Development of resistance to targeted therapies transforms the clinically-associated molecular profile subtype of breast tumor xenografts

Abstract

The effectiveness of therapies targeting specific pathways in breast cancer, such as the estrogen receptor or HER2, is limited because many tumors manifest resistance, either de novo or acquired, during the course of treatment. To investigate molecular mechanisms of resistance, we used two xenograft models of estrogen receptor-positive (ER+) breast cancer, one with and one without HER2 over-expression (MCF7/HER2–18 and MCF7 wt, respectively). Mice with established tumors were assigned to the following treatment groups: estrogen supplementation (E2), estrogen deprivation (ED), ED plus tamoxifen (Tam), all with or without the EGFR tyrosine kinase inhibitor gefitinib (G). Another group received ED plus the antiestrogen fulvestrant (MCF7 wt only). Tumors with acquired or de novo resistance to these endocrine therapies were profiled for gene expression and compared with tumors in the E2 control group. One class of genes under-expressed in endocrine-resistant tumors (relative to E2-treated tumors) were estrogen-inducible in vitro and associated with ER+ human breast cancers (Luminal subtype). Another class of genes over-expressed in tumors with acquired resistance in both models represented transcriptional targets of HER2 signaling and were associated with ER-/HER2+ human cancers (ERBB2+ subtype). A third class of genes over-expressed in MCF7/HER2–18 tumors exhibiting de novo resistance to Tam was associated with ER+ human cancers but not with estrogen-regulated genes. Thus, in response to various endocrine therapy regimens, these xenograft breast tumors shut down classical estrogen signaling and activate alternative pathways such as HER2 that contribute to treatment resistance. Over time, the molecular phenotype of breast cancer can change.

Introduction

Great strides have been made in breast cancer treatment using therapies that target specific molecular pathways. In the clinic, two tumor molecular markers are used to help predict therapeutic response in breast cancer: estrogen receptor alpha (ER) and the growth factor receptor HER2 (erbB-2/neu). Approximately 70% of breast cancers express ER (ER+ tumors) and tend to rely upon the classical estrogen signaling pathway for growth (1, 2). Patients with these tumors are usually treated with either anti-estrogens or aromatase inhibitors to block ER signaling. In addition, the 20–25% of tumors with gene amplification of HER2 (HER2+ tumors) rely upon its signaling pathway for growth (3), and HER2+ patients can be successfully treated by agents targeting HER2. Tumors that express neither ER nor HER2 rely on poorly-defined alternative pathways and currently have no effective targeted therapy. Of the breast cancer patients treated with targeted therapies, many do not respond, and of those who do, many acquire resistance over time. While adjuvant therapy studies of the anti-estrogen tamoxifen show a 40–50% reduction in the odds of disease recurrence and a 30–35% reduction in mortality in ER+ breast cancer, still de novo and acquired resistance to ER-targeted therapy remains a major problem (4). In the metastatic setting, 30–50% of ER+ tumors initially respond to tamoxifen, but essentially all of these tumors eventually acquire resistance, leading to disease progression and death (4).

In previous studies, we have used breast cancer xenograft model systems to investigate the mechanisms of resistance to targeted therapies in vivo (5–10). In our models, as in the clinical setting outlined above, resistance to treatment can take the form of de novo resistance, where the tumor shows no initial response, or acquired resistance, where the tumor initially responds but eventually escapes with progressive growth. Analysis of individual molecular marker expression have revealed a role for growth factor signaling via EGFR and HER2 (6, 8) and stress-related pathways associated with p38 and ERK1,2 mitogen activated protein kinases (5, 7) in de novo and acquired resistance to endocrine therapy.

In this present study, we profiled the expression of over 50,000 mRNA transcripts in xenograft tumors that initially were either ER+/HER2+ or ER+/HER2- and that acquired resistance to various targeted therapy regimens. In addition, we profiled ER+/HER2+ tumors that manifested de novo resistance to tamoxifen. The goal of this study was to define and characterize the global gene expression patterns associated with the resistant phenotypes. Several studies using clinical breast tumors have defined distinct global molecular subtypes associated with markers such as ER and HER2 (11–15). An important aspect of our study was to compare the gene signatures derived from our model system with gene signatures derived from clinical breast cancers that reflect specific molecular subtypes, in order to assess how our models might represent breast cancer in the clinical setting.

Materials and Methods

Xenograft studies

MCF7 wt and MCF7/HER2–18 xenograft experiments were carried out as previously reported (6, 8, 10, 16–18) and are described in Supplementary Material. Details of cell culture methods and animal experiments were previously described (6, 8, 10, 16–18). Tumors were harvested for molecular studies either after two-three weeks of treatment (sensitive, “S,” or early-growth, “E,” tumors, <400 mm³ in size) or when they reached the size of 1000 mm³ (resistant, “R,” or late-growth, “L,” tumors). Each tumor analyzed was from a different mouse; tumor tissues were removed from each individual mouse and kept at −190°C for later analyses. Graphs constructed for this paper (Figure 1) represent individual tumors from each of the different treatment groups as indicated.

Figure 1.

In vivo models of endocrine therapy resistance in breast cancer. Ovariectomized athymic nude mice bearing tumors derived from either MCF7 wt or MCF7/HER2–18 cells were randomly assigned (Day 1) to various treatment groups as indicated. E2, continued estrogen supplementation; G, EGFR tyrosine kinase inhibitor gefitinib; ED, estrogen deprivation; Tam, estrogen deprivation plus anti-estrogen tamoxifen; Fulv, estrogen deprivation plus fulvestrant; E2+Tam, estrogen plus tamoxifen. Graphs show growth curves of representative individual tumors in the different treatment groups over time for the MCF7 wt and HER2–18 models (representation of results from refs (8, 10, 16–18)).

Affymetrix microarray analysis

We used Affymetrix U133 Plus 2 Genechip arrays to determine gene expression patterns of tumor xenografts. RNA extraction was carried out in BioRobot EZ1 workstation using EZ1 RNA universal tissue kit according to the manufacturer's instructions (Qiagen, Valencia, CA). cDNA synthesis, cRNA labeling, and array hybridization were performed as previously described (19).. Arrays were processed and normalized using the dChiP software (20). Array data have been deposited in the public Gene Expression Omnibus (GEO) database, accessions GSE8139 and GSE8140. Expression dataset filtering and transformation are detailed in Supplementary Material.

Analysis of variance (ANOVA) using dChip identified genes that differed in mean gene mRNA levels in any one of the xenograft tumor treatment groups. Two-sample t-tests determined significant differences in gene expression between any two specific groups of samples (p-values were two-sided). Selection of t-test p-value cutoffs for defining gene groups was guided by a desire to have a sizeable number of genes for inter-dataset comparisons (where a smaller number of tumor profiles were involved in the t-test, a looser criteria such as p<0.01 versus p<0.001 might be used). Expression values were visualized as heat maps using the Cluster (21) and Java TreeView (22) software. One-sided Fisher’s exact tests and Q1–Q2 analysis (23) were used to determine statistically significant enrichment of subsets of genes. Further details on the microarray analysis are given in Supplementary Material.

Results

In vivo models of endocrine therapy resistance in ER+/HER2- (MCF7 wt) and ER+/HER2+ (MCF7/HER2–18) breast cancer xenografts

To further investigate mechanisms of growth inhibition and hormone treatment resistance, we used tumors from two of our previously defined in vivo xenograft models of ER+ breast cancer: one of MCF7 wild type cells (MCF7 wt) and one of MCF7 cells stably transfected with over-expressing HER2 (MCF7/HER2–18) (6, 8, 10, 16–18). In previous studies, MCF7 wt and MCF7/HER2–18 cells were each established as xenografts in ovariectomized athymic nude mice in the presence of estrogen. When tumors reached a sufficient size (150–200 mm³), mice were randomly assigned to various treatment groups (as detailed in Supplementary Methods and listed in Table 1). MCF7 wt tumors received continued estrogen supplementation (E2), E2 with the EGFR tyrosine kinase inhibitor gefitinib (E2+G), estrogen deprivation (ED), estrogen deprivation plus the antiestrogen tamoxifen (Tam), ED plus Tam with gefitinib (Tam+G), or ED plus ICI 182,780 (fulvestrant, or Fulv). MCF7/HER2–18 tumors received E2, E2+G, ED, ED+G, Tam, Tam+G, or E2+Tam.

Table 1.

Xenograft tumor treatment groups.

Group	# of samples	Description
MCF7 wild-type model
E2	3	continued E2 supplementation
E2+G	2	growing in the presence of E2 plus EGFR TKI gefitinib
ED	4	acquired resistance to estrogen deprivation
Fulv	4	acquired resistance to ED plus fulvestrant
Tam	4	acquired resistance to ED plus tamoxifen
MCF7/HER2–18 model
E2 (E)	4	continued E2 supplementation (“early”* tumor)
E2 (L)	3	continued E2 supplementation (“late”** tumor)
E2+G (E)	4	growing in the presence of E2 plus gefinitib (“early” tumor)
E2+G (L)	5	growing in the presence of E2 plus gefinitib (“late”tumor)
ED (S)	4	sensitive to estrogen deprivation
ED (R)	4	acquired resistance to estrogen deprivation
ED+G (S)	1	sensitive to ED plus gefitinib
ED+G (R)	5	acquired resistance to ED plus gefitinib
Tam (E)	5	growing in the presence of ED plus tamoxifen (“early” tumor, de novo resistant)
Tam (L)	4	growing in the presence of ED plus tamoxifen (“late” tumor, de novo resistant)
Tam+G (S)	4	sensitive to ED plus tamoxifen plus gefitinib
Tam+G (R)	4	acquired resistance to ED plus tamoxifen plus gefitinib
E2+Tam (L)	4	growing in the presence of E2 plus tamoxifen

^*harvested at early time points (tumor size 200–400 mm³)

^**harvested at later time points (tumor size ~1000 mm³)

Figure 1 shows a representative individual tumor growth curve from each of the treatment groups in each model (MCF7 wt and HER2–18) followed over time. These data and earlier experiments leading up to them have been presented and described in detail in prior reports (6, 8, 10, 16–18). While MCF7 wt tumors in E2 supplemented mice grew at a steady exponential rate, wt tumors in treated mice (the ED, Tam, Tam+G, Fulv groups) initially stopped growing (“sensitive” phase of treatment). However, after several months, the E2-deprived tumors begin to grow rapidly at a rate comparable to that of the E2 control group (“resistant” phase of treatment). The time to treatment resistance varied among the treatment groups, with the Tam+G and Fulv groups taking about 2–3 months longer to acquire resistance than the ED and Tam groups; resistance, however, eventually developed in all treatment groups. In the MCF7/HER2–18 model, the behavior of E2-deprived tumors was similar to that of the MCF7 wt model. However, unlike the MCF7 wt tumors, HER2–18 tumors were growth-stimulated by ED+Tam as the mechanism of de novo resistance. HER2–18 tumors were inhibited by Tam combined with Gefitinib, which restores Tam’s antitumor effect by blocking the HER2 pathway (6).

Gene expression signatures of endocrine therapy resistance in MCF7 wt and MCF7/HER2–18 xenografts

MCF7 wt and MCF7/HER2–18 tumors which had manifested resistance to various endocrine therapies (tumors in the ED, Tam, and Tam+G groups, as well as tumors in the MCF7 wt Fulv and MCF7/HER2–18 ED+G groups)--along with growing tumors from the E2 treatment groups (E2 alone as well as E2+G and MCF7/HER2–18 E2+Tam)--were profiled for 54675 mRNA transcripts (list of profiled tumor groups in Table 1). In addition to “resistant” tumors, we profiled MCF7/HER2–18 “sensitive” tumors in which growth was inhibited by ED, ED+G, and Tam+G groups. In addition, MCF7/HER2–18 tumors treated and stimulated by E2, E2+G, and Tam were profiled after three weeks of treatment (early tumors) and again after 8 weeks (late tumors), at tumor volumes that closely corresponded to the “sensitive” and “resistant” tumors of the other treatment groups. One profile of a HER2–18 ED+G sensitive tumor was taken, its profile appearing similar to the HER2–18 ED sensitive group (Figure 2A). For the HER2–18 E2+G group, gefitinib led to significant yet modest inhibition of growth (Figure 1), and thus we still referred to these tumors as early (“E”) versus late (“L”) (rather than sensitive versus resistant).

Figure 2.

Gene expression signatures of endocrine therapy resistance in xenograft mouse models. (A) Hierarchical clustering of patterns for genes that showed significant expression (p<0.001, ANOVA) in any one MCF7/HER2–18 treatment group. Expression patterns represented using color map (yellow: high expression; blue: low expression), each row representing a gene, each column representing a tumor. S, tumors sensitive to estrogen deprivation; R, tumors resistant to estrogen deprivation; E, L, tumors collected at early and late time periods, respectively (for treatment groups not showing initial sensitivity, see Figure 1). (B) Using clustering pattern of (A), template patterns representing the three major gene clusters were defined, and genes that best fitted each pattern were obtained (groups 1, 2, and 3). (C) Hierarchical clustering of patterns for genes that showed significant expression (p<0.01, ANOVA) in any one MCF7 wt treatment group (resistant or late tumors profiled only). (D) Using clustering pattern of (C), two groups of genes (groups 4 and 5) that best fitted pre-defined template patterns were obtained.

For the MCF7/HER2–18 profile data, we performed hierarchical clustering (24) of all 6883 transcripts that differed significantly in mean mRNA levels in any treatment group compared to the other groups (p<0.001, ANOVA). The clustered expression patterns were visualized as a color map (Figure 2A). From the overall clustering pattern, we observed three sizeable gene groups. Going back to the original dataset, we used supervised approaches to define the sets of genes with patterns that were representative of each of the three clustering patterns of Figure 2A (Figure 2B and Table 2): (1) a set of 525 uniquely identified genes, designated “group 1” genes, that were more highly expressed in the E2 treatment groups (both with and without gefitinib) than in both the ED/ED+G resistant groups (p<0.001, t-test) and the Tam/Tam+G late/resistant groups (p<0.001); (2) a set of 474 genes, the “group 2” genes, that were higher in the Tam/Tam+G late/resistant groups (p<0.01), the Tam/Tam+G early/sensitive groups (p<0.01), and the ED sensitive groups (p<0.01), all compared to the E2/E2+G groups, but were not higher in the ED/ED+G resistant groups compared to the E2/E2+G groups (p>0.05); and (3) a set of 720 genes, the “group 3” genes, that were higher in both the ED/ED+G resistant groups (p<0.001) and the Tam/Tam+G late/resistant groups (p<0.001), as compared to the E2/E2+G groups.

Table 2.

Gene sets associated with endocrine therapy resistance in MCF7 xenograft mouse models.

Gene set	MCF-7 Model	Description	# of genes	Selected example genes
Group 1	HER2–18	Higher in E2/E2+G compared to ED/ED+G and Tam/Tam+G resistant groups. Overlaps with group 4.	525	ESR1, FGFR2, GATA3, GPR30, PTEN, CCNB2, PARP1, IGF1R, MYC, IRS1, PGR
Group 2	HER2–18	Higher in Tam/Tam+G sensitive and resistant groups and ED/ED+G sensitive groups compared to E2/E2+G; no difference between E2/E2+G and ED/ED+G resistant groups.	474	FOXJ1, FOXN4, MDM2, CDKN2D
Group 3	HER2–18	Higher in both ED/ED+G and Tam/Tam+G resistant groups compared to E2/E2+G groups. Overlaps with group 5.	720	MUC4, PAK1, MYCN, S100A6, FZD8, GPR110, CEACAM5, GPR51, MUC1, TMPRSS2
Group 4	wt	Higher in E2/E2+G compared to ED/Fulv and Tam/Tam+G resistant groups. Overlaps with group 1.	310	NPY5R, MYB, FOXK2, TSPAN6, CCNA1, IGF1R, MYC, IRS1, PGR
Group 5	wt	Higher in ED/Fulv and Tam/Tam+G resistant groups compared to E2/E2+G. Overlaps with group 3.	483	ERBB2, CTNNB1, KRAS, TCF4, THBS1, FZD4, CEACAM5, GPR51, MUC1, TMPRSS2

We analyzed the MCF7 wt profile data in a similar manner to the MCF7/HER2–18 data. Hierarchical clustering of 1363 transcripts differing in mean mRNA levels in any treatment group (p<0.01, ANOVA) uncovered two major gene groups of interest (Figure 2C) (as there were fewer experimental groups in the MCF7 wt dataset compared to MCF7/HER2–18, using a higher ANOVA p-value cutoff for MCF7 wt seemed appropriate to account for the lower power to detect differences in any one group). From the original dataset, we obtained the following genes to represent each clustering pattern (Figure 2D and Table 2): (1) a set of 310 unique named genes, which we designated the “group 4” genes, that were higher in the E2 treatment groups (with and without gefitinib (G)) compared to the ED and Fulv resistant groups combined (p<0.01, t-test) and compared to the Tam/Tam+G resistant groups (p<0.01); and (2) a set of 483 genes, the “group 5” genes, that were higher in both the ED/Fulv resistant groups (p<0.01) and the Tam/Tam+G resistant groups (p<0.01) compared to the E2/E2+G groups. There are very likely other gene expression patterns of interest existing in our profile data that are more subtle and may include fewer genes. In this study, however, we focused on analysis of the five major gene groups identified here, leaving open the opportunity to further mine the data in future studies.

For the set of genes in any one of the five groups 1–5 (Figures 2B and 2D and Table 2, complete list provided in Supplementary Data File 1), we comp ared the expression patterns of the MCF7/HER2–18 dataset side-by-side with those of the MCF7 wt dataset. We observed high overlap between the MCF7/HER2–18 group 1 and the MCF7 wt group 4 (Figure 3), both of these gene groups being higher in the respective E2 treatment groups compared to the estrogen-deprived groups. Between the groups 1 and 4 genes, 85 were shared; when we relaxed the criteria for inclusion of genes from one group within the other group (looking for genes included in one of the two groups and that would have been included in the other group had cutoffs of p<0.05 been used to define it), 215 genes could be considered shared. Similarly, the MCF7/HER2–18 group 3 overlapped highly with the MCF7 wt group 5 (Figure 3), genes in either of these groups being more highly expressed in their respective estrogen-deprived compared to E2 treatment groups; between groups 3 and 5, 117 genes were shared (303 using relaxed criteria). For the MCF7/HER2–18 group 2 genes, which were over-expressed in the Tam-resistant and ED-sensitive tumor groups but not in the ED-resistant groups, no corresponding patterns were found in the MCF7 wt data; in particular, the group 2 genes were not over-expressed in the MCF7 wt Tam resistant groups as might have been expected (Figure 3).

Figure 3.

Side-by-side comparison of the gene signatures of endocrine therapy resistance from MCF7/HER2–18 and MCF7 wt xenograft models. In the MCF7/HER2–18 and MCF7 wt datasets, expression patterns for genes belonging to any one of the five groups defined in Figure 2C and 2D are shown as a color map. Also shown are the corresponding expression patterns in a profile dataset (25) of three different breast cancer cell lines stimulated with E2 over a time course from 0–24 hours (gray: data not represented). The order of the genes is the same across all three datasets represented, allowing one to observe where the various MCF7 xenograft gene sets from the two models overlap, and what proportion of these genes appear estrogen-regulated.

ESR1 (ER) and PGR (PR) mRNA were both low in most resistant tumors compared to E2 control (Figure 3). We have previously reported that ER expression at the protein level is lost in the MCF7/HER2–18 resistant tumors, except ED+Tam alone (8). Furthermore, HER2 mRNA appeared elevated in both MCF7 wt and MCF7/HER-18 resistant tumors compared to E2 controls (Figure 3), which is consistent with our prior observations of increased HER2 protein expression in patients with Tam resistance (7) and in our experimental model (17). These results indicate that the molecular phenotype of tumors can change with time as they develop treatment resistance, changing from an ER+ to an ER- phenotype. We further explored this idea below through analysis of the whole sets of genes associated with resistance.

Genes under-expressed in endocrine-resistant tumors, both MCF7 wt and MCF7/HER2–18, are estrogen-regulated and associated with ER+/PR+ human breast cancers

The MCF7/HER2–18 group 1 genes and the MCF7 wt group 4 genes were both more highly expressed in the respective E2 compared to estrogen-deprived treatment groups. To determine which of these genes might be targets of the classical estrogen signaling pathway, we examined the corresponding expression patterns of the group 1 and 4 genes in an expression profile dataset from a previous study (25) of breast cancer cell cultures stimulated by estradiol (E2) over a time course of 1 to 24 hours (Figure 3). The group 1 and group 4 genes had high enrichment of genes induced by E2 in vitro (i.e. genes in estrogen clusters B, C, and D as defined in (25)). Of the 525 group 1 genes (Figure 2A), 116, over 20%, were estrogen-induced (significance of overlap p<1E-35, one-sided Fisher’s exact), 63 of these within 4 hours of E2 treatment, suggesting that they represent primary estrogen-regulated genes. Of the 310 group 4 genes (Figure 2D and Table 2), 74 were estrogen-induced (Fisher’s exact p<1E-24), 51 within 4 hours.

We then examined the expression patterns of MCF7/HER2–18 group 1 genes and the MCF7 wt group 4 genes in human breast tumors, in order to ascertain the possible clinical context for our experimentally-derived genes. We obtained three independent gene expression profile datasets of human breast tumors: one from van de Vijver et al. (26) of 295 tumors (226 of them ER+), one from Wang et al. (11) of 286 tumors (209 ER+), and one from Miller et al. (13) of 247 tumors (213 ER+). We sought to determine whether any enrichment existed between our xenograft gene sets and relevant expression patterns in the human tumor datasets. Enrichment was assessed using the rigorous Q1–Q2 method (described in detail in refs (23, 27) and in Supplementary Material), which tested against randomly-generated xenograft gene sets or human tumor profile labels.

By Q1–Q2 (results shown in Figure 4A), the group 1 and group 4 gene sets (highly expressed in the E2 control groups for their respectively models, Table 2) were each significantly enriched in ER+tumors having high PR mRNA levels (these genes being positively correlated as a group with PR mRNA within the ER+ tumors). These patterns were observed in all three tumor datasets examined (group 1: van de Vijver p<1E-07, Wang p<1E-08, Miller p=0.0005; group 4: van de Vijver p<1E-11, Wang p<1E-11, Miller p<1E-6). Furthermore, group1 and group 4 genes were significantly enriched in ER+ over ER- human tumors in each of the clinical datasets (group 1: p<0.00005; group 4: p<1E-06). We evaluated the set of genes shared between groups 1 and 4 (Figure 3), as well as the genes specific to group 1 and not to 4 and genes specific to group 4 and not 1; only the genes that were shared between groups 1 and 4 were consistently enriched in ER+/PR+ tumors.

Figure 4.

Enrichment analysis of xenograft tumor genes within human breast tumor expression profiles. (A) Q1–Q2 enrichment patterns of the five xenograft gene sets (Figures 2 and 3) across three different profile data sets of human breast tumors (11, 13, 26). Four different human breast tumor gene rankings were evaluated: (1) genes with higher average expression in ER+ compared with ER- tumors (“ER+ (over ER-)”), (2) genes with high correlation (i.e. similarity) with PR mRNA within the subset of ER+ tumors (“ER+_PR”), (3) genes correlated with HER2 mRNA expression patterns within ER+ tumors (“ER+_HER2”), (4) genes correlated with HER2 mRNA within ER- tumors (“ER-_HER2”). (B) RNA expression patterns across MCF7 xenograft and clinical breast tumor datasets for the following sets of genes: (1) ER, PR, and HER2 (top panels); (2) genes in any one of the MCF7 xenograft groups 1–5 (Figure 2) that were also differentially expressed in ER+ versus ER- clinical tumors (middle panels); (3) genes in MCF7 xenograft groups 1–5 that were also correlated or anti-correlated with HER2 mRNA in ER-negative clinical tumors (bottom panels). Alongside the ER-status-associated genes are the corresponding patterns for E2-stimulated breast cell cultures. Alongside the HER2/ER-negative-associated genes are the corresponding patterns for MCF7 cells with MAPK-activated through various genes (28). The order of the genes is the same across all datasets represented. Patterns of enrichment (where xenograft and human tumor expression patterns share high overlap contributing to enrichment associations) are indicated. (C) Heat map of the Pearson’s correlations (genome-wide) between each xengraft tumor profile and the average expression for each of the four major breast tumor molecular profile subtypes (basal, ERBB2+, luminal A, luminal B) as defined by Hoadley et al. (29).

The ER+-associated human tumor enrichment patterns described above were evident when viewing the gene expression patterns involved using color maps (Figure 4B). Of the 215 genes that were shared between groups 1 and 4, a high overlap of 51 (chance expected 13, p<1xE-17, one-sided Fisher’s exact) were also higher in ER+ over ER-human tumors; in contrast, 22 genes were lower in ER+ human tumors, a number closer to the chance expected. Nearly all of the 51 group 1/group 4/ER+ human tumor genes in the profile dataset of E2-stimulated breast cancer cell cultures (25) showed induction by E2 in vitro (Figure 4B). These genes also tended to be correlated with higher PR mRNA levels within the ER+ group of tumors (Figure 4A, though this is not evident in the Figure 4B representation, as PR was uniformly higher in ER+ over ER- tumors).

Genes over-expressed in MCF7 wt and MCF7/HER2–18 endocrine-resistant tumors are associated with the HER2 pathway and with ER-/HER2+ human breast cancers

The MCF7/HER2–18 group 3 genes and the MCF7 wt group 5 genes (Table 2) were both more highly expressed in the respective endocrine-resistant compared to the E2 control groups. In the HER2–18 dataset, endocrine-sensitive tumors had levels of expression for the group 3 genes that were between those of the E2 controls and the fully resistant tumors (Figure 3), indicating that these genes were correlated with acquired resistance. When considering the expression profile dataset of E2-stimulated breast cancer cell cultures, many of the group 3 and 5 genes appeared to be targets of estrogen-mediated repression (Figures 3 and 4B and Supplementary Material).

We examined the expression patterns of the MCF7/HER2–18 group 3 genes and the MCF7 wt group 5 genes in human breast tumors (results shown in Figure 4A). For the subset of ER- human breast tumors in each of three profile datasets, the group 3 and group 5 genes were enriched within the genes positively correlated with HER2 (ERBB2) mRNA levels (group 3: van de Vijver p=0.006, Wang p<1E-05, Miller p=0.03; group 5: van de Vijver p<5E-6, Wang p<1E-7, Miller p<0.001, Q1–Q2). The subset of ER+ tumors, however, had no similar enrichment patterns with respect to HER2 mRNA (Figures 4A and 4B), indicating that many of the xenograft group 3 and group 5 genes are over-expressed in ER-/HER2+ compared to ER-/HER2- tumors but are not over-expressed in ER+/HER2+ compared to ER+/HER2- tumors. The ER-/HER2+-associated human tumor enrichment patterns uncovered by Q1–Q2 analysis (Figure 4A) were evident when viewing the gene expression patterns involved using color maps (Figure 4B and Supplementary Material).

The ER-/HER2+ tumors from the van de Vijver, Wang, and Miller datasets were an approximate surrogate for the “ERBB2+” tumor class defined by Sorlie et al. (14, 15), and the group 3 and 5 genes also overlapped with genes defining this class (Supplementary Material). In another recent study (28), gene expression profiles were taken of MCF-7 breast cancer cells with stably over-expressed EGFR or constitutively activated HER2 (erbB-2), Raf, or MEK. The MCF7/HER2–18 group 3 and MCF7 wt group 5 genes each overlapped significantly with the genes high in the MCF-7/HER2 cell lines (Figure 4B and Supplementary Material). In summary, we observed that breast tumors converted from one molecular phenotype (ER+ or luminal) to another (ER- /HER2+ or ERBB2+) after treatment.

Genes over-expressed in MCF7/HER2–18 tumors with de novo resistance to Tam are associated with ER+/PR+ human breast cancers but are not estrogen-regulated

In the MCF7/HER2–18 dataset, the group 2 genes were more highly expressed in both the ED+Tam groups (early and late) and ED sensitive group compared to the E2-treated groups, but were not more highly expressed in the ED resistant groups (Figures 2D and 3). These group 2 patterns were not observed in the MCF7 wt dataset (as they were not over-expressed in the wt Tam resistant tumors). Interestingly, the expression of the group 2 genes was attenuated in the Tam+G-treated resistant MCF7/HER2–18 tumors relative to the Tam tumors (Figure 3). The group 2 genes had no significant number of E2-regulated transcripts (Figure 3). Furthermore, the group 2 genes were enriched in ER+/PR+ compared to ER+/PR- human breast tumors, as well as in ER+ compared to ER- tumors (Figure 4A). Of the 474 group 2 genes, 52 were higher in ER+ over ER-human tumors (Fisher’s exact p=0.00001), yet virtually none of these genes were estrogen-regulated (Figure 4B).

MCF7 wt and MCF7/HER2–18 endocrine-resistant xenograft tumor profiles are globally associated with the ERBB2+ human breast tumor molecular profile subtype

Previous studies in gene expression profiling of human breast cancers have identified at least four major tumor subtypes (luminal A, luminal B, basal, ERBB2+) and a normal breast tissue group (15). While our above analyses associated specific sets of genes over-expressed in endocrine-resistant xenograft tumors with the HER2 pathway (Figures 4A and 4B), we went on to do a global comparison between each of our xenograft tumor profiles and the molecular profile subtypes as defined by a previously published human breast tumor dataset by Hoadley et al. (29). For each gene common to our array platform and the Hoadley platform, we computed the mean centroid of the four major Hoadley tumor subtypes and centered each group average on the centroid. We then took the Pearson’s correlation (using all 7316 genes common to both datasets) between the Hoadley centered-averages and the expression values of each xenograft tumor.

The profile-to-subtype correlation analyses (graphically represented in Figure 4C) showed both an overall trend for the E2-treated and sensitive/early xenograft tumors (both MCF7/HER2–18 and MCF7 wt models) to associate with the luminal subtype (either luminal A or luminal B or both), and an overall trend for the resistant/late tumors to associate with the ERBB2+ group. At the same time, a few individual tumors were found to show associations not consistent with those of their experimental group, and tumors could associate with more than one subtype (e.g. luminal A and luminal B). By our using all of the profiled genes, the Hoadley results were not influenced by our previous definitions of the five gene groups (Table 2). Therefore, by alternative analyses approaches, we could observe that the xenograft tumors converted from one molecular phenotype to another after acquiring treatment resistance.

Discussion

In our model system, we examined the behavior of two different breast cancer cell lines in response to therapies targeting the estrogen and HER signaling pathways: MCF7 wild-type (wt), which is thought to represent the ER+/HER2- clinical subtype, and MCF7/HER2–18, which is thought to represent the ER+/HER2+ subtype. When these cell lines are grown as tumor xenografts in the presence of estrogen, they rely upon the classical estrogen/ER signaling pathway for growth. When estrogen supplementation is removed, the tumors from both cell lines shut down classical estrogen signaling, as indicated by the loss of expression of the group 1 and group 4 gene sets, over 20% of which genes are up-regulated by estrogen. It has been hypothesized that breast cancers with acquired resistance to ED or Tam might somehow reactivate classical estrogen signaling (e.g. through increasing estrogen agonist effects of anti-estrogens, or by ligand-independent activation of ER via growth factor signaling, or by an activating mutation in ER) (6, 30, 31). However, reactivation of estrogen signaling was not observed in our system as a mechanism of resistance.

To acquire endocrine resistance, our MCF7 wt and MCF7/HER2–18 tumors activate growth factor signaling pathways including HER2 and EGFR (6–8). By correlating the gene expression patterns from our xenograft models with patterns from clinical breast tumors, we determined that our xenograft tumors with acquired resistance represented at the transcriptome level the ER-/HER2+ clinical breast cancer subtype (also known as the “ERBB2+” molecular subtype (14)). The group 3 and group 5 genes (over-expressed in the endocrine-resistant tumors) each shared extensive overlap with genes high in ER-/HER2+ clinical tumors, these genes also being highly enriched for transcriptional targets of HER2 signaling in vitro. In addition, we recently showed (32) that more specific HER-2 inhibition using a variety of compounds had greater efficacy in inhibiting growth of our HER2-overexpressing xenografts.

In this study, we observed in our model system that acquired resistance to various ER-targeted therapies transformed the clinically-associated molecular profile subtype of breast cancer from that of an estrogen-dependent, ER+ tumor to that of a growth-factor-dependent, ER-/HER2+ tumor. Resistant MCF7 wt tumors had been previously found to retain ER protein expression (16), though here these tumors appeared ER- at the transcriptome level. This switch from ER+ to ER-/HER2+ perhaps is not surprising in the MCF7/HER2–18 tumors, which over-express HER2 to begin with. However, our MCF7 wt tumors did not over-express HER2 and targets of this pathway before being deprived of estrogen (Figure 3, also (7, 17)). This up-regulation of HER2 as well as EGFR has been observed in other preclinical models of endocrine therapy resistance (33–36). Furthermore, recent findings show that HER2 gene amplification can be acquired as breast cancer progresses in patients receiving endocrine therapy (7, 37, 38).

One popular idea in molecular oncology is to identify molecular markers predicting tumor response to targeted therapies. Breast cancer treatment is a prime example of this idea being put into clinical practice, with patients whose tumors express ER receiving endocrine therapy and patients whose tumors express HER2 receiving anti-HER2 therapy. At the same time, our study here suggests that using gene expression of pre-treatment tumors to predict response might be limited to predicting initial response, and would not necessarily be predictive of possible mechanisms of acquired resistance. In our model, we could not have predicted HER2 as one mechanism of acquired resistance, solely on the basis of the pre-treatment tumor data. Our study lends support to a cancer treatment paradigm that targets not only the tumor in its pre-treatment state, but also the tumor as it might manifest itself during tumor progression and acquired therapeutic resistance.

Our current clinical practice of treating breast cancer disease recurrence based on tumor characteristics from the original diagnosis may, therefore, not take into account the dynamic nature of cancer progression in response to specific treatments. Repeating tumor biopsies at the time of cancer recurrence may be critical to correctly identify molecular targets for more tailored therapeutic options with the hope that this strategy can improve outcome in patients with breast cancer.

Abbreviations

ER

estrogen receptor alpha

HER2

HER2/neu

p-HER2

phosphorylated HER2

MAPK 42/44

mitogen-activated protein kinase

EGFR

EGF receptor

TKI

tyrosine kinase inhibitor

PR

progesterone receptor