up-regulated in the MAPK signature, −1 for down-regulated genes), tumor profiles that were similar to the MAPK signature were highly enriched (P = 6.3e−12) for ERα− tumors. The MAPK signature was able to classify ERα status with comparable accuracy in three other independent breast tumor data sets [Gruvberger data set ( 34) with 78% accuracy, P = 2.4e−05; Sotiriou data set ( 35), 68% accuracy, P = 4.6e−05; and Huang data set ( 30), 72% accuracy, P = 0.006] as shown in Fig. 4B. Verification at both the mRNA and protein levels of altered regulation of four genes from the MAPK profile not related to down-regulation of ERα was done. Because the EGFR cells do not exhibit constitutive activation of MAPK and require treatment with ligand (+EGF) for hyperactivation of MAPK, we reasoned that up- or down-regulation of protein expression would perhaps require more time than regulation of mRNA expression and thus time course analyses of the EGF regulation of protein expression in the EGFR+ cells was done. Figure 4C shows the up-regulation of angiopoietin-like 4 (ANGPTL4) and ets variant gene 5 (ETV5, an ets-related molecule), and the down-regulation of programmed cell death 4 (PDCD4, a neoplastic transformation inhibitor) and transducer of erbB2 1 (TOB1, a negative regulator of erbB-2) in the hyperactive MAPK cell lines compared with control cells.
As ERα− tumors are characterized by a loss of ERα expression, and as numerous estrogen-inducible genes were observed to be down-regulated in the MAPK signature ( Fig. 3), we wondered whether the similarities shared between the MAPK signature and ERα− tumors might be due solely to the observed down-regulation of ERα in vitro by MAPK activation. However, we found that the in vitro signature of estrogen-regulated gene expression from the Rae data set was unable to distinguish ERα− from ERα+ tumors in the same way as the MAPK signature. Furthermore, when removing from consideration 105 genes that seemed regulated in vitro by estrogen (P < 0.05) in the Rae data in the same direction as MAPK in our data, we found that the remaining 290 MAPK signature genes could still distinguish ERα+ from ERα− tumors (van't Veer data set with 85% accuracy; Gruvberger data set, 76%; Sotiriou data set, 60%; and Huang data set, 65%). Our findings are consistent with those of previous microarray analyses ( 36, 37), implying that the transition of tumors from ERα+ to ERα− status involves more than a loss of estrogen signaling; our analysis here indicates that an up-regulation of the MAPK pathway is also involved.
Some breast tumors have mRNA expression profiles similar to an ERBB2-specific mRNA signature, whereas other profiles are similar to an EGFR-specific signature. In addition to an mRNA signature of MAPK activation, our in vitro profile data revealed an mRNA signature specific to constitutive activation of erbB-2 and a signature specific to EGFR overexpression ( Fig. 2). As erbB-2 and EGFR are two growth factor receptors commonly overexpressed or hyperactivated in breast cancer, we examined the expression of their associated genes in a large panel of 295 breast tumor profiles from ref. 38, 69 of which were ERα− ( Fig. 5 ). Consistent with what was reported above, most of the ERα− tumor profiles had expression patterns for the MAPK signature genes very similar to the patterns observed in vitro; in 56 of these tumors, the similarity was significant by Pearson correlation (P < 0.05). Of these 56 profiles, seven were also significantly similar (P < 0.05) to our in vitro signature of erbB-2 activation, and 27 were similar (P < 0.05) to our in vitro signature of EGFR overexpression, three of these profiles being similar to both the erbB-2 and EGFR signatures. The rest of the ERα− profiles seem to share more similarities to the EGFR signature (although perhaps not to a statistically significant degree) than to the erbB-2 signature ( Fig. 5). It is worthy to note that EGFR mRNA seems overexpressed in most of the ER− tumors, although in the few tumors with similarities to the erbB-2-specific signature, ERBB2 seems overexpressed instead of EGFR. In addition, 50 of the ERα− tumor profiles had significant similarities (P < 0.05) with our in vitro signature specific to constitutively active Raf and 31 were similar (P < 0.05) to the MEK-specific signature.
We went on to examine the gene expression patterns of the 226 ERα+ tumor profiles, 44 of which were found to share significant similarities (P < 0.01) with the MAPK signature ( Fig. 5). Expression of ESR1 (ERα) mRNA was lower in these 44 MAPK+ profiles relative to the rest of the ERα+ profiles. As observed in the ERα− profiles, a subset of the low ERα+, MAPK+ tumor profiles also had significant similarities to the erbB-2-specific mRNA signature (22 profiles at P < 0.05), but not to the EGFR-specific signature; another subset (24 profiles) had significant similarities to the EGFR-specific signature and not to the erbB-2-specific signature; and a small subset (three profiles) had similarities to both EGFR and erbB-2 signatures. Also observed were a number of ERα+ tumor profiles that were not MAPK+ but still shared similarities with the erbB-2 signature ( Fig. 5).
Breast cancer cells with overexpressed EGFR in vitro share extensive transcriptomic similarities with ERα− breast tumors. The MAPK signaling pathway may be activated through either EGFR or erbB-2. We noticed that the EGFR mRNA transcript and not the ERBB2 transcript was more highly expressed in ERα− tumors compared with ERα+ tumors in each of the five breast tumor data sets examined (P < 0.01, each data set; refs. 29, 30, 34, 35, 38). Furthermore, only a small fraction of the ERα− tumor profiles we examined from ref. 38 shared significant similarities with an in vitro mRNA signature specific to erbB-2 signaling, whereas most of the profiles seemed more similar to an EGFR-specific signature ( Fig. 5). We went on to examine the transcriptomic differences between ERα− and ERα+ tumors in the other four tumor data sets to determine whether these differences were in general indicative of aberrant erbB-2 signaling or of aberrant EGFR signaling.
Principal components analysis was first used to define a two-dimensional coordinate space using the 22,283 mRNA transcript expression values from the MCF-7/lt-E2 and EGFR + EGF profiling experiments. As expected, the first principal component, which captures the greatest amount of variation in the data set, separated coMCF-7/lt-E2 from EGFR + EGF profiles ( Fig. 6 ). The differences between these two groups consisted not only of the expression patterns of the MAPK mRNA signature ( Fig. 4), but differences that were specific to the EGFR + EGF profiles over the (ca)erbB-2 profiles ( Fig. 2). Expression profiles from the four independent breast cancer data sets were then mapped onto the principal components analysis coordinate space. Although none of the tumor profiles were used to define the coordinate space, ERα− tumors seemed well separated from ERα+ tumors along the first principal component in each of the four data sets ( Fig. 6), with the ERα− tumors being more closely associated with the EGFR + EGF group. When we repeated this procedure using the (ca)erbB-2 profiles instead of the EGFR + EGF profiles to define the principal components analysis space, we could see no significant separation of ERα− from ERα+ tumors along the principal component defining the differences between (ca)erbB-2 profiles and coMCF-7/lt-E2 profiles ( Fig. 6).
In this study, we attempted to define a “hyperactive MAPK signature” set of genes whose expression is altered in breast cancer cell lines as well as tumors with high MAPK activity and low ERα expression/ERα negativity. Using previously characterized MCF-7-derived cell lines engineered for hyperactivation of MAPK through overexpression and/or constitutive activation of EGFR, erbB2, Raf, and MEK ( 19, 21– 23), we have not only generated gene expression profiles specific for each of these signaling effectors, but have also shown that a number of genes are similarly altered in expression irrespective of the means by which MAPK is hyperactivated ( Fig. 2). Because the cell lines with hyperactive MAPK are characterized by very low levels of ERα expression (similar to ERα− tumors), it is not surprising that several of the genes in the profile are known estrogen-regulated genes, which are oppositely regulated by hyperactive MAPK ( Fig. 3). This is in keeping with the previously shown down-regulation of ERα protein ( 19) and ERα message levels of these cell lines ( Fig. 3C) and also confirms the lack of ligand-independent activation of ER by MAPK ( 19) as, if this were occurring, one would expect the coordinate regulation of estrogen-regulated genes by MAPK. However, a large number of genes with altered expression levels are unrelated to ER expression and function. These include many transcription factors, genes involved in neoplastic transformation, cell survival and viability, angiogenesis, and regulation of mitogenic signaling molecules such as erbB2, all processes expected from the known role of MAPK downstream signaling. Thus, the MAPK signature gene set seems to affect every aspect of cellular function, thereby implicating high levels of MAPK activity in the growth, survival, and aggressive behavior of breast tumors lacking ERα signaling.
The significance of this hyperactive MAPK signature is greatly enhanced by the close correlation between the gene expression changes seen in our cell line models, and those seen in ERα− tumors ( Fig. 4A). Moreover, the hyperactive MAPK signature can predict ERα negativity in tumor samples with a remarkably high degree of accuracy ( Fig. 4B). This is striking not only because the in vitro cell line profile is closely matching in vivo tumor profiles, but all the more so as this MAPK profile was generated from four stably transfected cell lines all derived from a single parental breast cancer cell line. Even more remarkable is that this single parental cell line is an ERα+/progesterone receptor–positive cell line that was converted to an ERα− phenotype by the overexpression of various members of the MAPK signaling pathway. Interestingly, we have also found that among breast tumors classed as ERα+ (histologically), a subset that have relatively high MAPK activity and resemble the MAPK signature for gene expression are characterized by significantly lower levels of ERα mRNA expression ( Fig. 5), thereby strengthening the link between high MAPK activity and reduced ERα expression. These findings not only support the hypothesis that expression of ERα is a dynamic event dependent on other cellular signaling, but also suggest a gradual alteration from the ERα+, estrogen-dependent phenotype to an ERα− phenotype with MAPK activation being one of the driving factors.
We found that our mRNA expression signature of erbB-2 activation was distinctly different from the mRNA expression signature of EGFR activation. We were able to further subdivide breast cancer patients with profiles indicating high MAPK activity into profiles with the erbB-2 or EGFR signatures. Patient profiles tended to be associated with either erbB-2 or EGFR, rather than with both, indicating that MAPK may be activated by overexpression or constitutive activation of one growth factor receptor or the other. As most of the ERα− tumor profiles in the data sets we examined overexpressed EGFR rather than erbB-2, ERα− tumors tended to associate with the EGFR pathway, although a small subset with high erbB-2 levels associated with erbB-2. Our (ca)erbB-2 cells have constitutive activation of erbB-2 leading to strong, constitutive downstream signaling, similar perhaps to that exhibited by high levels of erbB-2 in tumors. Importantly, a study examining the activation state of overexpressed erbB-2 in breast tumors using a phosphospecific erbB-2 antibody found that tumors exhibiting activation of erbB-2 were most likely to be ERα− ( 39), thus supporting the correlation of our erbB-2 signature with those tumors exhibiting high erbB-2 levels.
We have shown using breast cancer cell line models of up-regulated growth factor signaling that hyperactivation of MAPK can clearly distinguish ERα− breast cancer from ERα+. The idea that a given tumor mRNA profile can indicate whether MAPK is activated and whether that activation occurs via erbB-2 or EGFR could have important implications for designing tumor-specific therapies. Although the pathways that cells may take on their way to malignancy are highly variable ( 40), our study establishes MAPK hyperactivation as an important factor in the determination of ERα status in breast cancer. These cell lines represent excellent models for investigating the mechanisms underlying the generation of the ERα− phenotype in breast cancer as well as the role that hyperactive MAPK plays in the aggressive behavior of ERα− breast cancer.
Grant support: American Cancer Society grant RSG-01-0330-01 (D. El-Ashry), Susan G. Komen Foundation grant BCTR2000565 (D. El-Ashry), NIH-University of Michigan Cancer Center support grant 5 P30 CA46592, and Cancer Biology Training Program (C.J. Creighton).
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.
We would like to acknowledge Drs. Marc Lippman and Michael Johnson for critical review of this manuscript and thoughtful discussions.
- Received December 7, 2005.
- Accepted January 31, 2006.
- ©2006 American Association for Cancer Research.