Abstract
Purpose: Estrogen withdrawal by treatment with aromatase inhibitors is the most effective form of endocrine therapy for postmenopausal estrogen receptor–positive (ER+) breast cancer. However, response to therapy varies markedly and understanding of the precise molecular effects of aromatase inhibitors and causes of resistance is limited. We aimed to identify in clinical breast cancer those genes and pathways most associated with resistance to aromatase inhibitors by examining the global transcriptional effects of AI treatment.
Experimental Design: Baseline and 2-week posttreatment biopsies were obtained from 112 postmenopausal women with ER+ breast cancer receiving neoadjuvant anastrozole. Gene expression data were obtained from 81 baseline and 2-week paired samples. Pathway analysis identified (i) the most prevalent changes in expression and (ii) the pretreatment genes/pathways most related to poor antiproliferative response.
Results: A total of 1,327 genes were differentially expressed after 2-week treatment (false discovery rate < 0.01). Proliferation-associated genes and classical estrogen-dependent genes were strongly downregulated whereas collagens and chemokines were upregulated. Pretreatment expression of an inflammatory signature correlated with antiproliferative response to anastrozole and this observation was validated in an independent study. Higher expression of immune-related genes such as SLAMF8 and TNF as well as lymphocytic infiltration were associated with poorer response (P < 0.001) and validated in an independent cohort.
Conclusions: The molecular response to aromatase inhibitor treatment varies greatly between patients consistent with the variable clinical benefit from aromatase inhibitor treatment. Higher baseline expression of an inflammatory signature is associated with poor antiproliferative response and should be assessed further as a novel biomarker and potential target for aromatase inhibitor-treated patients. Clin Cancer Res; 19(10); 2775–86. ©2013 AACR.
Translational Relevance
Most postmenopausal women with estrogen receptor (ER)–positive breast cancer receive an aromatase inhibitor at some stage during their treatment. However, responsiveness to aromatase inhibitors varies greatly. Previous studies of resistance mechanisms have been conducted in cell lines and have highlighted the importance of signal transduction pathways, but these studies omit the stromal influences that are increasingly recognized to have a major influence on tumor biology. In contrast, we report genome-wide expression profiling of 81 ER-positive breast carcinomas including stromal tissues before and during treatment with an aromatase inhibitor in the neoadjuvant setting. We identify an inflammatory signature as the strongest correlate of poor antiproliferative response and confirm this finding in an independent set of tumors as well as using gross lymphocytic infiltration as an alternative measure of inflammatory activity. This work provides a potential new avenue for drug development and for identifying patients with a reduced likelihood of response to aromatase inhibition.
Introduction
Approximately 80% of human breast carcinomas present as estrogen receptor α–positive (ER+). In postmenopausal women, estrogen withdrawal by treatment with aromatase inhibitors is the most effective form of endocrine therapy, but response to them varies markedly (1–3). Mechanisms of resistance appear to be multiple but are poorly characterized. In vitro studies of acquired resistance to estrogen deprivation have identified several putative mechanisms which largely involve growth factor–related signal transduction pathways (4–7) but there is limited clinical evidence to support these. It is also notable that these models do not involve an assessment of the contribution of human stromal elements.
The presurgical/neoadjuvant setting provides an exceptionally valuable scenario for linking biology to clinical response and to study mechanisms of resistance; within this setting, the proliferation marker Ki67 is a validated pharmacodynamic indicator of response to endocrine therapy (8–10). Treatment-dependent changes in sequential measurements of Ki67 in neoadjuvant trials of endocrine therapy in postmenopausal women (IMPACT, p024, and Z1031; refs. 9, 11, 12) have all revealed differences or lack of differences that were parallel to recurrence-free survival (RFS) differences in the equivalent adjuvant trials (ATAC, BIG 1-98, and MA27; refs. 1, 13, 14), respectively. In addition, Ki67 of patients after 2 weeks' treatment predicts RFS more closely than pretreatment values (8).
Nearly all patients on an aromatase inhibitor show reduction in Ki67 expression, suggesting that some benefit is derived, although this may be modest (9). This continuous marker of response is well-suited to assessment of mechanisms of resistance that are also likely to have a nonbinary effect. This may explain why changes in Ki67 have proven to be better predictors of benefit from endocrine therapy than clinical response during neoadjuvant endocrine therapy (15, 16).
Gene expression profiling of tumor biopsies before and during treatment has the potential to enable the identification of the most important genes/pathways involved in the response to estrogen deprivation therapy and the pretreatment determinants of response and resistance. Availability of comprehensive expression datasets, as provided here, will allow the evaluation of candidate genes from experimental research for their clinical relevance.
We present, to our knowledge, the largest study of the global transcriptional effects of aromatase inhibitor treatment in the neoadjuvant setting with the aims of identifying: (i) transcriptional features of response to aromatase inhibitors and (ii) phenotypic and genotypic determinants of benefit from aromatase inhibitors. This approach revealed for the first time the importance of immune/inflammatory influences that could not have been detected by studies of cell lines.
Materials and Methods
Patient samples
Fourteen-gauge core-cut tumor biopsies were obtained from 112 postmenopausal women with stage I to IIIB ER+ early breast cancer before and after 2-week anastrozole monotherapy in a neoadjuvant trial (17, 18). Patient demographics are shown in Supplementary Table S1. Tissue was stored in RNAlater at −20ºC. Two 4-μm sections from the core were stained with hematoxylin and eosin (H&E) and examined by a pathologist (A. Nerurkar and P. Osin) to confirm the presence of cancerous tissue, assess histopathology, and the presence or absence of lymphocytic infiltration. Tumors were deemed to be positive for lymphocytic infiltration if intraepithelial mononuclear cells could be seen within tumor cell nests or in direct contact with tumor cells. Biopsies in which lymphocytic infiltrate could be seen without direct contact to tumor cells were not considered to have lymphocytic infiltration. Samples were assessed blindly by both pathologists and the concordance rate was 81%. Discordant samples were then reassessed by A. Nerurkar who was blinded to the result of the initial assessment and the more frequent of the 3 assessments was used in the analysis. Total RNA was extracted using RNeasy (Qiagen). RNA quality was checked using an Agilent Bioanalyser: samples with RNA integrity values of less than 5 were excluded from further analysis. ER (H-score) and Ki67 (% cells positive) values by centralized immunohistochemistry were already available (17).
An additional 71 stored sections from paraffin-embedded, formalin-fixed core-cut biopsies were obtained from a subset of patients from the IMPACT trial who received neoadjuvant anastrozole or tamoxifen (15). Sections were H&E-stained, and the presence or absence of detectable lymphocytic infiltration was assessed by a pathologist (A. Nerurkar). Ki67 data were obtained from previous analysis of these patients (8, 9).
Gene expression analysis and data preprocessing
RNA amplification, labeling, and hybridization on HumanWG-6 v2 Expression BeadChips (Illumina) were conducted according to the manufacturer's instructions at a single Illumina BeadStation facility. Tumor RNA of sufficient quality and quantity was available to generate expression data from 104 pretreatment biopsies and 85 two-week biopsies (Supplementary Fig. S1). Data were extracted using BeadStudio software and normalized with variance-stabilizing transformation (VST) and Robust Spline Normalization (RSN) method in the Lumi package (19). Probes that were not detected in any samples (detection P > 1%) were discarded from further analysis. Gene expression data from this study is deposited (20).
Data analysis
Genes differentially expressed between baseline and 2-week samples were identified using a multivariate permutation test (21) implemented in BRB-Array Tools (22). Random variance t-statistics were calculated for each gene (23). Geneset comparison analysis (24) was conducted using Biocarta pathways (25).
Multiple correlation analysis was conducted in BRB-Array Tools. A statistical significance level for each gene for testing the hypothesis that the Spearman's correlation between gene expression and change in Ki67 was calculated to be 0 and P values were then used in a multivariate permutation test (21) from which false discovery rates (FDR) were computed. Proportional 2-week change in Ki67 was defined as (2-week Ki67/baseline Ki67) × 100. Pathway analysis using Ingenuity Pathways Analysis (Ingenuity Systems) was conducted on the list of genes correlated with P ≤ 0.005. Other statistical analyses were conducted in SPSS for Windows (SPSS Inc.), and Graphpad Prism (Graphpad Software Inc.).
Multivariable analyses were undertaken with log proportional change in Ki67 as the dependent variable and the immune metagene, ESR1, ER H-score, and grade as independent variables. Backward selection was used. Cases with missing values for any of the variables in the model were excluded from analysis.
Normalized data from the validation set (Edinburgh) were downloaded from Gene Expression Omnibus (ref. 26; Accession number = GSE20181). Expression of the inflammatory metagene was derived by extracting data for the component genes from the normalized series matrix file. Ki67 was obtained from a previous study of these samples (27).
Exploratory analysis of the type of immune cell represented by the signature associated with change in Ki67 was carried out in prediction analysis of microarrays (PAM; ref. 28). Relative expression of the correlated genes was determined by taking the square of correlation coefficient of the positively correlated genes and rescaling to the training population of all immune cell types from the Reference Database of Immune Cells (29) from which more than 4 examples were available. Leave-one-out cross-validation was used to determine the accuracy of the classifier and the normalized expression of the genes correlated with change in Ki67 was entered as an “unknown” to determine the likely identity of the unknown profile.
Results
Estrogen deprivation induces profound reductions in proliferation and estrogen-associated genes
Good quality gene expression data were available at baseline and 2 weeks postanastrozole treatment from 81 matched pairs of samples (Supplementary Fig. S1). Using multiple testing corrected class comparison analysis, 1,327 genes were identified that were significantly differentially expressed at an FDR of 1% (Table 1; Supplementary Table S2A). Of these, 926 were downregulated and 401 upregulated. Proliferation-associated genes such as TOP2A, CCNB2 and classical estrogen-dependent genes such as TFF1 and PDZK1 were highly represented among the downregulated genes. Less consistency in function was observed among upregulated genes, however, collagens and stromal components including immune-related molecules were prominent. Pathway analysis using Geneset Comparison (24) revealed that 19 of the top 30 statistically significantly altered pathways were proliferation-associated (Table 2). Pathways related to estrogen signaling, apoptosis, and the complement pathway also changed.
Genes differentially expressed between pretreatment and 2 weeks in 81 pairs of patients treated with anastrozole
Pathways affected by aromatase inhibitor treatment
Tumors exhibit heterogeneity in transcriptional response to estrogen deprivation
To visualize the degree of variability between the 81 tumors in their transcriptional response to estrogen deprivation, we conducted unsupervised hierarchical clustering analysis of the values representing the change (posttreatment − baseline) in expression of each of the 1,327 genes upon aromatase inhibitor treatment (Fig. 1). No single gene was up- or downregulated in all tumors; the most consistently altered gene TOP2A was downregulated in 94% of cases. No gene was upregulated in more than 82% of cases.
Heatmap of changes of expression of genes which alter upon aromatase inhibitor treatment. A, heatmap of changes in 1327 genes differentially expressed at FDR < 1%. B, dendrogram of 81 pairs of tumors color coded with immunohistochemical information corresponding to each sample. Heatmaps of genes comprising the core of the (C) estrogen-associated gene cluster; (D) proliferation-associated gene cluster; (E) ECM cluster; and (F) immune cluster. Red denotes upregulation, green denotes downregulation.
No strong patterns emerged according to this clustering for PgR or HER2 status or for ΔKi67 (Fig. 1B). Most tumors showed strong downregulation of classical estrogen-regulated genes (ERG; Fig. 1C) and proliferation-associated genes (Fig. 1D). However, while a small number of tumors showed consistently poor suppression of proliferation-associated genes, variability in the suppression of ERGs differed markedly between tumors (Fig. 1C). The majority of tumors showed consistent upregulation of the clusters of genes coding for collagens and chemokines (Fig. 1E and F). Although a small number of tumors with poor Ki67 responses also showed lesser increases in the chemokine and collagen clusters, this inverse association was not seen in all tumors. Both supervised segregation of the tumors based on ΔKi67 and HER2 status and unsupervised clustering of specific clusters showed an association of little decrease in proliferation genes for those tumors not showing a decrease in Ki67 but only a subtle difference for HER2-positive tumors (Supplementary Fig. S2).
Relationship between pretreatment expression of ESR1 and change in Ki67
This variation in transcriptional response to aromatase inhibitor treatment is consistent with the variation observed in antiproliferative response as measured by immunohistochemical assessment of Ki67 (Fig. 2A). Levels of ESR1 could be expected to contribute to the responsiveness of tumors to estrogen deprivation and this has previously been shown both by immunohistochemistry and mRNA analyses in tamoxifen-treated patients (30–32). In this study, ESR1 expression showed a relatively weak but statistically significant correlation with the proportional 2-week change in Ki67 (Spearman r = −0.29, P = 0.012; Fig. 2B). One patient showing poor change in Ki67, indicated in Fig. 2A with an asterisk, was excluded from further analyses as this patient's decrease in plasma estradiol was more than 3 SDs less than the mean and hence did not meet criteria for estrogen deprivation.
Ki67 changes in response to aromatase inhibitor treatment. A, profiles of %Ki67 staining before and 2 weeks after aromatase inhibitor treatment as measured by immunohistochemistry. B, relationship between pretreatment expression of ESR1 and change in Ki67.
Identification of genes that are correlated with antiproliferative response to aromatase inhibitors
Quantitative trait analysis (QTA) by Spearman correlation was used to identify 471 genes whose expression in baseline tumor biopsies correlated with proportional 2-week change in Ki67 at P < 0.005 (Table 3; Supplementary Table S2B). The list of genes associated with poor antiproliferative response to aromatase inhibitor treatment was dominated by immune-related genes with SLAMF8, a CD2 family member, the most highly correlated gene, and TNF, interleukins and receptors and other cytokines all among the top 20 genes (Table 3; Fig. 3A). Less consistency of function was observed among the genes predicting for a good response to aromatase inhibitors but estrogen signaling–associated genes such as GATA3 and STC2 both featured within the top 50 correlated genes (Table 3; Fig. 3B). Pathway analysis of the list of 471 correlated probes revealed that inflammatory response–related pathways were significantly overrepresented (Table 3; Supplementary Table S3A). The most significantly overrepresented network included modules focused upon Inflammatory Response, Inflammatory Disease, and Immunological Disease (P = 6.83 × 10−23 to 1.83 × 10−2).
Immune involvement and antiproliferative response to Ki67. A, scatter plot of relationship between pretreatment expression of SLAMF8 and proportional change in Ki67. B, relationship between pretreatment expression of GATA3 and proportional change in Ki67. C, relationship between pretreatment expression of the inflammatory response metagenes and proportional change in Ki67 in our discovery cohort. D, relationship between pretreatment expression of the inflammatory response metagenes and proportional change in Ki67 in the Edinburgh patient cohort. E, plot of proportional change in Ki67 in tumors with detectable lymphocytic infiltration versus those without in our discovery cohort. F, plot of proportional change in Ki67 in tumors with lymphocytic infiltration versus those with no detectable lymphocytic infiltration in the IMPACT series. Dotted lines are shown at 100% (no change in Ki67 after treatment) and 50% (data points above this have a less than 50% decrease in Ki67 and as such are defined as nonresponders; ref. 10).
Genes and pathways associated with antiproliferative response to anastrozole treatment
Validation of correlation of inflammatory response signature with antiproliferative response to aromatase inhibitors
For further analysis, we focused on the 45-gene immune response signature which was most significantly overrepresented in the genes correlated with antiproliferative response (Supplementary Table S3). The median baseline expression of these genes correlated significantly with 2-week proportional change in Ki67 (r = 0.44, P ≤ 0.0001; Fig. 3C). Multivariable analysis including grade, ER H-score, and baseline Ki67 found that the inflammatory response signature was an independent predictor of change in Ki67 (Supplementary Table S4). In an independent set of 58 tumors treated with the aromatase inhibitor letrozole (33) baseline expression of the immune response metagene was significantly correlated with change in Ki67 (r = 0.41, P = 0.0045; Fig. 3D) as in our own data.
Exploration of association between lymphocytic infiltration and antiproliferative response to aromatase inhibitors
One form of inflammatory activity that can be readily assessed in histologic specimens is lymphocytic infiltration. Two pathologists (A. Nerurkar and P. Osin) identified the presence of visible lymphocytic infiltration in 15 and absence in 64 tumors. Consistent with the gene expression observations, tumors with lymphocytic infiltration showed significantly poorer antiproliferative response to aromatase inhibitor treatment (Mann–Whitney, P = 0.011, n = 79; Fig. 3E). The presence of lymphocytic infiltration correlated with expression of the immune response metagene (Supplementary Fig. S3A). In addition, the proportion of samples containing a lymphocytic infiltrate in this cohort increased over the duration of treatment (Supplementary Fig. S3B). We also examined lymphocytic infiltration in a subset of material from the tamoxifen and anastrozole arms of the IMPACT trial (15). Although the relationship was not significant, possibly due to the smaller number assessed, a similar effect size was observed (Mann—Whitney, P = 0.15, n = 52; Fig. 3F). No significant association was observed using the anastrozole- or tamoxifen-only subgroups of this cohort (n = 29 and 32 patients, respectively).
Genes associated with response are correlated with the transcriptional profile of dendritic cells
The association between lymphocytic infiltration and change in Ki67 suggests that the gene expression signature predicting response to aromatase inhibitors may be derived from infiltrating immune cells, rather the tumor cells themselves. To investigate the likely source of the inflammatory signature predictive of change in Ki67, we defined PAM centroids typifying the main immune cell types using profiles from publicly available gene expression data (29). Leave-one-out cross-validation of the centroids identified hematopoietic stem cells, B cells, T cells, natural killer (NK) cells, and dendritic cells with 100% accuracy. Prediction analysis using the normalized relative expression of the genes in our 471-gene response predictor identified the profile as being most closely aligned to that of dendritic cells (Supplementary Table S5).
Discussion
In this study, the largest reported of the global transcriptional consequences of aromatase inhibitor treatment in breast tumors, we have identified an inflammatory gene expression signature in baseline samples that is independently associated with poor antiproliferative response to neoadjuvant aromatase inhibitor treatment (8, 15). The signature we identified appears to contain the transcriptional fingerprint of infiltrating immune cells, a possibility that is supported by our observation that tumors with detectable lymphocytic infiltration appear to obtain lesser benefit from aromatase inhibitors. Importantly, the relationship of the signature with change in Ki67 was validated in an independent set of tumors.
The immune system has conflicting potential roles in both suppressing tumor growth by destroying cancer cells and promoting tumors through the production of cytokines and growth factors (34). Investigation of the association between tumor-associated lymphocytes and response to neoadjuvant chemotherapy has revealed that the presence of lymphocytes is an independent predictor of good response to cytotoxic chemotherapy (35) in patients with breast cancer. Similarly, Mahmoud and colleagues recently showed that tumor-infiltrating CD8+ lymphocytes are indicators of good clinical outcome in patients with ER− breast cancer (36). Our observation that infiltrating immune cells are associated with poor response to endocrine therapy suggests that their role may be significantly different in tumors treated with endocrine therapy as opposed to chemotherapy and between the ER+ and ER− subgroups. The differential cytotoxicities of chemotherapy and endocrine therapy on lymphocytes and the effect of estrogen on the immune system may partially explain these differences with chemotherapeutic agents potentially killing infiltrating immune cells, hence negating their proproliferative effect.
Exploratory analysis of the inflammatory signature suggested that dendritic cells could be involved in the poor response of tumors with high expression of the signature to estrogen deprivation. Dendritic cells have been implicated in promoting breast tumorigenesis by polarizing CD4+ T cells (37) and hence could aid resistance through this manner. In addition, a dendritic cell metagene (38) has been shown to be associated with endocrine resistance in high proliferation, high estrogen-related score tumors (39). However, further work is needed to define the contribution of these cells to aromatase inhibitor response.
The prominence of the immune system as a determinant of resistance to endocrine therapy contrasts with evidence from cell lines of growth factor pathways as the major determinants (4–7). This is readily explained by the cell line work, even if conducted in rodent models, excluding human stromal components. It is also clear, however, that immune influences as characterized by the current work can explain resistance in only a proportion of patients and less statistically prominent pathways may be important determinants of resistance. Larger patient numbers and a greater focus on the pathways activated in individual patients are required to define the proportional influence of the various putative mechanisms. The public availability of our comprehensive molecular characterization of this set of tissues from a carefully monitored clinical trial provides a reference service for other experimentalists to assess the clinical relevance for their findings in relation to estrogen deprivation therapy.
The changes in gene expression were very substantial but variable between tumors as previously reported (40, 41). Some clusters of proliferation, estrogen-regulated, extracellular matrix (ECM), and immune genes were obvious, but within these clusters, major variability between the changes in gene expression existed. While proliferation gene changes correlated with immunohistochemical Ki67 changes, changes in estrogen-regulated genes showed little obvious pattern but this may be due to almost all patients showing suppression of nearly all of these genes and thus creating limited opportunity for segregation of patient groups on this basis. The most consistently downregulated gene was TOP2A, a key target for anthracycline-based therapies. This observation supports the current practice of sequencing aromatase inhibitor treatment after chemotherapy, particularly anthracyclins which directly target Topoisomerase-2. It should be noted, however, that proapoptotic genes are also downregulated, hence the balance of molecular effects in relation to their benefit or detriment for combination with specific chemotherapy is difficult to predict.
A small number of other groups have explored genomic profiles in relation to clinical response to aromatase inhibitors. Consistent with the current report, Miller and colleagues (40) found that the most prominent downregulated genes by aromatase inhibitors, in their case, letrozole, to be those related to proliferation and those that are increased to include many stroma-related genes. However, in that study, the genes most associated with resistance to the aromatase inhibitors were related to cellular biosynthetic processes, in particular those coding for ribosomal proteins (33). This difference from the current set of findings may be due to clinical response rather than Ki67 change being the endpoint for response in the Miller study: the need to get shrinkage of tumor for assignment of response may place an increased emphasis on metabolic processes. A recent study of 377 patients randomized to neoadjuvant treatment with 1 of the 3 third-generation aromatase inhibitors found that both luminal A and B subgroups responded, well although the luminal B group remained at poorer prognosis posttreatment (12). This study did not report on the exploration of other signatures to segregate responsive and nonresponsive subgroups.
The use of Ki67 change as an intermediate endpoint for treatment benefit may be considered a limitation of this study. However, this approach allows the identification of genes/signatures that are specific to response. In contrast, clinical response requires objective regression of tumors and this has dependence on initial growth rate as well the anti-growth effects of therapy. In addition, the continuous nature of Ki67 expression provides greater statistical power than the categorical nature of clinical response analyses. Finally, as pointed out above, Ki67 on treatment is a better predictor for long-term outcome than clinical response to endocrine treatment (8, 15).
In conclusion, as well as immune-related functions having recently described prognostic value and predictive value for chemotherapy in early breast cancer, they are related to reduced antiproliferative response to aromatase inhibitor therapy. Further study is needed to determine whether these relationships are causative and therefore potentially subject to intervention.
Disclosure of Potential Conflicts of Interest
H. Anderson has ownership interest (including patents) in shares in AstraZeneca. M. Dowsett has a commercial research grant, honoraria from speakers' bureau, and is a consultant/advisory board member of AstraZeneca. No potential conflicts of interest were disclosed by the other authors.
Authors' Contributions
Conception and design: A.K. Dunbier, Z. Ghazoui, P. Osin, M. Dowsett
Development of methodology: A.K. Dunbier, Z. Ghazoui, H. Anderson
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.K. Dunbier, Z. Ghazoui, H. Anderson, A. Nerurkar, W.R. Miller, I.E. Smith
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.K. Dunbier, Z. Ghazoui, H. Anderson, R. A'hern, W.R. Miller, M. Dowsett
Writing, review, and/or revision of the manuscript: A.K. Dunbier, Z. Ghazoui, H. Anderson, A. Nerurkar, R. A'hern, W.R. Miller, I.E. Smith, M. Dowsett
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z. Ghazoui, H. Anderson, J. Salter
Study supervision: A.K. Dunbier, M. Dowsett
Grant Support
A.K. Dunbier, H. Anderson, and Z. Ghazoui were supported by the Mary-Jean Mitchell Green Foundation. This work was also supported by a Breakthrough Breast Cancer Research Grant (to M. Dowsett), a Health Research Council of New Zealand Sir Charles Hercus Fellowship (to A.K. Dunbier), and National Health Service funding to the NIHR Biomedical Research Centre.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
- Received March 26, 2012.
- Revision received February 11, 2013.
- Accepted March 1, 2013.
- ©2013 American Association for Cancer Research.
References
Volume 19, Issue 10
