• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of jcoHomeThis ArticleSearchSubmitASCO JCO Homepage
J Clin Oncol. Sep 1, 2008; 26(25): 4078–4085.
PMCID: PMC2654368

Insulin-Like Growth Factor-I Activates Gene Transcription Programs Strongly Associated With Poor Breast Cancer Prognosis

Abstract

Purpose

Substantial evidence implicates insulin-like growth factor-I (IGF-I) signaling in the development and progression of breast cancer. To more clearly elucidate the role of IGF in human breast cancer, we identified and then examined gene expression patterns of IGF-I–treated breast cancer cells.

Methods

MCF-7 cells were stimulated with IGF-I for 3 or 24 hours and were profiled for greater than 22,000 RNA transcripts. We defined an IGF-I signature pattern of more than 800 genes that were up- or downregulated at both time points. The gene signature was examined in clinical breast tumors and in experimental models that represented other oncogenic pathways. The signature was correlated with clinical and pathologic variables and with patient outcome.

Results

IGF-I caused temporal changes in gene expression that were strongly associated with cell proliferation, metabolism, and DNA repair. Genes with early and sustained regulation by IGF-I were highly enriched for transcriptional targets of the estrogen receptor (ER), Ras/extracellular signal-related kinase 1/2, and phosphatidylinositol 3-kinase/Akt/mammalian target of rapamycin pathways. In three large, independent data sets of profiled human breast tumors, the IGF-I signature was manifested in the majority of ER-negative breast tumors and in a subset (approximately 25%) of ER-positive breast tumors. Patients who had tumors that manifested the IGF-I signature (including patients who did not receive adjuvant therapy) had a shorter time to a poor outcome event. The IGF gene signature was highly correlated with numerous poor prognostic factors and was one of the strongest indicators of disease outcome.

Conclusion

Transcriptional targets of IGF-I represent pathways of increased aggressiveness and possibly hormone independence in clinical breast cancers.

INTRODUCTION

Substantial evidence implicates insulin-like growth factor (IGF) signaling in the development and progression of many cancers, including breast cancer.1 High serum IGF-I levels predict an increased risk of breast cancer,2 and antiestrogens that are effective in the treatment and prevention of breast cancer reduce serum IGF-I levels.3 IGF-I receptor (IGF-IR) is hyperactive and overexpressed in many breast cancers.4 Preclinical data have validated IGF-IR as a therapeutic target, and several IGF-IR inhibitors have recently entered clinical trials.5

After IGF-I binds to its receptor, it activates a cascade of downstream signaling events. Although great progress has been made regarding the mechanism of IGF-IR signaling at the immediate level of protein phosphorylation cascades,1 relatively little is known regarding transcriptional targets of the IGF-I pathway. Analysis of these targets may shed light on the pathway's role in breast cancer and may provide biomarkers of IGF-I action. In this study, we used gene expression profiling to identify IGF-I–regulated genes in breast cancer cells. We identified a set of more than 800 genes in which expression was increased or decreased by IGF-I (ie, an IGF-I gene signature) that is present in a significant proportion of human breast cancers and that indicates a poor prognosis.

METHODS

Cell Line and Culture Conditions

MCF-7L ER-positive breast cancer cells were obtained originally from American Type Culture Collection and have been extensively described previously. Cells were routinely maintained in Dulbecco's Modified Eagle's Medium + 5% fetal bovine serum. Cells were plated in 10-cm dishes and then switched on the next day to serum-free medium. After an overnight incubation, cells were stimulated in triplicate with or without IGF-I (8 nmol/L or 50 ng/mL; GroPep, Adelaide, Australia) for 3 or 24 hours (serum-free medium 3-hours group was in duplicate) as well as with or without 1 nmol/L estradiol. Cells were then lysed, and total RNA was isolated by using the Qiagen midi-prep kit (Valencia, CA) according to the manufacturer's instructions.

Affymetrix Microarray Analysis

cDNA synthesis, cRNA labeling, and array hybridization were performed as previously described25 by using Affymetrix U133A 2.0 gene chips (Santa Clara, CA). Array data has been deposited in the public Gene Expression Omnibus database, accession GSE7561. Arrays were processed and normalized by using the dChiP software (http://biosun1.harvard.edu/complab/dchip/).26 Expression values were visualized as color maps by using the Cluster and Java TreeView software (http://rana.lbl.gov/EisenSoftware.htm).27,28 Data were analyzed through the use of Ingenuity Pathways Analysis (Ingenuity Systems, Redwood City, CA).

RESULTS

Global Gene Transcription Patterns Initiated by IGF-I Treatment of MCF-7 Cells

ER-positive and IGF-responsive MCF-7 human breast cancer cells were starved overnight in serum-free medium (SFM) and were then stimulated with or without IGF-I for 3 or 24 hours. RNA was isolated, and gene expression array profiles were analyzed with GS (gene expression data deposited at the Gene Expression Omnibus database, GSE7561). After filtering the genes (Appendix, online only), 13,731 RNA transcripts (8,496 unique, named genes) were represented. A comparison of IGF-I with SFM treatment revealed 2,154 transcripts (1,745 unique genes) that showed significant changes (P < .01, fold change > 1.5) at 3 or 24 hours or both. By hierarchical clustering, these genes separated into two approximately equal-sized groups of up- and downregulated genes (Appendix Fig A1, online only).

Genes that were increased or decreased by IGF-I included some expected targets but also many novel targets. For example, we have reported previously that IGF-I can repress progesterone receptor (PR) mRNA expression via phosphatidylinositol 3-kinase (PI3K)/Akt inhibition of the PR promoter6 and that PR levels in our data set were strongly downregulated (P < .001) at 3 hours. Several other genes in the insulin/IGF pathway were regulated, including INSIG1, IGF2BP2, IRS1, and IRS2.

We found significant concordance between genes that we identified as being regulated by IGF-I stimulation of MCF-7 cells and genes from a recently published study of MCF-7 cells that were stably overexpressing IGF-I7. Of the 24 genes reported as upregulated by IGF-I7, 15 likewise were increased (P < .05) by IGF-I at 24 hours in our data set (P < .000001, one-sided Fisher exact); of five genes previously downregulated by IGF-I, three likewise were decreased in our data set (Appendix).

To examine the biologic function of IGF-I–regulated genes, we performed ingenuity pathway analysis (IPA). IPA indicated significant representation within IGF-regulated genes related to the cell cycle and to cell growth and proliferation (Appendix Figs A2A and A2B, online only), which was consistent with the status of IGF-I as one of the most potent mitogens for human breast cancer cells. Indeed, IGF-I altered mRNA levels of numerous cyclins (CCNA1, A2, B1, D3, E1, E2, F, G1, and G2), cell-division cycle proteins (CDC14A, 14B, 2, A3, A4, A6, A8, 20, 25A, 25B, 25C; 2L6, 42EP1, and 45L), and cyclin-dependent kinases (CDK2, AP2, N1A, N1B, N2C, and N3). IPA analysis revealed temporal changes in IGF-I–regulated genes that affected different cellular pathways and functions. For example, 3 hours of IGF-I treatment showed regulation of G1/S checkpoint genes (Appendix Fig A3A, online only), whereas there was an effect on G2/M DNA damage checkpoint genes after 24 hours (Appendix Fig A3B). An example of IGF-I regulation of a gene expression network associated with G2/M DNA damage checkpoint regulation is shown (Appendix Fig A4, online only). IPA canonical pathway analysis revealed regulation of insulin and IGF-I signaling, as might be predicted; however, numerous other growth factor signaling pathways were associated, including epidermal growth factor, fibroblast growth factor, and vascular endothelial growth factor.

To more clearly articulate temporal changes in gene expression, we performed a supervised clustering analysis that compared SFM and IGF-I at 3 or 24 hours or both. Each transcript was assigned to an expression pattern of time-dependent induction or repression (Fig 1A). The global transcription patterns of IGF-I signaling showed temporal complexity; different genes showed induction or repression at different time points. Selected enriched Gene Ontology (GO) annotation terms for each gene cluster are indicated in Figure 1A.

Fig 1.
Global gene transcription patterns initiated by insulin-like growth factor (IGF) treatment in vitro. Affymetrix profiles were taken of MCF-7 cells in serum-free medium with or without the presence of IGF-I (8 nmol/L) at 3 and 24 hours. (A) Supervised ...

Genes With Early and Sustained Regulation by IGF Enriched for Transcriptional Targets of the Ras/Extracellular Signal-Related Kinase 1/2 and PI3K/Akt/Mammalian Target of Rapamycin Pathways

IGF-I is known to signal via rapid phosphorylation and activation of the Grb2/sos/Ras/ extracellular signal-related kinase 1/2 (ERK1/2) and the PI3K/Akt/mammalian target of rapamycin (mTOR) signaling cascades,1 which are generic to many growth factor signaling pathways. Surprisingly, although IGF-I activates these protein cascades at the level of phosphorylation, alterations in the mRNA levels of many of the same proteins were noted as well. Thus, IGF-I altered levels of numerous PI3K family members (PIK3C3, CA, R1, R3, and R4), mitogen activated protein kinase (MAPK) family members (MAP2K2, 3, 5, 6, 7, 8; MAPK9; and MAPKAPK5), and several dual-specificity phosphatases of MAPKs (DUSP2, 4, 5, 6, 8, 10, 12, and 13).

We continued to examine pathways activated by IGF-I by using public array data sets that defined downstream transcriptional targets of specific oncogenic pathways.8-10 We focused on a set of 1,084 RNA transcripts (903 unique, named genes) that were up- or downregulated by IGF at 3 and 24 hours (P < .01; fold change > 1.5 at either time point; P < .05 at the other time point). We reasoned that these genes represented early and sustained targets of IGF signaling (whereas genes regulated at 24 hours but not at 3 hours represented more secondary effects, such as cell proliferation).

When we viewed the expression patterns of our early and sustained targets of IGFs in one profile data set8 of breast cells activated with various oncogenes, we found widespread similarities between our IGF-regulated genes and a Ras signature pattern (Fig 1B). Notably, the HRAS gene itself showed upregulation by IGF in the array data (Fig 1B).

We compared our early and sustained targets of IGF (Fig 1B) with a public database (http://www.broad.mit.edu/cmap) of gene expression profiles of cultured cells (mostly MCF-7 cells) that were treated with 164 different small molecule inhibitors, called the connectivity map (CMap).9 CMap analysis indicated that our IGF gene signature showed an inverse correlation with the expression patterns induced by the PI3K inhibitors LY-294002 and wortmannin, by the mTOR inhibitor rapamyacin, or by the histone deacetylase (HDAC) inhibitor trichostatin A (Fig 1B). A significant number of 179 of the 422 genes upregulated by IGF were also downregulated by inhibition of PI3K; genes downregulated by IGF likewise tended to be upregulated by inhibition of PI3K (Appendix).

Similarities Among the Transcriptional Programs of IGF, Estrogen, and ERBB Signaling Pathways

Numerous studies have detailed extensive cross-talk between the IGF signaling pathway and two major growth regulators in breast cancer, namely ER and epidermal growth factor receptor/ERBB.11 Therefore, we sought to examine the level of similarity among these signaling cascades. As IGF-IR, ER, and ERBBs all stimulate cell proliferation, we first removed from additional consideration 108 transcripts (89 genes) from the 1,084 transcripts (903 genes) defined as early and sustained targets of IGF (Fig 1B), that were either annotated as cell cycle by GO or were found experimentally to be correlated with cell cycle progression12 so that any associations made between the remaining genes would not be caused by generic cell proliferation effects. We then compared our IGF gene signature of 976 transcripts (814 genes) with gene expression signatures from MCF-7 cells that were stimulated with estrogen or that had activated ERBB signaling10 (Fig 1C). To compare with estrogen, we profiled estrogen-treated MCF-7 cells by using the exact same conditions as for IGF-I (Appendix). We found extensive overlap (approximately 31% of genes) between the IGF and estrogen signatures (Fig 1C and Appendix). Additional data supported this observation, as we found a virtually identical number of genes (approximately 31%) that overlapped between our IGF signature and another public estrogen signature.13 However, despite this large overlap and similarity between estrogen and IGF signatures, many well-known targets of estrogen signaling, such as PGR, NPY1R, and IRS1, were not among the IGF-upregulated genes and were, in fact, downregulated by IGF.

We also found significant, albeit minor, overlap between the IGF upregulated genes and sets of genes upregulated in MCF-7 cells that had overexpression of ligand-activated EGFR or ERBB2.10 The IGF signature was partly distinct from an EGFR/ERBB2 signature, because many genes showed completely discordant regulation between IGF and EGFR/ERBB2, and the majority of IGF-regulated genes were not altered by expression of EGFR/ERBB2 (Fig 1C; Appendix Fig A5A, online only).

IGF-Induced Gene Expression Patterns Associated With ER-Negative Breast Cancer

We examined the expression patterns of the genes induced or repressed by IGF-I at both 3 and 24 hours (IGF gene signature; Fig 1C) in human breast tumors to ascertain the possible clinical context and relevance. As stated previously, the IGF gene signature has transcripts removed that were involved in cell cycle progression so that any associations made between the remaining genes and the clinical tumors would not be caused by generic cell proliferation effects (Appendix Fig A6A, online only). When we examined the levels of the genes in our signature in three independent expression profile data sets of clinical breast tumors from van de Vijver et al,14 Wang et al,15 and Miller et al,16 a striking correlation was observed (Fig 2A). Most of the ER-negative breast tumors showed high expression of the genes induced in vitro by IGF and low expression of the repressed genes, and this was highly consistent among the three independent data sets. In ER-positive tumors, approximately 25% (which had relatively lower levels of ER) showed a similar correlation with the IGF-I signature. Genes in our IGF gene signature appeared coordinately expressed in human breast tumors, even with rigorous testing for chance associations (Appendix).

Fig 2.
An insulin-like growth factor (IGF) gene signature in clinical breast tumors is associated with estrogen receptor (ER)–negative status and basal/luminal “B” tumor subtypes. (A) Genes in the IGF gene signature shown in Figure 1C ...

We examined an additional clinical breast tumor profile data set from Sorlie et al,17 in which the profiles were subtyped on the basis of gene clustering. The IGF gene signature appeared activated in most of the basal tumors (associated with ER-negative/HER2-negative status), in most of the luminal B tumors (ER-positive with poor prognosis), and in a sizable subset of the ERBB2-positive tumors (ER-negative/HER2-positive), but not in the luminal A or normal breast-like tumors (Fig 2B).

Gene Signature Patterns of IGF Signaling Associated With Adverse Pathologic and Clinical Markers and With Poor Prognosis in Breast Cancer

We defined an IGF activation score for a given tumor as the Pearson correlation between the IGF gene signature pattern (by using 1 and −1 for up and down, respectively) and the gene expression values of the tumor. In the van de Vijver, Wang, and Miller data sets, the activation score was inversely correlated with mRNA levels (as measured on the array) of both ER and PR within the ER-positive tumors (r < −0.3, P < .00001; and r < −0.18, P < .01, respectively, Appendix Table A1). In the van de Vijver and Miller data sets, the scores were highly correlated with increasing grade (r = 0.46, P < 5 × 10−13; and r = 0.46, P < 1 × 10−12, respectively), tumor size (r = 0.15, P < .02; and r = 0.23, P < .001, respectively) and lymph nodes (r = 0.20, P = .002; and r = 0.22, P = .002, respectively) within ER-positive tumors in both data sets and were inversely correlated with age in the Miller data set (r = −0.18, P < .01; Appendix Table A1).

Given the association of the IGF gene signature with known poor prognostic markers, we examined the association of the signature with patient outcome. Kaplan-Meier curves and univariate Cox analysis showed that patients with tumors that had high IGF activation scores had reduced disease-specific survival or metastasis-free survival compared with patients who had tumors that showed an inverse correlation with the IGF-I signature (Fig 3A). Interestingly, patients with tumors that had no significant correlation with the IGF-I signature (P > .01, Pearson) had an intermediate prognosis. As described above, the IGF gene signature showed activation predominantly in ER-negative breast cancer. As ER-negative breast cancer frequently shows poorer prognosis compared with ER-positive breast cancer,18 we examined the subset of ER-positive cancers and found the IGF gene signature to be an indicator of outcome there as well (Fig 3B).

Fig 3.
Gene expression patterns associated with insulin-like growth factor (IGF) signaling are correlated with poor prognosis. Kaplan-Meier analysis of three tumor profile data sets (A and D from Van de Vijver et al; B and E from Wang et al; and C and F from ...

The Wang data set represented patients who had small tumors, who were node-negative, and who received only local therapy, which thus indicated that the IGF gene signature reflects poor prognosis independent of treatment. The van de Vijver and Miller data sets both contained a proportion of patients who received adjuvant therapy. We examined the effect of the IGF gene signature in only the subset of 165 patients from the van de Vijver data set that did not receive hormone therapy or chemotherapy, and again we found that the IGF-I signature predicted poor outcome (Appendix Fig A7, online only). Therefore, the IGF-I signature is a highly significant indicator of poor prognosis.

In a multivariate Cox model, in which poor patient outcome was evaluated among tumors in relation to ER status, ER mRNA, PR mRNA, HER2 mRNA, age, grade, tumor size, and lymph node status, the IGF gene signature provided significant predictive power that was independent of the other clinical variables (Table 1; van de Vijver P < .001, Wang P = .004, and Miller P = .006). Analysis of the Cox multivariate models with and without the IGF gene signature indicated that the gene signature replaced grade in the model (presumably caused by the high correlation of the signature with grade; Appendix Table A1). Importantly, the prognostic ability of the IGF signature was not a result of genes generic to proliferation, because a proliferation score derived from 763 genes in the cell cycle signature from Whitfield et al.12 et al was prognostic (P < .05, all three data sets) in univariate Cox analysis, but the prognostic ability was lost in multivariate models that incorporated other clinical variables with or without the IGF signature activation score. (Appendix Table A2).

DISCUSSION

In this study, we experimentally defined a gene transcription signature of IGF-I signaling in human MCF-7 breast cancer cells. IGF-I altered levels of a large portion of the transcriptome (approximately 21%), with genes involved in cell growth, survival, metabolism, and biosynthesis. IGF-I–altered mRNA levels of numerous members of both the PI3K and ERK1/2 pathways and the IGF-I signature were enriched for transcriptional targets of PI3K/Akt/mTOR and Ras/MAPK pathways. The IGF gene signature was highly activated in ER-negative breast tumors but also in a sizeable subset (approximately 25%) of ER-positive tumors that had low levels of ER, which suggests that this transcriptional program may be important in the progression to loss of ER and hormone independence. The IGF-I signature was significantly associated with many poor prognostic markers in breast cancer (eg, negative ER/PR status; increasing grade, tumor size, and nodal involvement) and was highly prognostic when both ER-positive and ER-negative tumors were considered together and ER-positive tumors were considered alone (although the signature was not prognostic in ER-negative tumors; results not shown).

Although our data indicate that IGF-I–regulated genes are coordinately expressed in breast cancer and are associated with poor outcome, it is important to note that, because of the inherent cross-talk between numerous growth factor and hormone signaling pathways, the IGF gene signature may in part highlight tumors that are stimulated by hormones or growth factors that share some of the same downstream transcriptional targets. For example, the IGF signature encompasses transcriptional programs downstream of activated Erk1/2 and PI3K/Akt/mTOR. In addition, we found significant overlap between our IGF gene signature and gene signatures derived from MCF-7 cells stimulated with E2 or those that overexpressed EGFR or ERBB2. The overlap in transcriptional programs presumably represents not only the ability of IGFs, E2, and ERBBs to activate similar downstream signaling intermediates (eg, PI3K/Akt) but also the ability of IGFs to directly phosphorylate and activate both ER19 and ERBBs.20 Although significant overlap was noted with many transcriptional programs, the overlap was often minor (especially in the case of EGFR/ERBB); thus, the IGF signature contains transcriptional programs that are specific and unique to IGF signaling.

We compared our IGF-I signature with other gene signatures previously found prognostic in breast cancer (Appendix Fig A8, online only), including the 70-gene,21 wound healing,22 p53-mutation,16 and HOXB13:IL17BR23 signatures. We found a small but significant overlap of the IGF-I–induced genes in the wound healing signature (29 genes overlapped of 233 in the wound healing, activated set), but we found no significant overlap in the 70-gene set. In the van de Vijver breast tumor profile data set, the IGF, 70-gene, p53-mutation, and wound-healing signatures tended to identify many of the same tumors as poor versus good prognosis (Appendix Fig A8A), and such concordance among the prognostic gene signatures has been noted elsewhere.24 Not surprisingly, multivariate Cox analysis (data not shown) indicated that expression of the IGF gene signature was not independent of that of the 70-gene or the wound-healing signature. We found the PI3K gene signature itself (from CMap) to be prognostic and associated with ER negativity in a similar manner to the IGF-I gene signature (results not shown); however, when the PI3K gene signature was removed from the IGF-I signature, the IGF gene signature was still highly prognostic, which indicated that this signature provided information above and beyond PI3K (Appendix Fig A9, online only).

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The author(s) indicated no potential conflicts of interest.

AUTHOR CONTRIBUTIONS

Conception and design: Chad J. Creighton, Adrian V. Lee

Collection and assembly of data: Chad J. Creighton, Angelo Casa, ZaWaunyka Lazard, Shixia Huang, Anna Tsimelzon, Adrian V. Lee

Data analysis and interpretation: Chad J. Creighton, Angelo Casa, ZaWaunyka Lazard, Shixia Huang, Anna Tsimelzon, Susan G. Hilsenbeck, Charles Kent Osborne, Adrian V. Lee

Manuscript writing: Chad J. Creighton, Susan G. Hilsenbeck, Charles Kent Osborne, Adrian V. Lee

Final approval of manuscript: Chad J. Creighton, Angelo Casa, Susan G. Hilsenbeck, Charles Kent Osborne, Adrian V. Lee

Supplementary Material

[Data Supplements]

Acknowledgments

We thank Gary Chamness, PhD, for comments on the manuscript.

Appendix

Materials and Methods

Affymetrix microarray analysis.

Estimated probe set expression values were log-transformed and centered on the average of the serum-free medium (SFM) controls for each respective time point. Probe sets called as present in fewer than half of the samples were excluded from the analysis. Two-sample t tests using log-transformed data were performed as criteria for determining significant differences in mean gene mRNA levels between groups of samples. Fold changes between groups were estimated by taking the ratio of the average of the untransformed expression values of the one group to that of the other. Gene Ontology (GO) annotation terms were searched for enrichment within gene sets essentially as described in Creighton et al (Creighton C, Kuick R, Misek DE, et al: Genome Biol 4:R46, 2003). For the human breast tumor data sets, gene expression values were log-transformed and normalized to standard deviations from the median of the estrogen receptor (ER) –positive tumor profiles. For the Miller et al data set, only the U133A arrays were analyzed; the four profiles from the original Miller data set for which ER status was not given were removed from the analyses.

The supervised clustering analysis for differentially expressed genes (Fig 1A) was carried out in the following steps: (1) each pattern of interest was represented as a series of 1's, 0's, and −1's; (2) for each gene, the Pearson's correlation was computed between its expression values and each of the predefined patterns; (3) the pattern best correlated with the expression of each gene was determined, and the genes were manually sorted based on their assigned patterns. Mapping between independent expression profile datasets was done using either the Affymetrix probe set identifier (Fig 1B) or the Entrez gene identifier (Fig 2); in the case of the Entrez identifier, one probe set was selected to represent each gene (van de Vijver, Wang, and Miller datasets, the probe set with the highest variation; Sorlie data set, the probe with the most unflagged values, followed by the probe with the most variation). For the Sorlie data set, genes with flagged values in more than half of the tumors were removed from the analysis. The CMap data set (Fig 1B) was normalized by first log-transforming the values and centering on the batch control profiles, and then subtracting the median and dividing by the standard deviation within each array. For one-sided Fisher's exact tests for the significance of overlap between the insulin-like growth factor (IGF) gene signature and other gene sets, the entire population of 12,724 genes represented on the U133A array was used as the denominator.

Ingenuity Pathways Analysis.

Data were analyzed through the use of Ingenuity Pathways Analysis (IPA; Ingenuity Systems, Redwood City, CA; www.ingenuity.com) as described below.

Network generation.

A data set containing gene identifiers and corresponding expression values was uploaded into the application. Each gene identifier was mapped to its corresponding gene object in the Ingenuity Pathways Knowledge Base. A fold change cutoff of 1.85 was set to identify genes whose expression was significantly differentially regulated. This was the lowest fold cutoff that generated a number of genes that could be analyzed. Similar results were obtained using higher fold cutoffs of 2.0 and 2.5. Genes whose expression was differentially regulated, called focus genes, were overlaid onto a global molecular network developed from information contained in the Ingenuity Pathways Knowledge Base. Networks of these focus genes were then algorithmically generated based on their connectivity.

Functional analysis of an entire data set.

The functional analysis identified the biologic functions and/or diseases that were most significant to the data set. Genes from the data set that met the fold change cutoff of 1.85 and were associated with biologic functions and/or diseases in the Ingenuity Pathways Knowledge Base were considered for the analysis. Fisher's exact test was used to calculate a P value determining the probability that each biologic function and/or disease assigned to that data set is due to chance alone.

Functional analysis of a network.

The functional analysis of a network identified the biologic functions and/or diseases that were most significant to the genes in the network. The network genes associated with biologic functions and/or diseases in the Ingenuity Pathways Knowledge Base were considered for the analysis. Fisher's exact test was used to calculate a P value determining the probability that each biologic function and/or disease assigned to that network is due to chance alone.

Canonical pathway analysis—entire data set.

Canonical pathways analysis identified the pathways from the IPA library of canonical pathways that were most significant to the data set. Genes from the data set that met the fold change cutoff of 1.85 and were associated with a canonical pathway in the Ingenuity Pathways Knowledge Base were considered for the analysis. The significance of the association between the data set and the canonical pathway was measured in two ways: a ratio of the number of genes from the data set that map to the pathway divided by the total number of genes that map to the canonical pathway is displayed; and Fisher's exact test was used to calculate a P value determining the probability that the association between the genes in the data set and the canonical pathway is explained by chance alone.

Network graphical representation.

A network is a graphical representation of the molecular relationships between genes or gene products. Genes or gene products are represented as nodes, and the biologic relationship between two nodes is represented as an edge (line). All edges are supported by at least one reference from the literature, from a textbook, or from canonical information stored in the Ingenuity Pathways Knowledge Base. Human, mouse, and rat orthologs of a gene are stored as separate objects in the Ingenuity Pathways Knowledge Base, but are represented as a single node in the network. The intensity of the node color indicates the degree of upregulation (red) or downregulation (green). Nodes are displayed using various shapes that represent the functional class of the gene product. Edges are displayed with various labels that describe the nature of the relationship between the nodes (eg, P for phosphorylation, T for transcription).

Results

Strong similarity between a transcriptional signature of MCF-7 cells stimulated with IGF and an independently derived transcriptional signature of MCF-7 cells stably overexpressing IGF-I.

In a recent study by Pacher et al (Pacher M, Seewald M, Mikula M, et al: Carcinogenesis 28:49-59, 2007), gene expression profiles were taken of MCF-7 cells stably overexpressing IGF-I in in vitro analyses. The entire profile data set from Pacher study was not readily accessible. However, we did compare the published short list of most differentially expressed genes in the Pacher model with results from our profile data of IGF-I treatment. Out of the 24 genes upregulated by IGF-I in the Pacher study that were also represented on our array platform, 15 (ABCA12, ASNS, ASS, DCAMKL1, GADD45A, HERPUD1, KIAA0626, MARS, PHGDH, PLAB, SLC1A4, SLC3A2, SLC7A11, SLC7A5, VEGF) were likewise overexpressed (P < .05) by IGF-I at 24 hours in our data (chance expected overlap of four genes, P < 1 × 10−6, one-sided Fisher's exact test). Out of the five genes downregulated by IGF-I in the Pacher study, 3 (CAV1, KCNJ8, RAB31) were likewise downregulated in our data at 24 hours.

Genes with early and sustained regulation by IGF are enriched for transcriptional targets of the Ras/ERK1/2 and PI3K/Akt/mTOR pathways.

When viewing the expression patterns of early and sustained targets of IGF in one profile data set (Bild AH, Yao G, Chang JT, et al: Nature 439:353-357, 2006) of breast cells activated with the oncogenes Myc, Src, beta-catenin, E2F3, and H-Ras, we found widespread similarities between the early and sustained targets of IGF and the Ras signature pattern (Fig 1B); in other words, a significant number of genes upregulated by IGF in our data (422 named genes) were likewise upregulated (P < .001) by Ras in the other data set (181 overlapping of 2998 in total represented on the U133A array of 12,725 genes, chance expected overlap 100, one-sided Fisher's exact P < 1 × 10−17), and many of the 481 genes downregulated by IGF were also downregulated (P < .001) by Ras (overlap of 206, one-sided Fisher's exact P < 1 × 10−17). Notably, the HRAS gene itself showed upregulation by IGF in the array data (Fig 1B). We also found significant overlap between a previously defined set of 118 genes upregulated by MAP kinase (Creighton CJ, Hilger AM, Murthy S, et al: Cancer Res 66:3903-3911, 2006) and our IGF upregulated genes (14 overlapping, four expected by chance; Fisher's exact P < 5 × 10−5).

We went on to compare the early and sustained targets of IGF (Fig 1B) against a public database (www.broad.mit.edu/cmap) of gene expression profiles of cultured cells (mostly MCF-7 cells) under treatment with various small molecule inhibitors (164 in all), called the Connectivity Map, or CMap (Lamb J, Crawford ED, Peck D, et al: Science 313:1929-1935, 2006). CMap analysis indicated that early and sustained targets of IGF showed an inverse correlation with the expression patterns induced in vitro by PI3K inhibitors LY-294002 and wortmannin, by the mammalian target of rapamycin (mTOR) inhibitor rapamycin, or by the histone deacetylase inhibitor trichostatin A. These associations were evident when viewing the expression patterns of the CMap profile data set corresponding to the early and sustained targets of IGF (Fig 1B). A significant number of the 422 genes upregulated by IGF were also downregulated by inhibition of PI3K (179 overlapping of 1,046, with P < .01, comparing LY-294002 and wortmannin profiles with the rest of the CMaP profiles, Fisher's exact P < 1 × 10−80); genes downregulated by IGF likewise tended to be upregulated by inhibition of PI3K. The patterns associated with PI3K inhibition in CMaP were shared by the patterns associated with mTOR inhibition, and, to a lesser extent, with Ras signaling and histone deacetylase inhibition (Fig 1B). For an independent assessment of Akt/mTOR pathway activation, we obtained a set of 351 genes found activated by Akt with attenuated expression by mTOR inhibition in vivo in a transgenic mouse model (based on analyses in Majumder et al and Creighton [Majumder P, Febbo P, Bikoff R, et al: Nat Med 10:594-601, 2004; Creighton CJ: Oncogene 26:4648-4655, 2007]) and found these to overlap with the IGF upregulated genes (31 genes, Fisher's exact P < 1 × 10−6).

Genes with early and sustained regulation by IGF are enriched for transcriptional targets of the epidermal growth factor receptor (EGFR) and HER2 pathways.

We found a significant overlap of 79 genes between the 422 IGF upregulated genes and a set of 713 genes upregulated (with P < .01) in MCF-7 cells having overexpression of ligand-activated EGFR (P < 1 × 10−20, one-sided Fisher's exact), the EGFR-associated genes being obtained from a public data set from Creighton et al (Creighton CJ, Hilger AM, Murthy S, et al: Cancer Res 66:3903-3911, 2006). Likewise, there was significant overlap between IGF upregulated genes and genes upregulated by HER2 in the data set from (Creighton CJ, Hilger AM, Murthy S, et al: Cancer Res 66:3903-3911, 2006). From 1275 HER2-associated genes with P < .01, there was an overlap of 74 with the IGF genes (P < 1 × 10−8, one-sided Fisher's exact).

The similarities and differences between the IGF gene signature and gene signatures of EGFR or HER2 (ERBB2) can be illustrated using a heat map representation (Fig A5). We viewed the patterns of the IGF signature genes in the profile data set from Creighton et al (Creighton CJ, Hilger AM, Murthy S, et al: Cancer Res 66:3903-3911, 2006) of MCF-7 cells with activation of EGFR, HER2, Raf, or MEK. Many of the genes upregulated in the IGF signature were likewise upregulated in one or more of the oncogene-activated cell lines; similarly, many of the genes downregulated in the IGF signature were downregulated in at least one oncogene-activated cell line. Although the gene overlap between the IGF and EGFR/HER2 signatures described above was significant, the actual overlap still represents but a fraction of the entire IGF signature (Fig A5). In addition, many genes actually showed discordant regulation, being upregulated by IGF but downregulated by EGFR/HER2 and vice versa. The similarities in signatures presumably highlight the redundant and shared nature of these growth factor signaling cascades, and the differences may represent distinct transcriptional features of IGF signaling that are not generic to growth factor signaling. Notably, if we were to remove the set of genes common to both the IGF and HER2 or EGFR signatures, the remaining IGF signature as a group shows the same patterns of coordinate expression in breast tumors described elsewhere; for instance, the remaining genes are highly expressed in ER-negative breast cancer (Fig A5).

Genes with early and sustained regulation by IGF are enriched for transcriptional targets of estrogen.

We generated expression profiles of estrogen treatment of MCF-7 to compare with our profiles of IGF treatment. In an analogous manner to that for the IGF data set, MCF-7 cells were starved overnight in serum-free medium, and then stimulated with or without 1 nmol/L estradiol (E2) for 3 or 24 hours. RNA was then isolated and gene expression array profiles were analyzed. At 24 hours, 884 unique named genes were upregulated (P < .05) and 856 were downregulated by E2. Of the 422 genes upregulated by IGF in our signature, 132 were also upregulated by E2 within 24 hours (chance expected overlap = 29, one-sided Fisher's exact P < 1 × 10−50). Of these 132 genes, 33 were among a set of genes found to be correlated in expression with cell cycle progression in another profiling study (Whitfield ML, Sherlock G, Saldanha AJ, et al: Mol Biol Cell 13:1977-2000, 2002), and so these genes may be generic to growth and not specific to any signaling pathways. However, the remaining 99 shared genes still represented a highly significant overlap (P < 1 × 10−25). The overlap between the E2 and IGF signatures not due to cell cycle–associated genes are illustrated in Fig 1C (where cell cycle genes were not represented).

We furthermore compared our gene expression signature of IGF signaling (Fig 1B) with a previously published gene signature of estrogen signaling in multiple breast cancer cell lines in vitro (Creighton CJ, Cordero KE, Larios JM, et al: Genome Biol 7:R28, 2006), and found extensive overlap between these two as well. Of the 422 genes upregulated by IGF in our signature, 129 were also found upregulated by estradiol (E2) within 24 hours (one-sided Fisher's exact P < 1 × 10−35), and 52 were upregulated within 4 hours. Of these 129 genes, 24 were among the set of genes found to be correlated in expression with cell cycle progression (Whitfield ML, Sherlock G, Saldanha AJ, et al: Mol Biol Cell 13:1977-2000, 2002). However, the remaining 105 shared genes still represented a highly significant overlap (P < 1E-25).

IGF gene signature is not a generic proliferation signature.

Our IGF gene signature was obtained from serum-starved cells that were then stimulated to grow with IGF-I. One concern we had was that many of the genes expressed in these cells might be generic to the cell cycle and proliferation. If our IGF gene signature were merely a reflection of more rapidly proliferating tumors, then any associations made between the signature and breast tumor expression patterns would not be meaningful. We did observe higher enrichment for proliferation-related genes being expressed at 24 hours but not at 3 hours (Fig 1A), and so to define the IGF signature (Fig 1C), we selected genes expressed at both 3 and 24 hours as representing earlier and more direct targets of IGF signaling (Fig 1B), and then we removed 89 genes from the original 903-gene signature that were either annotated as cell cycle by GO or found experimentally to be correlated with cell cycle progression by Whitfield et al (Whitfield ML, Sherlock G, Saldanha AJ, et al: Mol Biol Cell 13:1977-2000, 2002).

To determine whether we had successfully removed any cell proliferation effects underlying the IGF signature, we considered a gene expression profile data set of the NCI60 cell lines done in triplicate. The cell lines were grown in culture, and it can be presumed that some cell cultures were growing at a faster rate than others, due to culture conditions as well as cell line genotype. We performed an in silico experiment where we first identified cell cultures from the NCI60 data set that showed a high manifestation of the Whitfield et al cell cycle signature (Fig A6, left panel). We then examined the expression patterns of genes in the IGF signature in these same cultures (Fig A6, right panel). Cell cultures that manifested the cell proliferation signature (which represented a variety of tissues of origin) did not likewise manifest the IGF gene signature.

Statistical significance of the IGF signature in human breast tumors.

Coordinate expression of IGF signature genes in ER-negative versus ER-positive samples was tested statistically using Q1Q2 analysis (Tian L, Greenberg SA, Kong SW, et al: Proc Natl Acad Sci U S A 102:13544-13549, 2005; Creighton CJ: Oncogene 26:4648-4655, 2007). This analysis compares a statistic summarizing differential expression of all 814 genes in the IGF signature between two phenotypes (ER negative v ER positive) to what would be expected with the same phenotype groups for a random sample of 814 genes (Q1) and to what would be expected with the IGF gene set for randomly assigned phenotypes (Q2). In this analysis in the Wang data set, genes induced by IGF were highly expressed as a group (P ≈ 0) in ER-negative over ER-positive tumors, and genes repressed by IGF were underexpressed (P ≈ 0). We next examined the degree to which individual tumors exhibit the IGF activation gene signature (or its opposite) by approximating the IGF gene signature pattern as a series of numbers 1 and −1 (for upregulated and downregulated, respectively) and computing the Pearson correlation between the ideal pattern and gene expression of a particular sample (IGF activation score). One hundred nineteen of 209 clinical ER-positive breast tumors from the Wang data set were significantly correlated or anticorrelated (P < .01, Pearson's) with the IGF-I pattern. When randomly assigning upregulated and downregulated designations to the IGF signature genes, only three on average of 209 were similarly correlated (close to that expected by chance, based on 1,000 tests). Therefore, the correspondence between the patterns of upregulation or downregulation of genes in our IGF gene signature and patterns of high and low expression of these same genes in human breast tumors is highly significant and presumably has important biologic significance.

Prognostic ability of the IGF signature not due to genes generic to proliferation.

From the IGF gene signature, we had removed genes believed to be involved in general processes of cell proliferation (Fig A6). We subsequently found the IGF signature to be correlated with poor prognosis in breast tumors. It has long been known that tumors with more rapidly proliferating cells (as measured by S phase markers) have poor prognosis. We therefore sought to test the IGF signature versus a proliferation signature to determine how correlated they are in human breast tumors. We used the 763 genes in the cell cycle signature from Whitfield et al (Whitfield ML, Sherlock G, Saldanha AJ, et al: Mol Biol Cell 13:1977-2000, 2002), to create an average proliferation score for each patient in the van de Vijver, Wang, and Miller breast tumor datasets. We compared this proliferation score to the IGF-I pathway activation score.

In univariate Cox analysis, the proliferation score was prognostic (P < .05 in all three datasets). However, in a multivariate Cox model in which distant recurrence was evaluated across tumors in relation to ER status, ER mRNA, PR mRNA, HER2 mRNA, age, grade, tumor size, and lymph node status, in addition to the Whitfield proliferation score, the IGF gene signature provided significant predictive power that was independent of the other variables (Table A2 van de Vijver, P = .0001; Wang, P = .06; and Miller P = .02), including the proliferation score. The prognostic ability of the proliferation score in univariate analysis was lost in multivariate models that incorporated other clinical variables, with or without the IGF pathway activation score. On the basis of these findings, we concluded that the IGF activation score is not merely tracking proliferation in human breast tumors.

Comparison of the prognostic ability of the IGF gene signature with that of previously published prognostic signatures.

Given that a number of gene signatures found to be prognostic in breast cancer have been published, we compared the IGF activation score with the following prognostic signatures: the 70-gene signature from van't Veer et al (van't Veer LJ, Dai H, van de Vijver MJ, et al: Nature 415:530-536, 2002), the wound-healing signature from Chang et al (Chang HY, Sneddon JB, Alizadeh AA, et al: PLoS Biol 2:E7, 2004), the p53 mutation signature from Miller et al (Miller LD, Smeds J, George J, et al: Proc Natl Acad Sci U S A 102:13550-13555, 2005), and the HOXB13:IL17BR signature from Ma et al (Ma X, Wang Z, Ryan PD, et al: Cancer Cell 5:607-616, 2004). Like the IGF signature, each of the previously published signatures consisted of a set of upregulated and downregulated genes, and so we used the same approach as for the IGF score to create a score for each of the other signatures, taking the Pearson's correlation between the given gene signature pattern (using 1 and −1, for up and down, respectively) and the gene expression values of the tumor.

For the van de Vijver breast tumor profile data set, the IGF, 70-gene, p53 mutation, and wound-healing signatures tended to pick out many of the same tumors as having poor versus good prognosis (Fig A8A). Such concordance among the prognostic gene signatures has been noted elsewhere (Fan C, Oh D, Wessels L, et al: N Engl J Med 355:560-569, 2006). For the van de Vijver tumors, Kaplan-Meier analysis was carried out separately for each of the five signatures, comparing patients showing high activation (P < .01), patients showing low activation (P < .01), and the rest of the patients (intermediate group). For the IGF, 70-gene, and wound-healing signatures, patients with tumors having high activation scores had reduced metastasis-free survival compared with patients with tumors having low activation scores (Fig A8B); patients in the intermediate group had an intermediate prognosis. These patterns were observed both for all 295 tumors and for the subset of 226 ER-positive tumors. The p53 mutation signature was prognostic for the 295 tumors but not for the 226 ER-positive tumors (Fig A8B). The HOXB13:IL17BR signature was not prognostic in the van de Vijver data set, though this has been noted elsewhere (Fan C, Oh D, Wessels L, et al: N Engl J Med 355:560-569, 2006).

PI3K/Akt/mTOR is not entirely responsible for the gene signature indicating poor outcome.

The PI3K/Akt/mTOR pathway is highly mutated or overexpressed in breast cancer and often correlates with poor outcome (Vivanco I, Sawyers CL: Nat Rev Cancer 2:489-501, 2002). Given that our IGF-I signature was highly correlated with PI3K signaling, we examined whether the signature was simply an indicator of this pathway. Removal of PI3K-related signaling genes (using CMaP profile data set definition with P < .01) did not affect the ability of the signature to predict poor outcome and indicates that the IGF signature provides information above and beyond the PI3K signaling pathway (Fig A9).

Fig A1.
Hierarchical clustering of genes regulated by IGF-I in MCF-7 cells at either 3 or 24 hours, or both (P < .01, fold change > 1.5).

Fig A2.
Ingenuity Pathways Analysis (IPA) of biologic functions and/or disease enriched in the IGFI-gene expression data set and ranked according to significance after (A) 3 hours or (B) 24 hours of insulin-like growth factor I (IGF-I). IPA of biologic functions ...

Fig A3.
Ingenuity Pathways Analysis (IPA) of canonical pathways significantly enriched in the gene expression data set ranked according to data at (A) 3 hours or (B) 24 hours of insulin-like growth factor I (IGF-I). IPA canonical pathways analysis identified ...

Fig A4.
Ingenuity Pathways Analysis (IPA) network map highlighting regulation of a G2/M DNA damage checkpoint after 24 hours insulin-like growth factor I (IGF-I) stimulation. Network graphical representation of the molecular relationships between genes/gene products ...

Fig A5.
Comparison of the IGF gene signature with gene signatures of EGFR and ERBB2. Heat map representation of genes in the insulin-like growth factor (IGF) signature of Figure 2B (with cell cycle–associated genes removed). Alongside the IGF treatment ...

Fig A6.
The IGF gene signature is not a generic proliferation signature. A data set of in-triplicate NCI60 gene expression data generated by Novartis on the Affymetrix U95v2 array platform (http://dtp.nci.nih.gov/mtargets/madownload.html) was considered. Cell ...

Fig A7.
Kaplan-Meier analysis of the insulin-like growth factor (IGF) signature in the subset of 165 patients in the van de Vijver data set that did not receive hormone or chemotherapy. P values by log-rank statistic.

Fig A8.Fig A8.
Comparison of the prognostic ability of the IGF gene signature with that of previously published prognostic signatures. In addition to the IGF gene signature, an activation score was derived for each of four previously published prognostic gene signatures ...

Fig A9.
Kaplan-Meier analysis after removing PI3K-associated genes from insulin-like growth factor (IGF) signature. P values by log-rank statistic.

Table A1.

Significance of Correlations Between Tumor Characteristics and IGF Gene Signature Pattern Within ER-Positive Clinical Tumors

Characteristicvan de VijverWangMiller
ER (mRNA)< 5 × 10−7 (−)< 1 × 10−5 (−)< 5 × 10−10 (−)
PR (mRNA).005 (−)< .01 (−)< 1 × 10−5 (−)
HER2 (mRNA).0003 (+)NS< 1 × 10−5 (+)
AgeNSNA< .01 (−)
Grade< 5 × 10−13 (+)NA< 1 × 10−12 (+)
Tumor size.02 (+)NA< .001 (+)
Lymph nodes.002 (+)NA.002 (+)

NOTE.P values by Pearson's coefficient. Correlation between IGF signature coefficient and tumor characteristic treated as a continuous variable (except for lymph nodes variable in Miller dataset, which was binned as + or −).

Abbreviations: ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2; NS, not significant (P > .05); NA, data not available; +, positive, −, negative correlation; IGF, insulin-like growth factor.

Table A2.

Multivariate Cox Proportional Analysis of Tumor Characteristics, Cell Cycle Gene Signature, and IGF Gene Signature in Relation to the Likelihood of Poor Outcome

Variablevan de Vijver
Wang
Miller
PHazard Ratio95% CIPHazard Ratio95% CIPHazard Ratio95% CI
Analysis without IGF gene signature
    Age at surgery.360.760.43 to 1.36.460.770.38 to 1.55
    Clinical tumor size.041.551.01 to 2.36.00163.061.53 to 6.12
    Tumor grade.011.551.11 to 2.17.561.160.70 to 1.93
    Lymph node status.480.870.58 to 1.30< .0012.911.62 to 5.23
    ER status.381.540.59 to 4.02.791.090.57 to 2.10.241.920.64 to 5.78
    ER mRNA (ESR1).770.940.62 to 1.42.11.270.95 to 1.70.681.090.72 to 1.64
    PR mRNA (PGR).0040.700.55 to 0.89.0010.700.57 to 0.87.450.880.63 to 1.22
    HER2 mRNA (ERBB2).221.130.93 to 1.36.840.980.81 to 1.19.0771.270.97 to 1.67
    Cell cycle signature.231.160.91 to 1.49.0231.281.04 to 1.59.171.270.90 to 1.79
Analysis with IGF gene signature
    Age at surgery.370.770.43 to 1.37.410.740.36 to 1.51
    Clinical tumor size.041.551.01 to 2.38.00362.821.40 to 5.68
    Tumor grade.251.230.87 to 1.76.710.900.53 to 1.54
    Lymph node status.20.770.51 to 1.15< .0013.071.70 to 5.53
    ER status.321.640.62 to 4.33.641.170.60 to 2.28.311.780.59 to 5.39
    ER mRNA.781.060.69 to 1.64.0371.391.02 to 1.88.241.300.84 to 2.00
    PR mRNA.020.750.59 to 0.96.0010.700.57 to 0.87.520.900.64 to 1.25
    HER2 mRNA.291.110.92 to 1.34.740.970.80 to 1.18.0781.270.97 to 1.65
    Cell cycle signature.780.960.74 to 1.25.541.090.82 to 1.44.681.080.75 to 1.56
    IGF signature.00011.881.37 to 2.60.0641.340.98 to 1.83.0161.751.11 to 2.75

NOTE. Age, size, lymph node status, and ER status were binary variables (0 or age < 50 years, and 1 for an age ≥ 50 years; 0 for diameter ≤ 2 cm, and 1 for a diameter > 2 cm; 0 for lymph node negative, and 1 for lymph node positive; and 0 for ER negative and 1 for ER positive, respectively). Tumor grade was binned as 1, 2, and 3, for low, medium, and high, respectively. ER mRNA, PR mRNA, HER2 mRNA, the cell cycle gene signature coefficient, and the IGF gene signature coefficient variables were each transformed to standard deviations from the mean.

Abbreviations: IGF, insulin-like growth factor; ER, estrogen receptor, PR, progesterone receptor; HER2, human epidermal growth factor receptor 2;

Table 1.

Multivariate Cox Proportional Analysis of Tumor Characteristics and IGF Gene Signature in Relation to the Likelihood of Poor Outcome

VariableData Sets*
van de Vijver14
Wang15
Miller16
PHR95% CIPHR95% CIPHR95% CI
Analysis Without IGF Gene Signature
Age at surgery.350.760.42 to 1.37.370.730.36 to 1.48
Clinical tumor size.031.581.03 to 2.43< .0013.201.58 to 6.59
Tumor grade< .0011.691.24 to 2.31.261.320.81 to 2.15
Lymph node status.610.900.60 to 1.36< .0012.791.54 to 5.07
ER status.401.510.57 to 3.96.831.080.55 to 2.09.261.80.61 to 5.78
ER mRNA.670.910.60 to 1.39.311.160.87 to 1.55.831.050.69 to 1.59
PR mRNA.0040.700.55 to 0.90< .0010.690.55 to 0.85.490.890.63 to 1.25
HER2 mRNA.391.080.90 to 1.30.690.960.79 to 1.17.071.280.97 to 1.70
Analysis With IGF Gene Signature
Age at surgery.370.770.43 to 1.38.380.730.36 to 1.50
Clinical tumor size.051.541.00 to 2.38.0042.831.38 to 5.78
Tumor grade.261.220.86 to 1.72.760.920.54 to 1.58
Lymph node status.190.760.50 to 1.15< .0013.041.66 to 5.54
ER status.311.650.61 to 4.47.621.190.60 to 2.34.321.760.57 to 5.46
ER mRNA.781.060.68 to 1.65.041.381.01 to 1.89.241.290.83 to 2.01
PR mRNA.020.750.58 to 0.96< .0010.700.56 to 0.87.540.900.64 to 1.27
HER2 mRNA.211.120.93 to 1.35.690.960.79 to 1.17.081.270.97 to 1.65
IGF signature< .0011.861.36 to 2.53.0041.431.12 to 1.82.0061.811.17 to 2.79

NOTE. Age, size, lymph node status, and ER status were binary variables (0 for age < 50 years and 1 for an age ≥ 50 years; 0 for diameter ≤ 2 cm and 1 for diameter > 2 cm; 0 for lymph node–negative and 1 for lymph node–positive disease; and 0 for ER-negative and 1 for ER-positive disease, respectively).

Abbreviations: IGF, insulin-like growth factor; HR, hazard ratio; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.

*End points for van de Vijver, Wang, and Miller data sets were metastasis-free survival, distant-metastasis–free survival, and disease-specific survival, respectively.
Tumor grade was binned as 1, 2, and 3 for low, medium, and high, respectively.
ER mRNA, PR mRNA, HER2 mRNA, and the IGF gene signature coefficient variables each were transformed to standard deviations from the mean.

Notes

Supported in part by National Institutes of Health Grants No. P01CA30195 (A.V.L.), P30CA125123 (C.K.O.), and P50CA58183 (C.K.O.) and by a research grant from the Baylor College of Medicine/AstraZeneca Alliance (A.V.L.).

Authors’ disclosures of potential conflicts of interest and author contributions are found at the end of this article.

REFERENCES

1. Pollak MN, Schernhammer ES, Hankinson SE: Insulin-like growth factors and neoplasia. Nat Rev Cancer 4:505-518, 2004. [PubMed]
2. Pollak M: Insulin-like growth factor physiology and cancer risk. Eur J Cancer 36:1224-1228, 2000. [PubMed]
3. Pollak M: IGF-I physiology and breast cancer. Recent Results Cancer Res 152:63-70, 1998. [PubMed]
4. Surmacz E: Function of the IGF-I Receptor in Breast Cancer. J Mammary Gland Biol Neoplasia 5:95-105, 2000. [PubMed]
5. Sachdev D, Yee D: Disrupting insulin-like growth factor signaling as a potential cancer therapy. Mol Cancer Ther 6:1-12, 2007. [PubMed]
6. Cui X, Zhang P, Deng W, et al: Insulin-like growth factor-I inhibits progesterone receptor expression in breast cancer cells via the phosphatidylinositol 3-kinase/akt/mammalian target of rapamycin pathway: progesterone receptor as a potential indicator of growth factor activity in breast cancer. Mol Endocrinol 17:575-588, 2003. [PubMed]
7. Pacher M, Seewald M, Mikula M, et al: Impact of constitutive IGF1/IGF2 stimulation on the transcriptional program of human breast cancer cells. Carcinogenesis 28:49-59, 2007. [PubMed]
8. Bild AH, Yao G, Chang JT, et al: Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439:353-357, 2006. [PubMed]
9. Lamb J, Crawford ED, Peck D, et al: The Connectivity Map: Using gene-expression signatures to connect small molecules, genes, and disease. Science 313:1929-1935, 2006. [PubMed]
10. Creighton CJ, Hilger AM, Murthy S, et al: Activation of mitogen-activated protein kinase in estrogen receptor α-positive breast cancer cells in vitro induces an in vivo molecular phenotype of estrogen receptor α-negative human breast tumors. Cancer Res 66:3903-3911, 2006. [PubMed]
11. Yee D, Lee AV: Crosstalk between the insulin-like growth factors and estrogens in breast cancer. J Mammary Gland Biol Neoplasia 5:107-115, 2000. [PubMed]
12. Whitfield ML, Sherlock G, Saldanha AJ, et al: Identification of genes periodically expressed in the human cell cycle and their expression in tumors. Mol Biol Cell 13:1977-2000, 2002. [PMC free article] [PubMed]
13. Creighton CJ, Cordero KE, Larios JM, et al: Genes regulated by estrogen in breast tumor cells in vitro are similarly regulated in vivo in tumor xenografts and human breast tumors. Genome Biol 7:R28, 2006. [PMC free article] [PubMed]
14. van de Vijver MJ, He YD, van't Veer LJ, et al: A gene-expression signature as a predictor of survival in breast cancer. N Engl J Med 347:1999-2009, 2002. [PubMed]
15. Wang Y, Klijn JG, Zhang Y, et al: Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. Lancet 365:671-679, 2005. [PubMed]
16. Miller LD, Smeds J, George J, et al: An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival. Proc Natl Acad Sci U S A 102:13550-13555, 2005. [PMC free article] [PubMed]
17. Sorlie T, Tibshirani R, Parker J, et al: Repeated observation of breast tumor subtypes in independent gene expression data sets. Proc Natl Acad Sci U S A 100:8418-8423, 2003. [PMC free article] [PubMed]
18. Allred DC, Brown P, Medina D: The origins of estrogen receptor alpha-positive and estrogen receptor alpha-negative human breast cancer. Breast Cancer Res 6:240-245, 2004. [PMC free article] [PubMed]
19. Thorne C, Lee A: Cross talk between estrogen receptor and IGF signaling in normal mammary gland development and breast cancer. Breast Dis 17:105-114, 2003. [PubMed]
20. Knowlden JM, Hutcheson IR, Barrow D, et al: Insulin-like growth factor-I receptor signaling in tamoxifen-resistant breast cancer: A supporting role to the epidermal growth factor receptor. Endocrinology 146:4609-4618, 2005. [PubMed]
21. van 't Veer LJ, Dai H, van de Vijver MJ, et al: Gene expression profiling predicts clinical outcome of breast cancer. Nature 415:530-536, 2002. [PubMed]
22. Chang HY, Sneddon JB, Alizadeh AA, et al: Gene expression signature of fibroblast serum response predicts human cancer progression: Similarities between tumors and wounds. PLoS Biol 2:E7, 2004. [PMC free article] [PubMed]
23. Ma X, Wang Z, Ryan P, et al: A two-gene expression ratio predicts clinical outcome in breast cancer patients treated with tamoxifen. Cancer Cell 5:607-616, 2004. [PubMed]
24. Fan C, Oh D, Wessels L, et al: Concordance among gene-expression-based predictors for breast cancer. N Engl J Med 355:560-569, 2006. [PubMed]
25. Huang S, Podsypanina K, Chen Y, et al: Wnt-1 is dominant over neu in specifying mammary tumor expression profiles. Technol Cancer Res Treat 5:565-571, 2006. [PubMed]
26. Li C, Wong WH: Model-based analysis of oligonucleotide arrays: Expression index computation and outlier detection. Proc Natl Acad Sci U S A 98:31-36, 2001. [PMC free article] [PubMed]
27. Eisen MB, Spellman PT, Brown PO, et al: Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95:14863-14868, 1998. [PMC free article] [PubMed]
28. Saldanha AJ: Java Treeview–extensible visualization of microarray data. Bioinformatics 20:3246-3248, 2004. [PubMed]

Articles from Journal of Clinical Oncology are provided here courtesy of American Society of Clinical Oncology

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...