WO2020178399A1 - Breast cancer signature genes - Google Patents

Abstract

The present invention relates to sets of genes, specific gene signatures, useful for predicting long-term prognosis of breast cancer in patients. Furthermore, the invention relates to a method of preparing a personalized treatment schedule for a breast cancer patient based on said long-term prognoses.

Description

TITLE: Breast cancer signature genes

FIELDS OF THE INVENTION

The present invention relates to a method of predicting long-term prognosis of breast cancer in a patient by determining the expression level of a at least six signature genes. The present invention furthermore provides a method of preparing a personalized treatment schedule for a breast cancer patient based on the long-term prognosis obtained by said methods.

BACKGROUND OF THE INVENTION

Breast cancer is the most common cancer that women may face in their lifetime (except for skin cancer), and one in eight women will be diagnosed with breast cancer in their life time (a man’s lifetime risk of breast cancer is one in thousand). Breast cancer is a malignant proliferation of breast cells, e.g. of the epithelial cells lining the ducts or lobules of the breast. There are several types, grades and stages of breast cancers with different characteristics and growth rates. Further, breast cancer can be invasive or non-invasive (in situ). Non-invasive breast cancer has not yet developed the ability to spread either within the breast or to another part of the body, whereas invasive breast cancer, which is the most common type, has the potential to spread, i.e. it is metastatic. Treatment aims to remove the cancer in the breast (local control) and destroy any cancer cell that may have already spread from the breast into the body through the bloodstream or the lymphatic system (systemic control), thus reduce the risk of the cancer to affecting other parts of the body in the future. Local control includes surgery (either breast-conserving surgery or a mastectomy with or without reconstruction) and radiotherapy. Systemic control includes chemotherapy, endocrine therapy and targeted/biological therapy.

Even though the treatment succeeds in eliminating or reducing the primary breast cancer, survival of breast cancer patients is limited by metastases, which can occur many years after the removal of the primary breast tumor. Even after decades of comprehensive cancer research, the possibilities to prevent or treat metastatic cancer is limited.

Furthermore, it is difficult to predict from standard clinical and pathological features if the breast cancer after successful treatment will affect future metastatic cancer in the patient, i.e. metastatic relapse. Accordingly, current treatment practice in most western countries is to offer systemic chemotherapy to the majority of breast cancer patients even though most of these patients would have good outcome (relapse-free survival) even without chemotherapy. Thus, the rate of“over-treated” breast cancer patient is high; it is estimated that 70-80% of breast cancer patients receiving chemotherapy based on traditional predictors would have survived without it.

As of today, methods of classifying breast cancers into subtypes and methods of determining long-term prognosis of a breast cancer patient, i.e. methods of predicting whether a patient treated for primary breast cancer will survive without relapse or will experience future metastatic relapse is known:

EP 2664679 B1 describes an algorithm, named the PAM50 classification model, which is based on the gene expression profile of a defined subset of at least 40 intrinsic genes as superior for classifying breast cancer intrinsic subtypes, and for predicting risk of relapse and/or response to therapy in a subject diagnosed with breast cancer.

US 7081340 B2 discloses a method of predicting the likelihood of long-term survival of a breast cancer patient without the recurrence of breast cancer, following surgical removal of the primary tumor, comprising determining the expression level of Bcl2, STK15, CEGP1, Ki-67, GSTM1, PR, SURV, TFRC, EstRl, CCNB1, BAGI1, and Her2, wherein overexpression of STK15, Ki-67, SURV, TFRC, CCNB1, and Her2, indicates a decreased likelihood of long-term survival without breast cancer recurrence, and the overexpression of Bcl2, CEGP1, GSTM1, PR, EstRl, BAG11, indicates an increased likelihood of long-term survival without breast cancer recurrence.

WO 2015080585 A1 describes a method of assigning treatment to a breast or ovarian cancer patient by determining expression of at least two genes in a cancer sample, and typing said sample as being BRCA-like or not. If the sample is classified as BRCA-like, DNA-damage inducing treatment is assigned to the patient.

Even though, methods involving gene signatures for predicting long-term prognosis of breast cancer patients is known, there is still a need for further signature genes useful in methods of determining long-term prognosis of breast cancer.

SUMMARY OF THE INVENTION

The present inventors have solved this need by identifying set of genes, specific gene signatures, which can be used to predict whether a patient will experience metastatic relapse or not.

The present invention provides in a first aspect an in v/tromethod for predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD 1 in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

According to one embodiment, expression level of one or more of the following genes AMER1, SQSTM1, and PALB2 is in addition measured.

According to one embodiment the expression level of one or more of the following genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA is measured in addition to the genes listed in the first aspect or the embodiment thereof.

According to one embodiment the expression level of the following genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA is measured in addition to the genes listed according to the first aspect.

According to one embodiment the expression level of the following genes AMER1, SQSTM1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA is measured in addition to the genes listed in the first aspect.

The present invention provides in a second aspect a method of predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples;

wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse - free survival of the patient. According to one embodiment of the above methods, the expression level is interpreted as overexpressed if it is above the median expression level in the reference set.

According to one embodiment of the above methods, the expression level is interpreted as overexpressed if it is increased more than about 0.5-4 times compared to the median expression level of the reference set.

In one embodiment, the expression level is interpreted as overexpressed if it is increased more than about 1-3 times, such as about 1, 1.5, 2, 2.5, or 3 times, preferable increased about 1.5 times compared to the median expression level of the reference set.

According to one embodiment of the above methods, the genetic material is mRNA and/or cDNA. According to one embodiment of the above methods, the genetic material is isolated from a tissue sample. According to one embodiment of the above methods, the genetic material is isolated from a fresh, a fresh-frozen or a wax-embedded tissue sample.

According to another embodiment of the above method, the genetic material is isolated from bodily fluid, secretion or derivative thereof such as blood, lymph, urine, saliva, nipple aspirates, or gynecological fluids samples.

According to one embodiment, the genetic material is isolated from circulating tumor cells (CTCs) of the primary breast tumor in bodily fluid, secretion or derivative thereof such as blood, lymph, urine, saliva, nipple aspirates, or gynecological fluids samples.

According to one embodiment of the above methods, the expression level is measured by quantifying RNA transcript level of the genes such as RNA

sequencing, probe-based quantification, microarrays and/or Real-time qRT-PCR.

The present invention provides in a third aspect a method of preparing a personalized treatment schedule for a breast cancer patient based on the long-term prognosis obtained by the methods or any of the embodiments thereof as disclosed above, wherein the patient having gene expression indicative of relapse-free survival is merely treated for the primary tumor and eventually offered mild systemic treatment such as endocrine therapy; and, wherein the patients having gene expression indicative of future metastatic relapse in addition to primary treatment is offered aggressive systemic treatment such as chemo-, immuno- or radio-therapy, individually or combinations of these, and active surveillance.

The present invention provides in a fourth aspect primers and/or probes for detecting mRNA or cDNA of at least the genes NQOl, SERPINE1, SRXN1, TALDOl, TXN and TXNRD1 in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from a patient for use in a method of predicting long-term prognosis of breast cancer in a patient.

In one further embodiment of the above aspect the present invention provides primers and/or probes for detection of mRNA or cDNA of at least the genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 AMER1, SQSTM1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from a patient for use in a method of predicting long-term prognosis of breast cancer in a patient.

In a further embodiment the primer and/or probes according to the fourth aspect and embodiments thereof are used in a method according to the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1 Tumors and cell cultures of the aggressive 66cl4 cells from the 4T1 model display activated NRF2 oxidative stress response. A) Gene expression differences consistent in both cell culture and tumors formed by the aggressive 66cl4 compared to the non-aggressive 67NR. RNA was isolated from culture or primary tumors and subjected to RNA sequencing. The significantly differently expressed genes

(p<0.05) was analyzed by gene enrichment analyses. NRF2 (NFE2L2) controlled genes is elevated in both cells and tumors formed by 66cl4. ChiP-X enrichment analysis of genes (1,270) upregulated in 66cl4 cells and primary tumors compared to 67NR cells and primary tumors (log2(ExpVal_66cl4/ExpVal_67NR > 0.59;

ExpVal_66cl4 > 1; p-value < 0.05). Analysis was performed with Enrichr database. Entries were sorted by combined score. B) Immunoblot analyses of NRF2 and NRF2 controlled proteins in cell lysates of 67NR and 66cl4. Acting (ACTB) was used as loading control.

Figure 2 NRF2 is needed for the formation of primary breast tumors and metastasis to the lungs in immunocompetent mice. A) Stable expression of NRF2 shRNA depletes the NRF2 mRNA in two independent clones of 66cl4 cells. B) Loss of NRF2 protein results in reduced levels of NQOl and HMOX1. Representative immunoblot of NRF2 (100 pg protein loaded), NQOl, HMOX1, FTL1, FTH1 and SQSTM1 (50 pg protein loaded). ACTB was used as a loading control (2 pg protein loaded). C) Loss in NRF2 in 66cl4 interfere with tumor formation in the fat pad of BALB/c mice. Cancer cells (500.000) were injected into the fat pad and tumor growth monitored by caliper. Each line represent one mice. D) The experiment was stopped at day 30 due to the size of the control tumors and tumor weight

determined. E) Tumor cell infiltration into lungs after tail vein injections monitored as lung weight at the end of the experiment.

Figure 3 Mutations in the NRF2 (NFE2L2) pathway is frequent in lung cancer but rare in breast cancer. The indicated breast and lung cancer cohorts were analyzed for genetic alterations in NRF2 or the negative regulators KEAP1 or CUL3 using cBioPortal cancer genomics software (cbioportal.org).

Figure 4 Breast cancer specific NRF2-gene signature predicts relapse-free survival. Analysis of relationships between gene expression and relapse-free survival (RFS) in breast cancer patients using the online tool KM plotter. High and low expression were defined as above and below median. In all breast cancer patients: (A)

Relationship between mean expression of six NRF2-target genes (NQOl,

SERPINE1, TXNRD1, TALDOl, TXN, SRXN1) and RFS. (B) Relationship between mean expression of the six NRF2-target genes (as stated in figure 4A) and three NRF2-regulating genes (SQSTM1, PALB2, AMER1) and RFS. HR, hazard ratio. Figure 5 The breast cancer specific NRF2-gene signature predicts survival in breast cancer patients. Analyses of relationship between the gene expression and the breast cancer specific survival (BCSS, A) or overall survival (OS, B) in the METABRIC cohort (1904 patients). The expression level of each of the transcripts determined using Affymetrix microarray and the expression of each transcripts scored according to the median expression of each transcript in the 6 gene signature (NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1). The p-value for the statistical difference between patients with above (orange) and under (green) median is indicated in the lower part of the plot.

Figure 6 The breast cancer specific NRF2-gene signature with three associated genes predicts survival in breast cancer patients. Analyses of relationship between the gene expression and the breast cancer specific survival (BCSS, A) or overall survival (OS, B) in the METABRIC cohort (1904 patients). The expression level of each of the transcripts determined using Affymetrix microarray and the expression of each transcripts scored according to the median expression of each transcript in the 9 gene signature (NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1, AMER1, SQSTM1, and PALB2). The p-value for the statistical difference between patients with above (orange) and under (green) median is indicated in the lower part of the plot.

Figure 7 The breast cancer specific NRF2-gene associated 7 gene signature predicts survival in breast cancer patients. Analyses of relationship between the gene expression and the breast cancer specific survival (BCSS, A) or overall survival (OS, B) in the METABRIC cohort (1904 patients). The expression level of each of the transcripts determined using Affymetrix microarray and the expression of each transcripts scored according to the median expression of each transcript in the 7 gene signature (CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA). The p-value for the statistical difference between patients with above (orange) and under (green) median is indicated in the lower part of the plot.

Figure 8 Expression levels of the individual transcripts in the NRF2 signature in breast cancer biopsies from different subtypes. Expression level for each of the indicated transcripts were determined using expression array (Affymetrix) in the biopsies of the METABRIC cohort (n= 1904). All transcripts are expressed and clearly detected in biopsies from all subtypes as well as from normal breast tissue.

Figure 9 NRF2 16 transcript signature score varies within and between BC subtypes. Based on median expression for each of the transcripts a signature score was determined for each biopsy.

Figure 10 Genetic alteration or gene expression differences in the gene or transcripts included in the breast cancer derived NRF2 signature. The NRF2 16 transcript signature score varies within and between BC subtypes. Analysis of the Breast Cancer Invasive Carcinoma (TCGA PanCancer Atlas) using cBioPortal cancer genomics software.

Figure 11 The signature of NRF2 related genes in breast cancer: panel A) NQOl alone, panel B) NQOl, SERPINE1, SRXN1, TALDOl, TXN and TXNRD1 and panel C) 16 transcripts showing correlation between gene expression and

progression. The web tool KMPlot.com was used to estimate progression in breast cancer patients separated by gene expression for the NRF2 signature.

Figure 12 The breast cancer derived NRF2 signature is a strong predictor of prognosis compared to the 50 transcripts included in Pam50. (A) median expression for each of the 50 transcripts included in the Pam50 gene or (B) NRF2 signature was used to separate ER positive patients (n=762) into high (lower curve) and low (black curve) and correlated for risk of relapse. Analysis of correlation between gene expression in biopsy and disease progression using the 30 different cohorts included in the KMPlot database from GEO, EGA and TCGA. The Kaplan Meier plots for Pam50 and NRF2 gene expression signatures superimposed for comparison (C).

Figure 13a Table 5: NRF2 16-transcript signature score is a strong and independent predictor of adverse outcome of BC. The statistical calculations were performed on the METABRIC cohort (n=1904 all BC subtypes; n=944 hormone receptor positive and HER2 negative). HR, hazard ratio; SE, standard error; coef., regression coefficient.

Figure 13b NRF2 16 transcript signature correlates with PAM50 ROR score in METABRIC cohort (n=1904). The Pam50 RAR score plotted against the NRF2 signature score for biopsies of the indicated and subtypes as well as normal breast tissue.

Figure 14 The breast cancer derived NRF2 signature according to the present invention (B) improves stratification compared to a NRF2 signature from lung cancer cell lines (A). The KMplot.com tool is used to separate breast cancer patients with estrogen receptor positive tumors according to the NRF2 signature suggested previously (Lu, K. et al. 2017; Singh, A. et al 2008). (B) The NRF2 signature according to the present invention. In the upper panels; the expression levels are split by median; in the lower panels, the expression of split by upper and lower quartile DETAILED DESCRIPTION OF THE INVENTION

Unless specifically defined herein, all technical and scientific terms used have the same meaning as commonly understood by a skilled artisan in the fields of genetics, biochemistry, and molecular biology.

All methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention, with suitable methods and materials being described herein. All publications, patent applications, patents, and other references mentioned herein are incorporated by reference in their entirety. In case of conflict, the present specification, including definitions, will prevail.

Where a numeric limit or range is stated, the endpoints are included. Also, all values and subs range within a numerical limit or range are specifically included as if explicitly written out.

The incidents of breast cancer increase, particularly for low-grade, small tumors. New diagnostic options and increased awareness among women likely explains some if this rise. These early-stage primary tumors are surgically removed, and the patients offered anti-estrogen blocking therapy (Tamoxifen or aromatase inhibitors) and additional chemotherapy. The additional chemotherapy reduces the risk of recurrence in the few patients with aggressive tumors but represent an overtreatment for the majority that likely will not relapse even without such addition treatment. Current diagnostic methods have limitations in predictive risk of recurrence and stratify the patients with early-stage breast cancer to guide the use of additional chemotherapy. This result in over-treatment and reduce quality of life for many breast cancer patients diagnosed with early-stage tumors.

This patent application describes a breast cancer derived, transcript signatures that stratify all breast cancer patients, as well as patients with estrogen receptor positive early-stage tumors, according to risk of relapse.

The gene expression signature in primary tumor biopsy report on the activation of the oxidative stress response coordinated by the transcription factor Nuclear factor erythroid 2 -related factor 2 (NRF2 protein; NFE2L2 gene).

The signature is identified by a combination of mouse model experiments and systems network analyses of patient cohorts and is different from a previously published NRF2 signature based on gene expression in lung cancer cell-lines. Also, the present NRF2 signature have only one transcript (MELK) that overlap with the 50 transcript signature, called Pam50, now introduced in breast cancer

diagnostics(Parker JS et al J Clin Oncol. 2009;27(8): 1160-1167; Nielsen TO et al Clin Cancer Res. 2010;16(21):5222-5232; Chia SK et al Clin Cancer Res.

2012;18(16):4465-4472). Compared to the signature of the 50 transcripts in the Pam50 algorithm, this breast cancer derived NRF2 signature more accurately identifies patients with favorable prognosis, even for those with estrogen receptor positive tumors, see Figure 12c and Table 5.

In addition, the breast cancer derived NRF2 signature clearly outperforms the previous lung cancer cell line based NRF2 signature in predictive strength, see Figure 14a compared to Figure 14b.

A signature of NRF2 related transcripts comprising at least the detection of mRNA or of NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 representing a novel gene expression pattern in tumor biopsies to stratify patients according to risk of relapse and to guide clinical decisions. In an alternative embodiment the NRF2 signature may in addition to detect NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 further comprise the detection of mRNA or cDNA of SQSTM1, AMER1, PALB2.

In another embodiment a signature of NRF2 related transcripts comprising at least detection of mRNA or cDNA of CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA may be used.

In a further embodiment the NRF2 signature comprises the detection of mRNA or cDNA of NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1, SQSTM1, AMER1 , PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA. As described above the present inventors have identified a set of genes, a specific gene signature comprising NRF2 regulated genes, which are expressed in primary breast tumors that give rise to future metastatic cancer. Said gene signatures can be used to predict whether a patient will experience metastatic relapse or not.

Thus, the present inventors have solved the need of providing further gene signature for use in methods for accurate determination of long-term prognosis of breast cancer. The invention is described in further details below.

For breast cancer cell, which are cells of epithelial origin, metastases are the result of a series of acquired changes, including the ability to leave the primary tumor, enter into and survive in circulation, extravasate and establish at a secondary site.

The acquired metastatic phenotype is driven by genetic changes that result in altered gene expression and protein function. Since the major cause of breast cancer related deaths is metastases, acquired changes that facilitate metastasis will result in reduced relapse-free survival.

The nature of the acquired changes in gene expression can be analyzed in publically available data sets and expression of individual genes or set of genes can be correlated with prognosis. However, these analyses are complicated by inter- and intratumor heterogeneity, different cellular composition of the biopsies and the fact that tumor cells interact with non-transformed cells, including immune cells.

Rather than searching for characteristics that discriminate aggressive and non- aggressive primary breast tumors in high number of patients, the inventors have used an established immunocompetent mouse model of breast cancer metastasis (4T1).

The so-called 4T1 model is an immunocompetent animal model for breast cancer; consisting of five different cell lines isolated from the same spontaneous BALB/c mouse tumor. The five different cell lines differ clearly in their metastatic potential, although they behave similarly in culture and all form primary tumors in the mammary fat pad of syngeneic BALB/c mice.

During metastasis, cancer cells experience high levels of oxidative stress and only cells that are able to cope with this challenge can form distant metastasis.

Alterations of the NRF2 pathway, which coordinates the induction of the

transcriptionally encoded oxidative stress response, are frequently seen in squamous-like cancer subtypes. Under normal conditions, NRF2 is ubiquitinated after direct binding to the Kelch-like ECH-associated protein 1 (KEAP1) as a part of the cullin 3 RING ubiquitin ligase complex and rapidly degraded by the proteasome. Upon oxidative stress, NRF2 dissociates from KEAP1, stabilizes and translocates to the nucleus where it controls the expression of numerous genes encoding enzymatic and non-enzymatic antioxidants or proteins that relive the damage caused by the insult.

The pathway is normally induced in response to elevated levels of reactive oxygen radicals and elevated NRF2 can result from metabolic changes in transformed cells even in the absence of genetic alterations directly affecting NRF2. In cancer, somatic mutations in NFE2L2, KEAP1, or CUL3, which activate the NRF2 pathway, have been firmly established. These genetic changes also include deletions of exon two of NFE2L2. Moreover, the level and activity of NRF2 can be increased by several indirect mechanisms, including SQSTM1 mediated degradation of the negative regulator KEAP1. Aside from that, SQSTM1 expression is controlled by NRF2 and its induction can set up a positive feedback loop of the pathway.

The inventors hypothesized that even though somatic mutations in NFE2L2,

KEAP1, and CUL3 are rarely detected in breast cancer, activation of NRF2 signaling can be important also in this cancer type, as it provides growth advantage to the cells and may contribute to the development of an aggressive phenotype. The present inventors used the non-metastatic cell line 67NR and the metastatic cell line 66cl4 of the immunocompetent 4T1 mouse mammary tumor model to study the role of NRF2 signaling in aggressive tumor development. The present inventors report that NRF2 constitutively elevated in the malign 66cl4 cells and results in activation of a selective set of NRF2 controlled genes. Depletion of NRF2 abolished 66cl4’s ability to form primary tumors and metastases in the lungs. Consistent with a role for selective NRF2 controlled genes in human breast cancer metastasis, the expression of several of these genes were elevated in tumor biopsies and increased expression correlated with poor prognosis. Furthermore, a specific subset of NRF2 controlled transcripts could be combined into a breast cancer-specific NRF2-target gene signature that correlates with reduced relapse -free survival (RFS) in human breast cancer patients.

NRF2 signaling pathway is constitutively activated in 66cl4 cells and primary tumors

Transcriptome sequencing was conducted on RNA isolated from cells grown in culture, primary tumors and macroscopically visible lung metastasis of 66cl4- bearing BALB/cJ mice. From the individual sequenced samples, 85%-92% of the paired-end reads were aligned to the mouse genome. Among these, the mapped reads with multiple hits to the reference genome accounted for between 4.7% and 6.7%, indicating high quality of the dataset. By using the Cufflinks software, the expression level of 23,994 genes for each sample was estimated based on gene annotation model of UCSC release mmlO. Principal component analysis (PCA) showed a clear segregation of cell culture versus primary tumor samples and 67NR versus 66cl4 (data not shown)The PCA plot also indicated that 66cl4 lung metastases most closely resembled 66cl4 primary tumors. There was a clear correlation between the relative differences of the cells grown in culture and as primary tumors (data not shown) These data indicate that the two cell lines originating from the same primary tumor harbor stable differences, which are present both in culture and in the tumor.

Thus, the inventors compared the gene expression in cell cultures and primary tumors formed by the metastasis unable cell line 67NR and the metastatic 66cl4. As described above, both cell lines originate from the same spontaneous breast tumor, appear similar in culture and efficiently form primary tumors when injected into the fat pad of BALB/c. The transcriptome analyses showed that the oxidative stress is strikingly elevated in cultured and tumors of the aggressive 66cl4 cell line

(combined score 35, P-value 7.7e ⁹) compared to the non-aggressive 67NR (Figure 1A).

Description of the mouse model experiments are described in more details in the following paragraphs. Of the 23,994 genes identified in the transcriptome sequencing, 1,270 genes showed a higher expression in 66cl4 than in 67NR, both in culture and primary tumors (log2(ExpVal_66cl4/ExpVal_67NR > 0.59;

ExpVal_66cl4 > 1; p-value < 0.05). On the other side, 1,252 genes were higher expressed in 67NR cells and tumors. Enrichment analysis of the genes upregulated in 66cl4 was done using Enrichr (Chen et al., Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC bioinformatics. 2013;14: 12815; Kuleshov et al., Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 2016;44(Wl):W90-7). ChiP-X enrichment analysis (ChEA) suggested that Myb and Nfe212 are the two most activated transcription factors in 66cl4 (Figure la). Moreover, analysis using the KEGG and Reactome databases indicated that cell metabolism was altered in these cells (data not shown)Since NRF2 is known to control transcription of various enzymes in diverse metabolic pathways and hyper-activation of this transcription factor has been linked to tumorigenesis and progression in various cancer types, the present inventors decided to focus on this pathway. To discriminate if activation of the oxidative stress response was due to direct genetic alterations, the exomes of 66cl4 and 67NR were sequenced. DNA isolated from BALB/cJ mouse was used as a control. When compared to the mm9 mouse reference genome (C57BL/6J), the majority of gene variants were common in BALB/cJ, 67NR and 66cl4. Of the remaining variants, 18 were common for the two cell lines, whereas 640 and 834 mutations were exclusively found in 67NR and 66cl4, respectively (data not shown)

No alterations were found in Nfe212 or Keapl, however, 66cl4 harbored a C to G transversion in exon 14 of Cul3 (c.C1768G; p.H590D). CUL3 is part of the KEAP1 - CUL3 E3 ligase complex, which is needed to polyubiquitylate NRF2 and target it for degradation under basal conditions. Sanger sequencing showed that 66cl4 is the only among the five cells lines of the 4T1 model that harbors this mutation (data not shown).

Consistent with a functional role of this mutation, NRF2 protein levels were clearly elevated (Figure lb) even though mRNA levels were in fact reduced in 66cl4 (data not shown)lb). In line with a constitutive NRF2 signaling, NQOl was increased at both mRNA and protein level (Figure lb) in 66cl4. Collectively, these data demonstrate a constitutive activation of the NRF2 pathway in the metastatic 66cl4.

Constitutive NRF2 reduced basal ROS levels, but abrogated the ability to adopt to additional oxidative stress

In order to sustain aberrant proliferation, cancer cells produce high amounts of ATP, which results in increased ROS and imbalanced redox status. In accordance with the activation of NRF2 in 66cl4, the present inventors found that basal ROS levels were reduced but that these cells were also clearly less able to mobilize endogenous antioxidants in response to the oxidative stress inducing agents L- sulforaphane and hydrogen peroxide (data not shown). In line with the reduced ability to adopt to additional oxidative stress, hydrogen peroxide led to lower metabolic activity in 66cl4 (data not shown). Constitutive NRF2 coincides with metabolic reprogramming The transcriptome analysis revealed that a subset of well-known NRF2 target genes was upregulated in 66cl4 (data not shown). The inventors then asked if constitutive NRF2 activation correlated with doubling time in culture, the ability to form colonies in soft agar or growth rate of the primary tumor. Compared to 67NR, the 66cl4 cells multiplied slightly faster in culture and formed larger colonies in soft- agar (data not shown)(. However, there was no significant difference in the growth of the primary tumors formed by the two cell lines. In fact, tumors formed by 67NR displayed a tendency to grow faster than 66cl4 tumors (data not shown).

NRF2 regulates genes involved in glucose and glutamine metabolism and pathway enrichment analysis revealed alterations in cell metabolism in 66cl4 (data not shown). In addition, GO cellular component analysis showed that mitochondrial components were upregulated in 66cl4. Analysis of cellular bioenergetics using a Seahorse XF Analyzer demonstrated that 66cl4 had a higher glycolytic flux. The glycolytic capacity did not differ, and the glycolytic reserve was largest in 67NR. Moreover, the basal mitochondrial respiration and ATP production were similar.

The maximum respiration was highest in 67NR and, accordingly, the spare respiratory capacity was significantly higher in these cells. Together, these results indicate that upregulation of mitochondrial components in 66cl4 is a compensatory mechanism that does not result in increased mitochondrial activity.

NRF2 depletion impairs primary tumor growth and metastasis

Consistently, metastatic 66cl4 cells show elevated protein levels of NRF2 as well as the NRF2 regulated genes NQOl, HMOX1 and FTL1. Furthermore, NRF2 was shown to be important for the tumor forming capacity of the 66cl4 cells. NRF2 were depleted by stable expression shRNA targeting NRF2. In two independent clones, the transcript and protein levels of NRF2 was reduced by more than 90% compared to cells expressing a non-targeting shRNA (Figure 2a and 2b). Depletion of NRF2 coincided with up to 90% reduction in mRNA levels of Nqol, Hmoxl and Gclc (data not shown) and protein levels of NQOl and HMOXL However, mRNA and protein expression of FTH1, FTL1 and SQSTM1 were not changed, even if these genes are known to be regulated by NRF2 in other model systems. As NRF2 independent controls, mRNA expression of Cul3 and Lc3b were unaffected. NRF2 depletion led to a clear increase in basal ROS and loss of NRF2, NQOl, and

HMOX1 induction in response to oxidative stress (data not shown). The inventors also observed reduced growth rate and reduction in the ability to form colonies in soft-agar (data not shown).

These results indicate that the constitutive active NRF2 is different in breast cancer cells compared to other cells or conditions. Importantly, depleting NRF2 coincided with a severely reduced ability to form primary tumors in the fat pad (breast) even if the growth characteristics in culture and soft-agar was grossly unchanged (data not shown). The lack of primary tumors precluded an evaluation of metastasis from the primary tumors. However, tail vein injection of the cells demonstrated that the NRF2 depletion resulted in a complete loss in ability to colonize the lungs.

Together, these results demonstrate constitutively activated NRF2 in the aggressive 66cl4 breast cancer cells and that the protein is needed for these cells to form primary tumors and colonize the lungs. Moreover, NRF2 depleted clones displayed reduced efficiency of glycolysis and significantly increased mitochondrial respiration. Together, these data indicate that inhibition of constitutive NRF2 activation restored energy production via oxidative phosphorylation. Importantly, NRF2 depletion abolished 66cl4’s ability to establish primary tumors in the fat pad and lung metastases after tail vein injection (Figure 2c, d and e).

Nqol may play a role primary and secondary tumor growth

In NRF2 depleted cells, mRNA and protein expression levels of Nqol were consistently downregulated. Previously, it has been found that high NQOl expression correlates with poor prognosis in various cancer types including breast, lung, and cervix (Yang et al., Clinical implications of high NQOl expression in breast cancers. Journal of experimental & clinical cancer research: CR. 2014;33 : 14). Thus, the inventors were interested if NQOl is crucial for the metastatic phenotype of 66cl4. To this end, six clones using two different Nqol -targeted guide RNAs and three non-target controls were generated using a lentiviral CRISPR/Cas9 system to generate cells stably expressing guide RNAs and CAS9. Both the doubling time in culture and the ability to form colonies in soft-agar varied greatly between the six Nqol clones as well as NT clones. After injection into immunocompetent BALB/cJ mice, none of the CAS9 expressing clones formed primary tumors, irrespective of any gene editing. However, all cell lines were able to establish tumors in

immunocompromised, nude mice.

Compared to the NT mix control cells, which formed primary tumors similarly to the 66cl4 wild-type cells, the Nqol-KO clones formed smaller primary tumors and lung metastases after tail vein injection. Even though NQOl expression was similar in the two single cell non-target control clones, they displayed very variable ability to form primary and secondary tumors compared to the NT mix and the parental 66cl4 cells. These data demonstrate that results from genetically manipulated single clones must be carefully considered, as heterogeneity within a cell line cannot be neglected. However, consistent with previous observations, NQOl expression may be important for the formation of primary and secondary tumors also in this breast cancer model.

NRF2 target gene signature in breast cancer

Somatic mutations in NFE2L2, KEAP1, and CUL3, resulting in activation of NRF2 signaling, have been described in various cancer types, particularly in squamous- like cancers. Although these specific genetic alterations are rare in breast cancer, the pathway may also be activated by several indirect mechanisms, including competitive binding of SQSTM1, DPP3, PALB2, or AMER1 to KEAP1 or direct binding of CDKN1A, which all lead to the stabilization of NRF2. Since metastases are the major cause of breast cancer related deaths, the present inventors

hypothesized that if any particular gene controlled by NRF2 should be important for the ability to metastasize, its mRNA expression level should predict poor prognosis in breast cancer patients. The inventors selected 46 established NRF2-target genes (Supplementary Table SI) and analyzed whether high mRNA expression of any of these correlates with poor prognosis in breast cancer patients using KM plotter (Lanczky et al., miRpower: a web-tool to validate survival-associated miRNAs utilizing expression data from 2178 breast cancer patients. Breast cancer research and treatment. 2016; 160(3):439-46; Gyorffy et al., An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients. Breast cancer research and treatment.

2010; 123(3):725-31) and BreastMark (Madden et al., BreastMark: an integrated approach to mining publicly available transcriptomic datasets relating to breast cancer outcome. Breast cancer research: BCR. 2013;15(4):R52).

Supplementary Table SI: Selected NRF2-target genes

These analyses revealed that high mRNA expression of the same 10 out of 46 NRF2 target genes, correlated with reduced RFS in both databases (Supplementary Table S2).

Supplementary Table S2: NRF2-target genes found to be associated with poor prognosis in breast cancer (BC) in both BreastMark and KM plotter. High mRNA expression of 10 out of 46 NRF2-target genes correlated with reduced disease-free/relapse-free survival (DFS/RFS) in at least one intrinsic BC subtype in both online databases BreastMark and KM plotter (Hazard ratio (HZ) > 1.2, p -value < 0.05).

A functional role of NRF2 in aggressive breast cancer is surprising since somatic mutations of NRF2 is infrequent in breast cancer compared to other tumor types (Figure 3). However, it seems that NRF2 controls a discrete set of genes in breast cancer cells compared to other cell types. This conclusion is based on two observations: 1) Depleting the cancer cells for NRF2 caused reduced expression of only some of the“established” NRF2 target genes known from other cellular systems such as lung cancer cells, mouse embryonic fibroblasts and macrophages.

Surprisingly, a number of established NRF2 target genes from other model systems was not down regulated in the breast cancer cells upon depleting of NRF2 in the breast cancer cells (Figure 2b). Transcriptome sequencing of RNA isolated from two independent NRF2 knock down clones demonstrate that a number of

established target genes are not reduced in expression. Moreover, the majority of the transcripts down regulated by NRF2 depletion have not previously been found to be controlled by NRF2 (data not shown). In summary, these experiments suggests that: 1) NRF2 is important also for breast cancer development. 2) NRF2 control other genes in breast cancer cells compared to genes regulated after oxidative stress or somatic mutations in lung cancer cells.

Furthermore, by using the Ualcan cohort database (Ichimura et al., Phosphorylation of p62 activates the Keapl-Nrf2 pathway during selective autophagy, Molecular cell. 2013;51(5):618-31) the inventors found that 17 NRF2-target genes were higher expressed in breast tumor tissue than normal tissue.

Based on these findings, databases for gene expression in human breast cancer biopsies was searched for NRF2 controlled genes. A list of 50 established NRF2 controlled transcripts (from other cell models) was used to identify transcripts elevated in breast cancer biopsies and for witch elevated levels correlate with poor prognosis.

Of the 50 NRF2-target genes, only six correlated with poor prognosis and were upregulated in tumor tissue NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1. Each of these genes alone where less suitable for predicting long-term prognosis (data not shown). However, by combining the six genes into a gene signature, the predictive value increased substantially (Figure 4a).

In line with the assumption that NRF2 controls discrete genes in breast cancer, the inventors further used systems network biology approach to search for transcripts that highly correlates in expression in breast tumor biopsies (and not normal breast tissue from the same patient) with the six established NRF2 transcripts.

Of the top 50 correlated transcripts, another seven transcripts CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA also display a clear correlation between elevated expression level and poor prognosis (Figure 7a and b). Since somatic mutations are rare in breast cancer, the inventors anticipated an alternative NRF2 activation in these tumors due to elevated expression in any of the three established NRF2 activating proteins SQSTM1 (p62), AMER1 and PALB2 that all predict poor prognosis when elevated in breast cancer biopsies.

Further, by including SQSTM1, PALB2, and AMER1 in the above define NRF2- gene signature NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1 generating a signature comprising nine transcripts, the hazard ratio further increased (Figure 4b).

The gene signature according to the invention was identified as described above by a combination of mouse model experiments and systems network analyses of patient cohorts and is different from a previously published NRF2 signature based on gene expression in lung cancer cell-lines (Lu, K. et al., NRF2 induction supporting breast cancer cell survival is enabled by oxidative stress -induced DPP3-KEAP1

interaction, Cancer Res. 2017 Jun 1;77(11):2881 -2892). Furthermore, the breast cancer derived NRF2 signature according to the present invention clearly

outperforms the previous lung cancer cell line based NRF2 signature in predictive strength (Lu, K. et al.).

Of note, the same NRF2 gene expression signature displayed no predictive value in gastric or lung cancer patients.

Together, these results show that the NRF2-gene signature identified here are breast cancer specific and that not all NRF2 target genes are equally important for cancer aggressiveness.

The clinical relevance of said results is highlighted by the observed correlation between high expression of the six NRF2-target genes (NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1) and reduced relapse free survival (RFS) in breast cancer patients.

As described above by also including the transcript levels encoding negative regulators of (SQSTM1, PALB2, and AMER1) further improved the prognostic values of the signatures (Figure 6a and b).

Thus the gene signature according to the present invention may also comprise the detection of mRNA or cDNA of the NRF2 related transcripts NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1, SQSTM1, AMER1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA and represent a novel gene expression pattern in tumour biopsies to stratify patients according to risk of relapse and to guide clinical decisions. It is demonstrated that compiling at least 6 of these transcripts into a signature give a tool that can be used to predict clinical outcome and aid in therapeutic decisions see figure 11B.

In a preferred embodiment the at least six genes are NQOl, SERPINE1, SRXN1, TALDOl, TXN and TXNRD1.

The expression level of each of the NRF2 related transcripts were analyzed with respect to the different subtypes of breast cancer using the METABRIC cohort, see https://www.cbioportalorg/studv/clinicalData?id=brca metabric and Pereira, B. et al., The somatic mutation profiles of 2,433 breast cancers refines their genomic and transcriptomic landscapes, Nature Communication, 10 May 2016, doi:

10.1038/ncommsl l479.

The analyses demonstrated that all transcripts are expressed at clearly detectable levels and vary between patients (Figure 8). The expression of each of the transcripts were compiled into a NRF2 signature score based on the median levels and found to differ within each subtype of breast cancer (Figure 9).

The cBioPortal, the Breast Invasive Carcinoma (TCGA PanCancer Atlas) cohort consisting of 994 patients were used for analysis of genetic or gene expression changes in the NRF2 signature. This analyzes showed that the transcripts are co- altered in a high number of the patients and high mRNA levels are by far the most frequent alteration (Figure 10). Analyzing three other breast cancer cohorts using cBioPortal resulted in the same conclusion, the NRF2 transcripts are frequently elevated at mRNA levels and co-altered in many of the patients.

Having found that the breast cancer derived NRF2 signature of 16 transcripts is frequently co-altered in biopsies, the inventors asked if the signature could possess predictive value.

In total 30 different cohorts with gene expression and clinical data were analyzed by the open access program KMplot.com. Elevated expression of the classical NRF2 controlled gene NQOl by itself correlate with poor prognosis both when analyzing 3951 breast cancer patients of all subtypes and in the 2061 patients with estrogen receptor positive tumors (Figure 11a). By extending from NQOl to the 6 transcripts NQOl, SERPINEl, SRXN1, TALDOl, TXN, TXNRD1 (Figure l ib) or the described 16 transcript (Figure 11c), the hazard ratio and statistical significance increase both for all subtypes and for estrogen receptor positive tumors. From (Figure 1 lb) compared to (Figure 1 lc) it is clear that the above described 6 genes alone may also be used as a signature predicting long-term prognosis of breast cancer in a patient. By comparing the gene transcripts of a 50 transcript signature, called Pam50, now introduced in breast cancer diagnostics(Parker JS et al J Clin Oncol.

2009;27(8): 1160-1167; Nielsen TO et al Clin Cancer Res. 2010;16(21):5222-5232; Chia SK et al Clin Cancer Res. 2012;18(16):4465-4472). The present NRF2 signature have only one overlapping transcript (MELK) with the Pam50.

Pam50 is a signature that aid subgrouping of breast cancer patients into the subgroups named LumA, Lumb, Basal, HER2. When combined with clinical parameters as grade, size and metastasis, the transcripts in the Pam50 algorithm can be combined into a risk of recurrence score (Pam50 ROR).

Compared to the signature of the 50 transcripts in the Pam50 algorithm, the breast cancer derived, 16 transcript NRF2 signature according to the present invention more accurately identifies patients with favorable prognosis, even for those with estrogen receptor (HR) positive tumours.

By comparing these two signatures, the NRF2 signature better separate the patients according to prognosis compare to the Pam50 signature genes (Figure 12a) compared to (Figure 12b). Importantly, the NRF2 signature is better in predicting patients with favorable prognosis. Out of 1000 patients with ER positive tumours,

83 patients with favorable 5-year prognosis will be misclassified using Pam50 transcript signature but not with the NRF2 signature (Figure 12c).

Consistently, using the dataset of another cohort (METABRIC), we found that the

16 trascript NRF2 signature have increased hazard ratios and statistical significance, using cox regression, compared to Pam50 ROR, see Table 5.

We then plotted the Pam50 ROR score against the NRF2 signature score and found a correlation (Figure 13). Consistent with NRF2 as a strong and independent predictor of outcome, combining the Pam50 and NRF2 signature cause significant (p= 1.017e ⁰⁵) improved prediction for patients with hormone receptor positive and HER2 negative tumors in the METABRIC cohort. Together, these data show that the breast cancer derived NRF2 signature is a strong and independent predictor of breast cancer outcome.

A previous NRF2 signature, based on NRF2 dependent gene expression in two lung cancer cell-lines have been found to predict outcome also in breast cancer patients with estrogen receptor positive tumors. Only three of the transcripts in the NRF2 lung cancer signature overlap with genes of the breast cancer derived NRF2 signature NQOl, TXN and TXNR/TXNRD 1.

However, by direct comparisons by meta-analyses of the cohorts in the KMplot tool, the previous lung cancer derived NRF2 signature only shown tendency of separating the patients by clinical development both by separating all the patients in two groups by median expression of each transcript or by comparing the upper and lower quartile (Figure 14a). On the other side, the hazard ratio of the breast cancer derived NRF2 signature according to the present invention increase further by comparing the lower and upper quartile and the statistical significance is still very strong (P= l.le ⁰⁵) (Figure 14b).

Thus, a breast cancer signature comprising transcripts related to NRF2 activity comprising at least detection of NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1, SQSTM1, AMER1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA is a strong and independent predictor of outcome and can be used alone or added to other gene expression signatures and combined with clinical variables to predict prognosis and guide treatment for breast cancer patients.

Accordingly, the present invention relates to a method of predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes of the following gene signatures:

1) NQOl, SERPINE 1 , SRXN 1 , T ALDO 1 , TXN, TXNRD 1 ; or

2) NQOl, SERPINE 1, SRXN1, TALDOl, TXN, TXNRD 1 , AMER1, SQSTM1, PALB2;

3) CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA; or

4) NQOl, SERPINE 1, SRXN1, TALDOl, TXN, TXNRD 1 , AMER1, SQSTM1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA

in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and comparing said determined expression level against an expression level of said genes in a standard comprising samples of both metastatic and non-metastatic primary breast cancer or merely non-metastatic primary breast cancer; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

Alternatively, the present invention relates to a method of predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes of the following gene signatures:

1) NQOl, SERPINE 1, SRXN1, TALDOl, TXN, TXNRD 1 ; or

2) NQOl, SERPINE 1, SRXN1, TALDOl, TXN, TXNRD 1 , AMER1, SQSTM1,

PALB2; 3) CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA; or

4) NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 , AMER1, SQSTM1,

PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, CENPA

in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and comparing said determined expression level against an expression level of said genes in a standard merely comprising metastatic breast cancer; wherein expression of said genes at the same level as the standard is indicative of future metastatic relapse in the patient; and expression of said genes below the standard is indicative of relapse-free survival of the patient.

In a preferred embodiment the comparison is a normalization and the determined expression level is normalized against a reference set of breast cancer samples.

As used herein,“reference set” refers to a reference set of breast cancer samples selected from primary and/or metastatic tumor samples or a set of complete genetic information obained from such samples. The number of samples in such a reference set should be sufficiently high to ensure that different reference sets (as a whole) behave essentially the same way. If this condition is met, the identity of the individual breast cancer samples present in a particular set will have no significant impact on the relative amounts of the genes assayed. Usually, the breast cancer sample reference set consists of at least about 30, preferably at least about 40 different breast cancer sample specimens.

A signature score relates to the relative expression of 6 or more transcripts from the NRF2 signature can determined by any technical platform used to quantify gene expression.

The expression level measured by measuring mRNA level of each transcript in the biopsy is normalized. The nature of the transcripts used for normalization may vary by the technical platform used.

As known to a person skilled in the art the nature of the transcripts used for normalization may vary by the technical platform used and may for example be a set of housekeeping genes which is constitutively expressed and carries out essential cellular functions. A set of housekeeping genes may also be combined with spiked transcripts and/or negative controls.

Also, the normalization will depend on the pre -analytic treatment of the biopsies (fresh, snap frozen, fixated and paraffin embedded).

Alternatively, global normalization or normalization against a geometric mean of the expression level of all genes analyzed may be used, in which expression of each gene in the gene signature is normalized against the geometric mean of a larger population or number of assayed genes. As is appreciated by the skilled person in the art, normalization, particularly for microarray assay platforms, is conventionally performed to adjust for effects arising from variation in the microarray technology, rather than from biological differences between the samples, such as RNA samples, or between the addressable probes. In general, global normalization in microarray provides a solution for adjusting for errors that effect entire arrays by scaling the data so that the average measurement is the same for each array (and each color). Scaling is typically accomplished by computing the average expression level for each array, calculating a scale factor equal to the desired average, divided by the actual average, and multiplying every measurement from the array by that scale factor. The desired average can be arbitrary, or it may be calculated from the average of a group of arrays.

Thus, according to a first aspect of the present invention relates to a method of predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1 in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

In one embodiment of the above method, expression level of one or more of the following genes AMER1, SQSTM1, and PALB2 is in addition measured.

Thus, an alternative gene signature of the present invention is a nine genes signature comprising the nine genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 , AMER1, SQSTM1, and PALB2.

Furthermore, the present inventors have identified a correlated gene signature comprising the seven genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA.

Said correlated seven genes signature was identified by searching in a database covering gene expression from 421 breast cancer biopsies for the transcripts that correlate the best in expression with the above six NRF2-target genes transcripts (NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1 ) in biopsies from breast cancer patients using a published breast cancer cohort. See:

http://www.cbioportal.org/study?id=brca_tcga_pub2015 and;

htps://www.ncbi.iilm.nih.gov/pubmed/26451490. From this search, 48 transcripts were identified as highly correlated. All 48 transcripts were analyzed individually for association between elevated expression and poor prognosis using kmplot.com. Of these, 22 transcripts showed a significant association between elevated level and poor prognosis. Seven of the 22 transcripts with predictive value showed very clear predictive value alone and improved predictive strength when combined (increased hazard ratio and/or reduced p -value) when using kmplot.com (all breast cancer patients and in all subgroups and with no difference in weight). These seven transcripts form the seven genes signature CDCA5, MELK, CCNB2, TTK,

DLGAP5, KIF4A, and CENPA.

Upregulation of these seven genes (CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA) also correlates with poor prognosis in breast cancer patients.

Thus, a second aspect of the present invention relates to a method of predicting long-term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

In a third aspect of the present invention relates to a method of predicting long term prognosis of breast cancer in a patient, comprising determining the expression level of at least the genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, TXNRD1 , AMER1, SQSTM1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

The below table list information of the gene signature of the present invention. The information includes the Accession Nos., link to NCBI sequence information, and SEQ ID. Nos.

Table 1 The genes of the 6 genes signature of the present invention

Table 2 The genes of the 9 genes signature of the present invention

Table 3 The genes of the 7 genes signature of the present invention

Table 4 The genes of the 16 genes signature of the present invention

Growth of some breast cancers are stimulated by the hormone estrogen, which binds to estrogen receptors at the cancer cells. Such cancers are referred to as estrogen receptor positive (ER+) or hormone sensitive breast cancer. Tissue from biopsy or after surgery is routinely being tested for the presence of estrogen receptors. If the primary tumor is estrogen receptor positive (ER+) it is often indicative of a favorable aggressive progression of the cancer disease, however some of these patients still relapse and end up with progressive disease. In one embodiment of the present invention the cancer cells of the primary tumor are estrogen receptor positive (ER+). In another embodiment of the present invention the cancer cells of the primary tumor are estrogen negative (ER-).

In one embodiment of the present application the sample comprising genetic material from cancer cells of a primary breast tumor is a tissue sample of the primary breast tumor of the patient obtained by method well known in the art, e.g. during surgery of the breast or by needle biopsy of the breast. The surgical step of removing the breast cancer sample from the patient is not part of a method according to the present invention.

In another embodiment of the present application the sample comprising genetic material from cancer cells of a primary breast tumor is a bodily fluid, secretion or derivative thereof. Non-limiting examples are blood, lymph, urine, saliva, nipple aspirates, or gynecological fluids. In such fluid or secretion sample the genetic material is preferably isolated from circulating tumor cells (CTCs) of the primary breast tumor.

In a preferred embodiment of the present invention the sample is a tissue biopsy.

An obtained tissue sample can be stored before further analysis is applied by method well-known to the skilled person, e.g. it can be preserved at minus 70 °C or archived by paraffin-embedding and formalin-fixation. It is well known in the art that it is possible to successfully use such fixed-paraffin-embedded tissue samples as a source of RNA for e.g. RT-PCR.

In order to measure the gene expression, the genetic material of either freshly collected sample or from a preserved sample has to be isolated. In a preferred embodiment of the present invention RNA is isolated from the sample.

General methods for mRNA extraction including RNA extraction from paraffin embedded tissues are well known in the art. RNA isolation can be performed using purification kit, buffer set and protease from commercial manufacturers .

The expression level of the genes of the gene signatures of the present invention is determined. As disclosed in e.g. US 7081340 B2, methods of gene expression profiling are well known in the art and includes the following non -limiting examples for the quantification of mRNA expression: northern blotting and in situ hybridization; RNAse protection assays; and reverse transcription polymerase chain reaction (RT-PCR). Alternatively, antibodies may be employed that can recognize specific duplexes, including DNA duplexes, RNA duplexes, and DNA-RNA hybrid duplexes or DNA-protein duplexes. Representative methods for sequencing-based gene expression analysis include Serial Analysis of Gene Expression (SAGE), and gene expression analysis by massively parallel signature sequencing (MPSS).

In one embodiment of the present invention, the expression level is measured by quantifying RNA transcript level of the genes by any technological platform including RNA sequencing, probe-based quantification such as Nanostring, microarrays and/or Real-time qRT-PCR.

Another alternative is to measure the level of the product of the gene expression, i.e. the protein, of a signature gene according to the present invention. Protein quantification can be determined by any method known in the art. Methods to determine protein levels are known to a skilled person and include, but are not limited to, Western blotting or ELISA.

The measured amount of a patient tumor mRNA of the genes of the gene signatures of the present invention is in a preferred embodiment of the present invention normalized against the mRNA amount of said gene signatures in a reference set of breast cancer samples.

A non-limiting example of a reference set is the METABRIC dataset which comprises information from 1094 breast cancer patients,

https://www.cbioportal.org/studv/clinicalData?id=brca metabric.

If the expression level of the gene signature in the patient sample is above the median expression level of said gene signature in a reference set, the risk of future metastatic relapse is increased in the patient.

If the expression level of the gene signature in the patient sample is below the median expression level of said gene signature in a reference set, it indicates relapse free survival. Figures 5-8, and 13 show expression levels of the gene signatures of the present invention in the METABRIC reference dataset.

In one embodiment of the present invention the expression level is interpreted as overexpressed if it is increased more than about 0.5-4 times compared to the median expression level of the reference set, such as increased about 0.5, 1, 1.5, 2, 2.5, 3,

3.5 or 4 times, preferable increased about 1.5 times.

The determination of the long-term prognosis of breast cancer is useful in the planning of the most optimal, personalized treatment schedule for a breast cancer patient.

Thus, another aspect of the present invention relates to a method of preparing a personalized treatment schedule for a breast cancer patient based on the long-term prognosis obtained by a method of the invention as disclosed above, wherein the patient having gene expression indicative of relapse -free survival is merely treated for the primary tumor and eventually offered mild systemic treatment such as endocrine therapy; and, wherein the patients having gene expression indicative of future metastatic relapse in addition to primary treatment is offered aggressive systemic treatment such as chemo-, immuno- or radio-therapy, individually or combinations of these, and active surveillance.

By applying the above described gene signatures and methods patients that probably not will experience metastatic relapse can be identified. Such a diagnostic approach will contribute to reduce the number of patients that develop short-term and long term adverse effects of anti-cancer therapies.

In a further aspect the present invention relates to primers and/or probes for the detection of the transcripts according to the present invention for use in the method according to the claims. Based on the sequence information of the genes according to the invention a person skilled in the art is able to design appropriate primers and/or probes for use in a suitable detection metod.

Definitions

As used herein,“long-term prognosis of breast cancer” refers to a prediction of the likelihood whether metastases from a primary breast tumor in a patient, even if said primary tumor have been treated or removed, will arise in the patient at any time in the remaining lifespan of the patient (metastatic relapse), or if the patient with survive without future relapse (relapse-free survival).

As used herein,“relapse” refers the situation where metastases originating from a primary breast tumor arises, also called disease recurrence. In the present application it is also referred to as“metastatic relapse”. As used herein,“relapse-free survival” refers to the situation where a primary breast cancer does not give rise to any metastatic cancers at any time in the remaining lifespan of the patient.

As used herein, a“breast cancer patient” is a patient that suffers, or is expected to suffer, from breast cancer.

As used herein,“breast cancer” refers to ductal carcinoma in situ, lobular carcinoma in situ, ductal carcinoma, lobular carcinoma, inflammatory carcinoma, paget disease of the nipple, phyllodes tumor and/or angiosarcoma.

The method of the present invention is useful on all subtypes of breast cancer.

As used herein,“primary breast cancer” or“primary breast tumor” refers to the original or main cancer or tumor in the breast and possibly in the armpit lymph nodes from which metastases possibly might be initiated.

As used herein,“genetic material” refers to a gene, a part of a gene, a group of genes, a DNA molecule, a fragment of DNA, a group of DNA molecules. Further, it refers to gene expression products, i.e. RNA such as mRNA.

As used herein,“overexpression” refers to an expression level above the median in a reference set.

As used herein,“non-elevated expression” refers to an expression level below the median in a reference set.

As used herein,“normalizing” refers to a comparison of data that includes identification and removal of systematic variability in the data, i.e. normalization of the data, and normalization to adjust for possible differences in total level of analyte (mRNA) that enters into the analyses in order to increase the statistical power of comparison analyses.

The skilled person will acknowledge that an oligonucleotide primer according to the present invention may be a fragment of DNA or RNA of variable length used herein in order to determine the expression level of the target sequence, e.g. single - stranded DNA or RNA, upon alignment of the oligonucleotide primer to

complementary sequence(s) of the said target sequence to be analyzed. An oligonucleotide primer according to the present invention may furthermore be labeled with a molecular marker in order to enable visualization of the results obtained. Various molecular markers or labels are available. An oligonucleotide primer according to the present invention typically comprises the appropriate number of nucleotides allowing that said primer align with the target sequence to be analyzed. The term“probe” refers to an entity that binds to a target molecule, directly or indirectly, and enables the target to be detected, e.g., by a readout instrument. The probe may be labeled. A probe may be designed to detect cDNA or mRNA. A probe is typically a single-stranded polynucleotide that comprises one or more label which directly or indirectly provides a detectable signal. The label can be covalently attached to the polynucleotide, or the polynucleotide can be configured to bind to the label (e.g., a biotinylated polynucleotide can bind a streptavidin -associated label). The label probe can, for example, hybridize directly to a target nucleic acid, or it can hybridize to a nucleic acid that is in turn hybridized to the target nucleic acid or to one or more other nucleic acids that are hybridized to the nucleic acid. Thus, the label probe can comprise a polynucleotide sequence that is

complementary to a polynucleotide sequence of the target nucleic acid, or it can comprise at least one polynucleotide sequence that is complementary to a polynucleotide sequence in a capture probe, amplifier, or the like.

Having generally described this invention, a further understanding can be obtained by reference to certain specific examples, which are provided herein for purposes of illustration only, and are not intended to be limiting unless otherwise specified.

Examples

Example 1: Cell culture and generation of stable cell lines

67NR and 66cl4 were obtained from Barbara Ann Karmanos Cancer Institute.

168FARN, 4T07 and 4T1 were kindly provided by Dr. Fred Miller (Wayne State University, Detroit, MI). ShRNA-NRF2 knockdowns, CRISPR/Cas9-NQ01 knockouts and respective controls were generated by viral transduction (Sigma Aldrich: TRCN0000054658, SHC216V-1EA, MM0000251257, MM0000251258, CRISPR12V-1EA).

Example 2: Orthotopic mouse tumors and in vivo lung colonization assay

Experiments were approved by the National Animal Research Authorities and carried out according to the European Convention for the Protection of Vertebrates used for Scientific Purposes (FOTS ID 4536 and FOTS 10049). For all experiment, female mice (8-11 weeks old, Janvier Labs, France) were used. For orthotopic tumors, mice were anaesthetized and injected with 5 x 10⁵ viable cells into the fourth mammary fat pad. For the in vivo lung colonization assay 5 x 10⁵ cells were injected in the lateral tail vein. 67NR, 66cl4 and NRF2 depleted cells were injected into BALB/cJRj, whereas CRISPR/Cas9 NQOl knockouts and controls were also injected into Balb/cAnNRj-Foxnlnu.

Example 3: Transcriptome analysis RNA was isolated from three passages of 67NR and 66cl4 cells; four and seven primary tumors of 67NR and 66cl4, respectively and six 66cl4 lung metastases.

Lung metastases were collected two to three weeks after the removal of the primary tumors. All samples were stored in RNAlater (Qiagen, 76104) prior to RNA isolation. Tissue samples were homogenized with 1,4 mm ceramic beads form Precellys and QIAshredder (Qiagen, 79654). RNA from cells and tumors was isolated using RNeasy Plus Mini Kit (Qiagen, 74134). RNA from metastases was isolated using RNeasy Micro Kit (Qiagen, 74004).

RNA seq libraries were prepared using TruSeq Stranded mRNA kit (Illumina, San Diego, CA, USA), normalized, pooled to 22 pM and subjected to clustering (by a cBot Cluster Generation System on a HiSeq2500 high output run mode flowcell (Illumina Inc. San Diego, CA, USA). The sequencing (2X100 cycles paired end reads) were performed on an Illumina HiSeq2500 instrument (Illumina, Inc., San Diego, CA, USA). FASTQ files were created with bcl2fastq 2.18 (Illumina, Inc., San Diego, CA, USA). Everything was done according to manufacturer's

instructions. All sequencing reads of RNA-seq were mapped to the Mouse genome by splice-aware aligner Tophat2 with default settings

(http://ccb.jhu.edu/software/tophat/index.shtml; version 2.0.11). Gene expression level for each sample, measured in fragments per kilobase of mRNA million mapped reads (FPKM), was calculated by Cufflinks v2.1.1 using gene model annotation from UCSC release mmlO. FPKM values were log2 -transformed to make variation similar across orders of magnitude. The differential gene expression analyses were carried out using t-test for two-condition comparison and ANOVA for three-condition comparison followed by post hoc Turkey’s honestly significant difference (HSD) test.

Example 4: Exome sequencing

DNA from 67NR, 66cl4, and blood of BALB/cJ mice was isolated using QIAGEN Blood & Cell Culture DNA Kit (Qiagen, #13323). Exome sequencing libraries were prepared from 1 pg gDNA using SureSelectXT target enrichment system for Illumina paired-end sequencing libraries (Agilent Technologies, Santa Clara, CA, USA). Exon capture was performed from 1000 ng of each sequencing library using the SureSelectXT SureSelect Mouse All Exon Kit (Agilent Technologies, Santa Clara, CA, USA). A 20 pM solution of the sequencing libraries was subjected to cluster generation on a HiSeq2500 rapid ruin mode flowcell by the cBot instrument (Illumina, Inc., San Diego, CA, USA). Paired-end sequencing was performed for 2X100 cycles on a HiSeq2500 instrument (Illumina, Inc. San Diego, CA, USA). Everything was done according to manufacturer's instructions. Base calling was done on the HiSeq instrument by RTA 1.17.21.3. Fastq sequence files were generated using CASAVA 1.8.2 (Illumina, Inc. San Diego, CA, USA). Raw fastq files from three replicates were combined and aligned to the mm 9 genome reference by BWA (http://bio-bwa.sourceforge.net/), version 0.6.2, for each of 66cl4, 67NR, and blood, respectively. Sequence Alignment Map (SAM) files were converted to Binary Alignment Map (BAM) files by Picard

(https://broadinstitute.github.io/picard/), version 1.102 and sorted by Samtools (http://samtools.sourceforge.net/), version 0.1.18. The BAM files were subsequently preprocessed by the GATK pipeline. Single nucleotide variants were called by Mutect (https://www.nature.com/articles/nbt.2514), version 1.1.7, and short insertions and deletions (indels) were called by Strelka

(https://www.ncbi.nlm.nih.gov/pubmed/22581179), version 1.0.14. SNVs and indels were annotated by Annovar (http://annovar.openbioinformatics.org/en/latest/), version 2013May09, and exonic, nonsynonymous mutations were kept for further analysis.

Example 5: Quantitative real-time PCR

Quantitative real-time PCR was performed in parallel using QuantiTect SYBR Green PCR master mix (Qiagen, 204141) and Qiagen QuantiTect Primer Assays. Relative gene expression levels were calculated with the 2^L( -delta delta CT) method. Transcripts were normalized to Actb and Tbp.

Example 6: Immunoblotting

Cells were harvested in urea lysis buffer. Equal amounts of proteins were run on Invitrogen NuPAGE Bis-Tris protein gels, transferred onto nitrocellulose

membranes and probed with antibodies against NRF2 (Cell signaling, 12721),

NQOl (Abeam, ab34173), HMOX1 (Enzo, ADI-OSA-110), FTL1 (ThermoFisher Scientific PA5-27357), FTH1 (Abeam, ab65080), SQSTMl/p62 (Progen, GP62-C) or ACTB (Abeam, ab6276). Proteins of interest were detected with near-infrared fluorescent (IRDye) secondary antibodies (Li-Cor Biosciences, 926-32211, 926- 32411, 926-68070).

Example 7: Determination of long-term prognosis of breast cancer in a patient mRNA is isolated from a sample of a primary breast tumor of a patient, e.g. from a formalin-fixed, paraffin-embedded sample (FFPE), fresh or fresh-frozen tissue sample, by standard methods known to the skilled person, e.g. by FFPET RNA Isolation Kit from Roche (Roche-FFPET-025).

The gene expression level of the six genes NQOl, SERPINE1, SRXN1, TALDOl, TXN, and TXNRD1 is then measured by using standard methods known to the skilled person, e.g. by probe-based quantification. Then, the measured expression level of the gene signature in the sample of the patient is normalized against the expression level of said gene signature in a breast cancer tissue reference set, e.g. the METABRIC dataset.

The amount and quality of the mRNA isolate from the sample is first determined using routine assays like the NanoDrop spectrophotometer. A standardize amount of mRNA from the sample is added a known amount of 3-5 mRNA that is not present in a human sample. These mRNAs serve as internal controls. For normalization of the mRNA isolation from the sample, five to ten transcripts are selected for being the most equally expressed and serves as loading controls. These transcripts are also selected based on an expression level that is in the range of the median level of the transcripts that constitute the respective signature. A large number of transcripts may serve as internal controls for normalizing the dataset and define if the expression of any of the transcripts in the signatures are elevated.

If the measured expression level of the gene signature is above the median level as defined by the reference level of said gene signature, the risk of future metastatic relapse is increased in the patient. If, on the other side, the measures expression level of the gene signature is below said median level, it is indicative of relapse free survival (see e.g. figures 4-7, 11-12, and 14 which shows expression levels of the gene signatures of the present invention in the METABRIC dataset below and above the median level).

Statistics

Statistical analyses were performed in GraphPad Prism 7. Values are expressed as mean ± standard deviation (SD) if not otherwise stated. Statistical analyses were performed by paired 2-tailed Student t test after log-transformation (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

Claims

1. An in vitro method for predicting long-term prognosis of breast cancer ina patient, comprising: - determining the expression level of at least the genes NQOl, SERPINE1,

SRXN1, TALDOl, TXN and TXNRD1 in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and - normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

2. The method according to claim 1, wherein the expression level of one or more of the following genes AMER1, SQSTM1, and PALB2 is measured in addition to the genes listed in claim 1.

3. An in vitro method for predicting long-term prognosis of breast cancer in a patient, comprising:

- determining the expression level of at least the genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A, and CENPA in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from the patient; and

- normalizing said determined expression level against an expression level of said genes in a reference set of breast cancer samples; wherein overexpression of said genes is indicative of future metastatic relapse in the patient; and, wherein non-elevated expression of said genes is indicative of relapse-free survival of the patient.

4. The method according to claim 1 or claim 2, wherein the expression level of one or more of the following genes CDCA5, MELK, CCNB2, TTK,

DLGAP5, KIF4A, and CENPA is measured in addition to the genes listed in claim 1 or claim 2.

5. The method according to claim 1, wherein the expression level of the following genes CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA is measured in addition to the genes listed in claim 1.

6. The method according to claim 1, wherein the expression level of the

following genes AMER1, SQSTM1, PALB2, CDCA5, MELK, CCNB2, TTK, DLGAP5, KIF4A and CENPA is measured in addition to the genes listed in claim 1.

7. The method according to any of the above claims, wherein the expression level is interpreted as overexpressed if it is above the median expression level in the reference set.

8. The method according to claim 6, wherein the expression level is interpreted as overexpressed if it is increased more than about 0.5-4 times compared to the median expression level of the reference set.

9. The method according to claim 7, wherein the expression level is interpreted as overexpressed if it is increased more than about 1 -3 times, such as about 1, 1.5, 2, 2.5, or 3 times, preferable increased about 1.5 times compared to the median expression level of the reference set.

10. The method according to any of the above claims, wherein the genetic

material is mRNA and/or cDNA.

11. The method according to any of the above claims, wherein the genetic

material is isolated from a tissue sample.

12. The method according to claim 10, wherein the tissue sample is a fresh, a fresh-frozen or a wax-embedded tissue sample.

13. The method according to any of the above claims, wherein the expression level is measured by quantifying RNA transcript level of the genes by RNA sequencing, probe-based quantification, microarrays and/or Real-time qRT- PCR.

14. A method of preparing a personalized treatment schedule for a breast cancer patient based on the long-term prognosis obtained by a method according to any of the claims 1-13, wherein the patient having gene expression indicative of relapse-free survival is merely treated for the primary tumor and eventually offered mild systemic treatment such as endocrine therapy; and, wherein the patients having gene expression indicative of future metastatic relapse in addition to primary treatment is offered aggressive systemic treatment such as chemo-, immuno- or radio-therapy, individually or combinations of these, and active surveillance.

15. Primers and/or probes for detecting mRNA or cDNA of at least the genes

NQOl, SERPINE1, SRXN1, TALDOl, TXN and TXNRDl in a sample comprising genetic material from cancer cells of a primary breast tumor obtained from a patient for use in a method of predicting long-term prognosis of breast cancer in a patient.

16. Use of primers and/or probes according to claim 15 in the method according to claim 1.