Prognostic significance of high metabolic activity in breast cancer: PET signature in breast cancer

Biochemical and Biophysical Research Communications 511 (2019) 185e191

Contents lists available at ScienceDirect

Biochemical and Biophysical Research Communications journal homepage: www.elsevier.com/locate/ybbrc

Prognostic significance of high metabolic activity in breast cancer: PET signature in breast cancer Sanghee Kang a, b, 1, Eui Hyun Kim a, c, 1, Jun-Eul Hwang d, Ji-Hyun Shin a, Yun Seong Jeong a, Sun Young Yim a, e, Eun Wook Joo f, Young Gyu Eun g, Dong Jin Lee h, Bo Hwa Sohn a, Sung Hwan Lee a, Bora Lim i, Ju-Seog Lee a, * a

Department of Systems Biology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA Department of Surgery, Korea University College of Medicine, Seoul, Republic of Korea Department of Neurosurgery, Severance Hospital, Brain Tumor Center, Yonsei University College of Medicine, Seoul, Republic of Korea d Department of Hematology-Oncolgy, Chonnam National University Medical School, Gwangju, Republic of Korea e Division of Gastroenterology and Hepatology, Department of Internal Medicine, Korea University College of Medicine, Seoul, Republic of Korea f Department of Gynecology, School of Medicine, Kyung Hee University, Seoul, Republic of Korea g Department of Otolaryngology-Head and Neck Surgery, School of Medicine, Kyung Hee University, Seoul, Republic of Korea h Department of OtolaryngologydHead and Neck Surgery, School of Medicine, Hallym University, Seoul, Republic of Korea i Department of Breast Medical Oncology, The University of Texas MD Anderson Cancer Center, Houston, TX, USA b c

a r t i c l e i n f o

a b s t r a c t

Article history: Received 3 February 2019 Accepted 7 February 2019 Available online 15 February 2019

High metabolic activity, reflected in increased glucose uptake, is one of the hallmarks of many cancers including breast cancer. However, not all cancers avidly take up glucose, suggesting heterogeneity in their metabolic demand. Thus, we aim to generate a genomic signature of glucose hypermetabolism in breast cancer and examine its clinical relevance. To identify genes significantly associated with glucose uptake, gene expression data were analyzed together with the standardized uptake values (SUVmax) of 18 F-fluorodeoxy-glucose on positron emission tomography (PET) for 11 breast cancers. The resulting PET signature was evaluated for prognostic significance in four large independent patient cohorts (n ¼ 5417). Potential upstream regulators accountable for the high glucose uptake were identified by gene network analysis. A PET signature of 242 genes was significantly correlated with SUVmax in breast cancer. In all four cohorts, high PET signature was significantly associated with poorer prognosis. The prognostic value of this PET signature was further supported by Cox regression analyses (hazard ratio 1.7, confidential interval 1.48e2.02; P < 0.001). The PET signature was also strongly correlated with previously established prognostic genomic signatures such as PAM50, Oncotype DX, and NKI. Gene network analyses suggested that MYC and TBX2 were the most significant upstream transcription factors in the breast cancers with high glucose uptake. A PET signature reflecting high glucose uptake is a novel independent prognostic factor in breast cancer. MYC and TBX2 are potential regulators of glucose uptake. © 2019 Elsevier Inc. All rights reserved.

Keywords: Breast cancer 18 F-fluorodeoxy-glucose MYC Positron emission tomography TBX2

1. Introduction One of the hallmarks of cancer is metabolic reprograming, which includes induction of glucose hypermetabolism to meet the tumor's high energy demand, 18F-fluorodeoxy-glucose positron

* Corresponding author. Department of Systems Biology, Division of Cancer Medicine, Unit 950, The University of Texas MD Anderson Cancer Center, 1515 Holcombe Blvd, Houston, TX, 77030, USA. E-mail address: [email protected] (J.-S. Lee). 1 Sanghee Kang and Eui Hyun Kim contributed equally to the work. https://doi.org/10.1016/j.bbrc.2019.02.035 0006-291X/© 2019 Elsevier Inc. All rights reserved.

emission tomography (FDG-PET) is an imaging technique that visualizes uptake and consumption of glucose by cancer cells and thus is widely used in cancer imaging, for applications such as cancer detection, staging, and response assessment [1]. The prognostic value of FDG-PET in cancer has been extensively investigated, leading to a consensus that higher FDG uptake is generally associated with worse prognosis [2,3]. Not all cancers avidly take up FDG, however [4]. FDG-PET may reveal substantial intertumoral heterogeneity associated with any of several factors, such as tumor histopathologic subtype, size, extent of microvasculature, proliferation, hormone receptor status,

186

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

and levels of expression of glucose transporters. In practice, FDGPET is a diagnostic tool that is especially useful for surveillance of metastasis. It is also useful for detecting recurrent tumors after surgery or radiotherapy, which may be obscured by scarring and necrosis. Numerous reports suggest that high level of FDG uptake in primary breast cancers is associated with greater tumor burden, higher tumor grade, increased tumor cell proliferation, poor response to neoadjuvant chemotherapy, shorter recurrence-free survival, and more aggressive tumors in metastatic setting [5,6]. However, the clinical and prognostic significance of high glucose uptake in breast cancer is not known. The purpose of our study was to generate a genomic signature reflecting glucose hypermetabolism and to examine its clinical relevance in breast cancer. We generated a signature based on the maximum standardized uptake value (SUVmax) of FDG on PET in a sample of breast cancers, referred to as the “PET signature,” and evaluated its clinical significance in four large independent cohorts of breast cancer patients. We also investigated the association of the PET signature with various clinical parameters and its correlation with previously established risk prediction models. Finally, we performed gene network analysis to identify genes that regulate glucose metabolism in breast cancers. 2. Materials and methods 2.1. PET genomic signature As training set, we used previously generated genome-wide gene expression data from 11 patients to identify genes whose expression was significantly associated with glucose uptake by breast cancers [4]. Primary breast tumors were imaged with FDGPET 4 weeks prior to surgery. Gene expression data were generated by using the Affymetrix U133A microarray platform (GSE21217). Microarray data were normalized by the RMA method. Genes whose expression was significantly (P < 0.005) correlated with the SUVmax value from FDG-PET imaging were selected for the “PET genomic signature” and subjected to further analysis (242 genes, Supplementary Table 1). 2.2. Gene expression data from breast cancer patients For testing and validating the clinical associations of the PET genomic signature, multiple datasets were used. One dataset was a cohort of breast cancer patients from MD Anderson Cancer Center and the gene expression data from these patients, which were described in earlier studies [7]. Additional gene expression data for two patient cohorts were obtained from the public database Gene Expression Omnibus (GEO); raw gene expression data on the Affymetrix U133A platform from eight previous studies were downloaded from GEO and re-normalized together (POOL1), as were data on the Affymetrix U133 2.0 platform from another seven previous studies (POOL2). Finally, gene expression data from the breast cancer patient cohorts in the METABRIC and The Cancer Genome Atlas (TCGA) databases, which are described in earlier reports [8e10], were downloaded. In total, data on 6512 patients from five large independent cohorts were used for validation of the PET signature. The clinical characteristics of the patients in the four cohorts used for prognostic analysis are summarized in Supplementary Table 2. 2.3. Data analysis The gene expression data were analyzed and the prediction model constructed with the BRB Array Tools software program (http://linus.nci.nih.gov/BRB-ArrayTools.html) [11]. Statistical

analysis was performed using the R language (http://www.rproject.org). To construct the prediction models, we used a Bayesian compound covariate prediction (BCCP) model as described previously to estimate the probability that a particular tumor would have a high SUVmax value [12,13]. Briefly, gene expression data from 11 patients in the training set were combined to form a classifier according to a BCCP. The robustness of the classifier was assessed using a misclassification rate determined by using leave-one-out cross-validation in the training set. The sensitivity and specificity of predicting a high PET subtype in the training set were 1.0 and 1.0, respectively. The BCCP classifier estimated the likelihood that an individual tumor would be of either subtype (high or low) of the PET genomic signature and dichotomized tumors according to Bayesian probability (cutoff of 0.5). Three previously reported prognostic genomic subtypes were used in this study. Intrinsic subtype was assigned to each case using the PAM50 centroid-based classifier as published [14]. An Affymetrix-based approximation of the 70-gene prognostic index (NKI subtype) was calculated as previously described with some modification [15]. A microarray-based approximation of the 21gene recurrence score (OncotypeDX, Genomic Health, Santa Clara) was calculated as previously described [16]. We used a gene network analysis tool built in to the Ingenuity Pathway Analysis (IPA) software to identify potential upstream transcription factors that regulate expression patterns of genes enriched in the high PET subtype. The analysis is based on prior knowledge of expected effects between transcriptional factors and their target genes stored in the Ingenuity Knowledge Base. Briefly, the analysis examined the known targets of each transcription factor in the PET signature and compared their direction of change (i.e., expression in the high PET subtype relative to the low PET subtype) to what would be expected from the literature. If the direction of change was consistent with the literature across the majority of targets, then the transcriptional factor was predicted to be active in the high PET subtype, whereas if the direction of change was mostly inconsistent (anti-correlated) with the literature, then the transcriptional factor was predicted to be inactive in the high PET subtype. If there was no clear pattern, then the factor was judged to have no predictive value either way. Regulation z-score was used to estimate the activation state of the transcription factors. An absolute z-score >2 was considered significant as suggested by IPA. The overlap P values generated by Fisher exact test were used to estimate the statistical significance of overlap between the dataset genes and the genes regulated by a transcription factor. 3. Results 3.1. Gene expression signature associated with glucose uptake in breast cancer Since SUVmax values of FDG-PET imaging accurately reflect glucose uptake by cancer cells, we first selected genes whose expression was significantly correlated with SUVmax by using genomic expression data from 11 treatment-naïve breast cancer tissues. This analysis yielded 242 genes (P < 0.005, Fig. 1A). Of 242 genes, 27 are metabolic genes (Fig. 1B). Interesting, major glucose transporters in breast epithelial cells, such as SLC2A1 (or GLUT1) and SLC2A8 (or GLUT8) [17], were not included in the list of metabolic genes, suggesting that these transporters’ activities might be regulated at a post-transcriptional or post-translational level. However, expression of other transporters, such as SLC2A5 (or GLUT5) and SLC50A1 (or SWEET1), showed positive correlation with SUVmax (Fig. 1C), although they were not selected for the PET

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

187

Fig. 1. Expression of genes associated with glucose uptake in breast cancer (PET genomic signature). (A) Expression of 242 genes whose expression was significantly associated (P < 0.005) with SUVmax on PET imaging in 11 breast cancer patients. (B) Expression patterns of metabolic genes in the PET signature (27 genes). (C) Association of expression of selected glucose transporters with SUVmax.

genomic signature because the correlation of their expression with SUVmax did not meet the cut-off P < 0.005 (P ¼ 0.01 for SLC2A5 and P ¼ 0.06 for SLC50A1). 3.2. Clinical significance of glucose uptake in breast cancer We next sought to test the clinical relevance of the PET genomic signature in breast cancer by applying it to gene expression data from other breast cancer patients. After constructing predictor from trained set (Fig. 2A), we applied it to gene expression data from five different breast cancer patient cohorts (Fig. 2B). In the MD Anderson Cancer Center cohort (n ¼ 806), the distant metastasisefree survival (DMFS) of patients with the high PET genomic signature was significantly shorter than that of patients with the low PET genomic signature (Fig. 2C), suggesting that metabolic activity may dictate clinical outcome of patients with breast cancer after treatment. For further validation of the clinical significance of the PET genomic signature, we used the two GEO datasets, each comprising results from a series of studies normalized together by microarray platform used: POOL1 consisting of data generated by using Affymetrix U133A (n ¼ 1494) and POOL2 consisting of data generated by using the U133 v2 platform (n ¼ 1137). The association of PET genomic signature with prognosis remained significant in these two large independent cohorts (Fig. 2C). We then applied the PET genomic signature to genomic data from the clinically and pathologically well-defined METABRIC cohort of breast cancer patients

(n ¼ 1980) [9]. In parallel with the other three cohorts, patients with a high PET genomic signature in this cohort had a significantly poorer rate of disease-free survival (DFS) (Fig. 2C). We also included gene expression data from TCGA (n ¼ 1095) in this analysis [8], and the proportions of patients classified to the high and low PET genomic signature subgroups were similar to those in the other cohorts (Fig. 2B). Since the follow-up period for patients in the TCGA cohort is very short, prognostic significance was not addressed for this cohort. Taken together, our analysis strongly indicates that high tumor glucose uptake reflected in gene expression pattern is indeed significantly associated with poor prognosis in breast cancer patients. We next carried out subset analysis according to estrogen receptor (ER) and node status to further evaluate the prognostic significance of the PET signature. For convenience of analysis, all genomic and clinical data from Affymetrix platforms (MD Anderson, POOL1, and POOL2) were pooled together. The PET genomic signature accurately identified patients at high risk of recurrence even among the patients with ER-positive and node-negative disease (Fig. 3A), a subset generally held to have a very favorable clinical outcome, in the pooled cohorts. The clinical significance of the PET genomic signature remained the same in the METABRIC cohort as in the pooled cohorts (Fig. 3B), suggesting the heterogeneity of metabolic demand in this patient group and the potential for using the PET genomic signature as a biomarker for identifying patients with a poor prognosis even among a group typically

188

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

Fig. 2. Clinical relevance of PET genomic signature in breast cancer. (A) Schematic overview showing construction of a prediction model based on the PET genomic signature. BCCP, Bayesian compound covariate prediction; LOOCV, leave-one-out cross-validation. (B) Concordance of PET genomic signature in five independent cohorts of breast cancer patients. The data are presented in a matrix format in which each row represents a gene and each column represents a tissue sample. (C) Association of the PET genomic signature with prognosis in four of the five independent cohorts. The high PET genomic signature was associated with poorer survival in all four cohorts. The TCGA cohort was excluded from survival analysis because of relatively short follow-up period.

considered as having a good prognosis. 3.3. Association of PET genomic signature with other prognostic genomic subtypes We next examined the association of the PET genomic signature with previously discovered molecular subtypes or prognostic genomic subtypes. When patients in the Affymetrix pooled cohorts were stratified according to intrinsic subtypes by the PAM50 genomic signature, patients with the basal, Her-2, or luminal B subtype showed poor prognosis, as expected (Fig. 4A). Bayesian probability of PET genomic signature (PET probability) was highest in the basal subtype and second highest in the Her-2 subtype, while it was very low in the normal-like and luminal A subtypes, which typically have good prognosis (Fig. 4B). PET probability was significantly higher in groups at high risk of recurrence in the Oncotype DX classification for breast cancer than in groups with intermediate or low risk of recurrence (Fig. 4B). Likewise, high PET probability was significantly associated with the poor prognostic subtype in the NKI classification (Fig. 4B). The strong association of PET genomic signature with molecular subtype was also

maintained in the METABRIC cohort (Supplementary Fig. 1). To examine the prognostic significance of the PET signature in association with well-known clinical parameters and other risk prediction models, univariate and multivariate Cox proportional hazards regression analyses were performed in the METABRIC cohort (Supplementary Table 3), since clinical data in this cohort are collected prospectively. The high PET signature was associated with poorer DFS (hazard ratio 1.728 with P < 0.001 in univariate analysis; hazard ratio 1.268 with P ¼ 0.01 in multivariate analysis), suggesting that the prognostic significance of the PET signature is independent of other prognostic clinical parameters. 3.4. Gene networks associated with cancer metabolism To gain biological insights into the molecular mechanisms that regulate glucose uptake in breast cancer, we next carried out gene network analysis of the PET signature using IPA. Network analysis identified many potential upstream regulators of genes in the PET signature (Supplementary Table 4). Not surprisingly, MYC, a master transcription regulator of cell metabolism and proliferation [18], was one of the highly activated genes in the PET high subtype

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

189

Fig. 3. Clinical significance of the PET genomic signature is independent of clinical variables. (A) Kaplan-Meier survival plots of patients in the pooled Affymetrix platform breast cancer cohorts (MD Anderson, POOL1, and POOL2). Patients were stratified according to PET signature, estrogen receptor (ER) status, or lymph node invasion status. In the rightmost graphs, patients with ER-positive or node-negative disease are further stratified according to PET signature. High PET signature was associated with shorter distant metastasisefree survival (DMFS). (B) Kaplan-Meier survival plots of patients in the METABRIC breast cancer cohort. Patients with ER-positive or node-negative disease are stratified according to PET signature. High PET signature was associated with shorter disease-free survival (DFS).

(Supplementary Fig. 2), indicating that glucose uptake might be regulated by MYC in breast cancer. Another transcription regulator, TBX2, also was highly activated in the PET high subtype, suggesting that TBX2 might also play an important role in cellular metabolism in breast cancer. 4. Discussion To assess the genome-wide characteristics associated with high glucose uptake in breast cancers, we carried out a systematic analysis, starting with gene expression data from 11 breast cancer patients with available SUVmax data. From these data, we identified a signature of 242 genes in these tumors whose expression was significantly associated with glucose uptake. When applied to gene expression data from clinical cohorts, this “PET genomic signature” showed a significant association with clinical outcome. The clinical relevance of the PET genomic signature is strongly supported by several lines of evidence in our study. First, the prognostic significance of the PET genomic signature was validated in four large clinical cohorts (5417 patients). In all four cohorts, a high PET genomic signature was significantly associated with shorter DMFS (MD Anderson, POOL1, POOL2) or DFS (METABRIC). Second, subset analysis showed that the PET genomic signature had the capacity to identify patients with a high risk of recurrence from among clinically homogeneous patients with high likelihood of favorable clinical outcome (ER-positive and node-negative disease). Third, the PET genomic signature was strongly correlated with established prognostic genomic signatures such as PAM50, Oncotype DX, and NKI. Finally, in Cox proportional hazards regression analysis, the prognostic value of the PET genomic signature was independent of other clinical parameters. Taken together, these

results suggest that that the PET genomic signature we identified can add additional prognostic value to previously established risk prediction models and clinical parameters. Our gene network analysis suggested that MYC and TBX2 may be important transcription regulators that determine key breast cancer features associated with glucose uptake. MYC is a pleiotropic transcription factor that participates in many cellular processes, including cell proliferation, apoptosis, differentiation, metabolism, and genome stability [19]. Cancer cells prefer aerobic glycolysis to oxidative phosphorylation, since it provides energy and intermediates necessary to rapidly proliferating cells (the “Warburg effect”) [20]. Activated MYC promotes aerobic glycolysis, driving the expression of almost all glycolytic enzymes, particularly LDH, hexokinase 2, and enolase 1, as well as the glucose transporters [21]. Although direct targeting of MYC has not been successful in numerous clinical trials, MYC-driven glycolysis is now being targeted in various cancer models with promising results [22]. About 30% of breast cancers harbor MYC amplification, and MYC overexpression is often observed in triple-negative breast cancers, known to be the worst prognostic subtype [23]. Our finding suggesting that MYC is a strong upstream regulator of glucose uptake is supported by a similar observation in another study [4]. Whereas the strong association of MYC with glycolysis addiction has been extensively evaluated, the association of TBX2 with cancer metabolism has been barely investigated. TBX2 is a member of the T-box family of transcription factors, which are critical in embryonic development, including mammary gland development. Recently, TBX2 has emerged as an interesting target in cancer as it is known to play a role in cell cycle regulation and oncogenesis [24]. The expression of TBX2 is increased in breast cancers, and this may be associated with the fact that TBX2 maps to chromosome 17q23, a

190

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

Fig. 4. Association of glucose uptake with molecular subtypes of breast cancer in an Affymetrix pooled cohort. (A) Prognostic differences among breast cancer patients with different intrinsic subtypes (PAM50), Oncotype DX subtypes, and NKI subtypes in an Affymetrix pooled cohort (MD Anderson, POOL1, and POOL2; n ¼ 3437). DMFS, distant metastasisefree survival. (B) Average Bayesian probability of PET signature in each molecular subtype. For each box, the boundary closest to zero indicates the 25th percentile, the color line within the box marks the median, and the boundary farthest from zero indicates the 75th percentile. Whiskers above and below the box indicate the 10th and 90th percentiles. For each molecular subtyping system, high PET signature was associated with poorer prognosis. (For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)

region that is frequently amplified in breast cancers [25]. Jacobs et al. also observed that TBX2 is amplified in a subset of breast cancers and suggested that TBX2 functions as an immortalizing gene that enables the cells to bypass senescence by repressing CDKN2A [26]. It is interesting that our gene network analysis also identified CDKN2A, which induces cell cycle arrest in G1 and G2 and thus acts as a tumor suppressor, as a key upstream regulator inhibited in breast cancers with high PET signature. As TBX2 overexpression is observed in many types of cancers and generally is associated with poor prognosis, the potential for using TBX2 as target for cancer therapy is now being explored, although direct targeting of TBX2 is not currently feasible. Further identification and characterization of target genes transcriptionally regulated by TBX2 may help to elucidate whether TBX2 is correlated with cellular energy metabolism [27]. We used gene expression analysis to generate a PET signature which we propose reflects key features of glucose hypermetabolism in breast cancers. Subgroup analysis revealed an apparent survival difference between groups of patients with a low PET signature and those with a high PET signature, which was also observed in the analysis of large cohorts, suggesting that the PET signature add its own prognostic value to known clinical and prognostic parameters in breast cancer. Gene network analysis suggested that MYC is an essential transcriptional regulator that promotes glycolysis of cancer cells. TBX2 was selected as a new

candidate cancer cell metabolism regulator that has not been investigated yet. Acknowledgement The authors also thank Kathryn L Hale of the Department of Scientific Publications at MD Anderson for editing the manuscript. Transparency document Transparency document related to this article can be found online at https://doi.org/10.1016/j.bbrc.2019.02.035. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi.org/10.1016/j.bbrc.2019.02.035. Funding This study was supported by Cancer Prevention & Research Institute of Texas (RP170307), Congressionally Directed Medical Research Programs (CA160616), National Institutes of Health grants (CA150229), and IRG and SINF from The University of Texas MD Anderson Cancer Center. Additional support was provided by the

S. Kang et al. / Biochemical and Biophysical Research Communications 511 (2019) 185e191

National Institutes of Health through a Cancer Center Support Grant to MD Anderson (CA016672).

[13]

Conflicts of interest [14]

The authors declare no conflict of interests. References

[15]

[1] D. Groheux, A. Cochet, O. Humbert, et al., (1)(8)F-FDG PET/CT for staging and restaging of breast cancer, J. Nucl. Med. : Off. Publ., Soc. Nucl. Med. 57 (Suppl 1) (2016) 17se26s. [2] S. Chen, N.K. Ibrahim, Y. Yan, et al., Risk stratification in patients with advanced-stage breast cancer by pretreatment [(18) F]FDG PET/CT, Cancer 121 (2015) 3965e3974. [3] T. Higuchi, A. Nishimukai, H. Ozawa, et al., Prognostic significance of preoperative 18F-FDG PET/CT for breast cancer subtypes, Breast (Edinburgh, Scotland) 30 (2016) 5e12. [4] N. Palaskas, S.M. Larson, N. Schultz, et al., 18F-fluorodeoxy-glucose positron emission tomography marks MYC-overexpressing human basal-like breast cancers, Cancer Res. 71 (2011) 5164e5174. [5] F. Crippa, E. Seregni, R. Agresti, et al., Association between [18F]fluorodeoxyglucose uptake and postoperative histopathology, hormone receptor status, thymidine labelling index and p53 in primary breast cancer: a preliminary observation, Eur. J. Nucl. Med. 25 (1998) 1429e1434. [6] D.A. Mankoff, L.K. Dunnwald, J.R. Gralow, et al., Blood flow and metabolism in locally advanced breast cancer: relationship to response to therapy, J. Nucl. Med. : Off. Publ., Soc. Nucl. Med. 43 (2002) 500e509. [7] C. Hatzis, V. Pusztai L Fau - Valero, D.J. Valero V Fau - Booser, et al., A genomic predictor of response and survival following taxane-anthracycline, JAMA 305 (2011) 1873e1881. LID - 1810.1001/jama.2011.1593 [doi]. [8] T. Consortium, Comprehensive molecular portraits of human breast tumours, Nature 490 (2012) 61e70. LID - 10.1038/nature11412 [doi]. [9] C. Curtis, S.-F. Shah Sp Fau - Chin, G. Chin Sf Fau - Turashvili, et al., The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel, Nature 486 (2012) 346e352. LID - 310.1038/nature10983 [doi]. [10] J.S. Lee, Exploring cancer genomic data from the cancer genome atlas project, BMB Rep. 49 (2016) 607e611. [11] R. Simon, A. Lam, M.C. Li, et al., Analysis of gene expression data using BRBArrayTools, Canc. Inf. 3 (2007) 11e17. [12] J.S. Lee, A. Chu Is Fau - Mikaelyan, D.F. Mikaelyan A Fau - Calvisi, et al.,

[16]

[17] [18] [19] [20]

[21] [22]

[23]

[24]

[25] [26]

[27]

191

Application of comparative functional genomics to identify best-fit mouse models, Nat. Genet. 36 (2004) 1306e1311. C.Y. Ock, J.E. Hwang, B. Keam, et al., Genomic landscape associated with potential response to anti-CTLA-4 treatment in, Nat. Commun. 8 (2017) 1050. LID - 1010.1038/s41467-41017-01018-41460 [doi]. J.S. Parker, M.C.U. Mullins M Fau - Cheang, S. Cheang Mc Fau - Leung, et al., Supervised risk predictor of breast cancer based on intrinsic subtypes, J. Clin. Oncol. : Off. J. Am. Soc. Clin. Oncol. 27 (2009) 1160e1167. LID - 1110.1200/ JCO.2008.1118.1370 [doi]. M.J. van de Vijver, L.J. He Yd Fau - van't Veer, H. van't Veer, Lj Fau - Dai, et al., A gene-expression signature as a predictor of survival in breast cancer, N. Engl. J. Med. 347 (2002) 1999e2009. T. Iwamoto, G. Lee Js Fau - Bianchini, R.E.Bianchini G. Fau - Hubbard, et al., First generation prognostic gene signatures for breast cancer predict both, Breast Canc. Res. Treat. 130 (2011) 155e164. LID - 110.1007/s10549-1001111706-10549 [doi]. F.Q. Zhao, Biology of glucose transport in the mammary gland, J. Mammary Gland Biol. Neoplasia 19 (2014) 3e17. Z.E. Stine, Z.E. Walton, B.J. Altman, et al., MYC, metabolism, and cancer, Cancer Discov. 5 (2015) 1024e1039. LID - 1010.1158/2159-8290.CD-1015-0507 [doi]. N. Meyer, L.Z. Penn, Reflecting on 25 years with MYC, Nat. Rev. Canc. 8 (2008) 976e990. M.G. Vander Heiden, L.C. Cantley, C.B. Thompson, Understanding the Warburg effect: the metabolic requirements of cell proliferation, Science (New York, N.Y.) 324 (2009) 1029e1033. D.M. Miller, S.D. Thomas, A. Islam, et al., c-Myc and cancer metabolism, Clin. Cancer Res. : Off. J. Am. Assoc. Canc. Res. 18 (2012) 5546e5553. B. Li, M.C. Simon, Molecular Pathways: targeting MYC-induced metabolic reprogramming and oncogenic stress in cancer, Clin. Cancer Res. : Off. J. Am. Assoc. Canc. Res. 19 (2013) 5835e5841. E.S. Knudsen, A.K. McClendon, J. Franco, et al., RB loss contributes to aggressive tumor phenotypes in MYC-driven triple negative breast cancer, Cell Cycle 14 (2015) 109e122. A. Abrahams, M.I. Parker, S. Prince, The T-box transcription factor Tbx2: its role in development and possible implication in cancer, IUBMB Life 62 (2010) 92e102. C.S. Sinclair, C. Adem, A. Naderi, et al., TBX2 is preferentially amplified in BRCA1- and BRCA2-related breast tumors, Cancer Res. 62 (2002) 3587e3591. J.J. Jacobs, P. Keblusek, E. Robanus-Maandag, et al., Senescence bypass screen identifies TBX2, which represses Cdkn2a (p19(ARF)) and is amplified in a subset of human breast cancers, Nat. Genet. 26 (2000) 291e299. J. Lu, X.P. Li, Q. Dong, et al., TBX2 and TBX3: the special value for anticancer drug targets, Biochim. Biophys. Acta 1806 (2010) 268e274.