Title:
METHOD FOR BREAST CANCER RECURRENCE PREDICTION UNDER ENDOCRINE TREATMENT
Kind Code:
A1


Abstract:
The present invention relates to methods, kits and systems for the prognosis of the disease outcome of breast cancer, said method comprising:
    • (a) determining in a tumor sample from said patient the RNA expression levels of at least 2 of the following 9 genes: UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP
    • (b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient; and kits and systems for performing said method.



Inventors:
Dartmann, Mareike (Cologne, DE)
Feder, Inke Sabine (Cologne, DE)
Gehrmann, Mathias (Cologne, DE)
Hennig, Guido (Cologne, DE)
Weber, Karsten (Cologne, DE)
Von, Törne Christian (Cologne, DE)
Kronenwett, Ralf (Cologne, DE)
Petry, Christoph (Cologne, DE)
Application Number:
13/638360
Publication Date:
03/14/2013
Filing Date:
03/29/2011
Assignee:
Sividon Diagnostics GmbH (Cologne, DE)
Primary Class:
Other Classes:
435/6.12, 506/16, 435/6.11
International Classes:
C40B30/04; C12Q1/68; C40B40/06
View Patent Images:



Primary Examiner:
ZHANG, KAIJIANG
Attorney, Agent or Firm:
MYRIAD GENETICS INC. (INTELLECTUAL PROPERTY DEPARTMENT 320 WAKARA WAY SALT LAKE CITY UT 84108)
Claims:
1. A method for predicting an outcome of breast cancer in an estrogen receptor positive and HER2 negative tumor of a breast cancer patient, said method comprising: (a) determining in a tumor sample from said patient the RNA expression levels of at least 2 of the following 9 genes: UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP; and (b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient.

2. The method of claim 1 comprising: determining in a tumor sample from said patient the RNA expression levels of at least 3 of the following 9 genes: UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP.

3. The method of claim 1 comprising: (a) determining in a tumor sample from said patient the RNA expression levels of the following 8 genes: UBE2C, RACGAP1, DHCR7 , STC2, AZGP1, RBBP8, IL6ST, and MGP; and (b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient.

4. The method of claim 1 comprising: (a) determining in a tumor sample from said patient the RNA expression levels of the following 8 genes: UBE2C, BIRC5, DHCR7 , STC2, AZGP1, RBBP8, IL6ST, and MGP; and (b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient.

5. The method according to claim 4 wherein BIRC5 may be replaced by UBE2C or TOP2A or RACGAP1 or AURKA or NEK2 or E2F8 or PCNA or CYBRD1 or DCN or ADRA2A or SQLE or CXCL12 or EPHX2 or ASPH or PRSS16 or EGFR or CCND1 or TRIM29 or DHCR7 or PIP or TFAP2B or WNT5A or APOD or PTPRT with the proviso that after a replacement 8 different genes are selected; and UBE2C may be replaced by BIRC5 or RACGAP1 or TOP2A or AURKA or NEK2 or E2F8 or PCNA or CYBRD1 or ADRA2A or DCN or SQLE or CCND1 or ASPH or CXCL12 or PIP or PRSS16 or EGFR or DHCR7 or EPHX2 or TRIM29 with the proviso that after a replacement 8 different genes are selected; and DHCR7 may be replaced by AURKA, BIRC5, UBE2C or by any other gene that may replace BIRC5 or UBE2C with the proviso that after a replacement 8 different genes are selected; and STC2 may be replaced by INPP4B or IL6ST or SEC14L2 or MAPT or CHPT1 or ABAT or SCUBE2 or ESR1 or RBBP8 or PGR or PTPRT or HSPA2 or PTGER3 with the proviso that after a replacement 8 different genes are selected; and AZGP1 may be replaced by PIP or EPHX2 or PLAT or SEC14L2 or SCUBE2 or PGR with the proviso that after a replacement 8 different genes are selected; and RBBP8 may be replaced by CELSR2 or PGR or STC2 or ABAT or IL6ST with the proviso that after a replacement 8 different genes are selected; and IL6ST may be replaced by INPP4B or STC2 or MAPT or SCUBE2 or ABAT or PGR or SEC14L2 or ESR1 or GJA1 or MGP or EPHX2 or RBBP8 or PTPRT or PLAT with the proviso that after a replacement 8 different genes are selected; and MGP may be replaced by APOD or IL6ST or EGFR with the proviso that after a replacement 8 different genes are selected.

6. The method according to claims 1, wherein said patient has received endocrine therapy or is contemplated to receive endocrine treatment.

7. The method of claim 6, wherein said endocrine therapy comprises tamoxifen or an aromatase inhibitor.

8. The method according to claim 1 wherein a risk of developing breast cancer recurrence or cancer related death is predicted.

9. The method according to claim 1, wherein said expression level is determined as a Messenger-RNA expression level.

10. The method according to claim 8, wherein said expression level is determined by at least one of a PCR based method, a microarray based method, or a hybridization based method.

11. The method of claim 1, wherein said determination of expression levels is in a formalin-fixed paraffin embedded tumor sample or in a fresh-frozen tumor sample.

12. The method of claim 1, wherein the expression level of at least one marker gene is determined as a pattern of expression relative to at least one reference gene or to a computed average expression value.

13. The method of claim 1, wherein said step of mathematically combining comprises a step of applying an algorithm to values representative of expression levels of given genes.

14. The method of claim 13, wherein said algorithm is a linear combination of said values representative of expression levels of given genes.

15. The method of claim 14 wherein a value for a representative value of an expression level of a given gene is multiplied with a coefficient.

16. The method of claim 1, wherein one or more thresholds are determined for said combined score, that discriminate into high and low risk, high, intermediate and low risk, or more risk groups by applying the threshold on the combined score.

17. The method of claim 1, wherein a high combined score is indicative of benefit from cytotoxic chemotherapy.

18. The method of claim 1, wherein information regarding nodal status of the patient is processed in the step of mathematically combining expression level values for the genes to yield a combined score.

19. The method of claim 17, wherein said information regarding nodal status is a numerical value if said nodal status is negative and said information is a different numerical value if said nodal status positive and a different or identical number if said nodal status is unknown.

20. A kit for performing a method of claim 1, said kit comprising a set of oligonucleotides capable of specifically binding sequences or to sequences of fragments of the genes in a combination of genes, wherein said combination comprises at least the two of the 9 genes UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP.

21. A computer program product capable of processing values representative of expression levels of a set of genes, mathematically combining said values to yield a combined score, wherein said combined score is indicative of efficacy from endocrine therapy of said patient, according to the method of claim 1.

22. The method of claim 1, comprising: determining in a tumor sample from said patient the RNA expression levels of at least 6 of the following 9 genes: UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP.

23. The method of claim 1, wherein two or more thresholds are determined for said combined score, that discriminate into high and low risk, high, intermediate and low risk, or more risk groups by applying the threshold on the combined score.

Description:

TECHNICAL FIELD

The present invention relates to methods, kits and systems for the prognosis of the disease outcome of breast cancer. More specific, the present invention relates to the prognosis of breast cancer based on measurements of the expression levels of marker genes in tumor samples of breast cancer patients.

BACKGROUND OF THE INVENTION

Breast cancer is one of the leading causes of cancer death in women in western countries. More specifically breast cancer claims the lives of approximately 40,000 women and is diagnosed in approximately 200,000 women annually in the United States alone. Over the last few decades, adjuvant systemic therapy has led to markedly improved survival in early breast cancer. This clinical experience has led to consensus recommendations offering adjuvant systemic therapy for the vast majority of breast cancer patients (EBCAG). In breast cancer a multitude of treatment options are available which can be applied in addition to the routinely performed surgical removal of the tumor and subsequent radiation of the tumor bed. Three main and conceptually different strategies are endocrine treatment, chemotherapy and treatment with targeted therapies. Prerequisite for treatment with endocrine agents is expression of hormone receptors in the tumor tissue i.e. either estrogen receptor, progesterone receptor or both. Several endocrine agents with different mode of action and differences in disease outcome when tested in large patient cohorts are available. Tamoxifen has been the mainstay of endocrine treatment for the last three decades. Large clinical trials showed that tamoxifen significantly reduced the risk of tumor recurrence. An additional treatment option is based on aromatase inhibitors which belong to a new endocrine drug class. In contrast to tamoxifen which is a competitive inhibitor of estrogen binding aromatase inhibitors block the production of estrogen itself thereby reducing the growth stimulus for estrogen receptor positive tumor cells. Still, some patients experience a relapse despite endocrine treatment and in particular these patients might benefit from additional therapeutic drugs. Chemotherapy with anthracyclines, taxanes and other agents have been shown to be efficient in reducing disease recurrence in estrogen receptor positive as well as estrogen receptor negative patients. The NSABP-20 study compared tamoxifen alone against tamoxifen plus chemotherapy in node negative estrogen receptor positive patients and showed that the combined treatment was more effective than tamoxifen alone. However, the IBCSG IX study comparing tamoxifen alone against tamoxifen plus chemotherapy failed to show any significant benefit for the addition of cytotoxic agents. Recently, a systemically administered antibody directed against the HER2/neu antigen on the surface of tumor cells have been shown to reduce the risk of recurrence several fold in a patients with Her2neu over expressing tumors. Yet, most if not all of the different drug treatments have numerous potential adverse effects which can severely impair patients' quality of life (Shapiro and Recht, 2001; Ganz et al., 2002). This makes it mandatory to select the treatment strategy on the basis of a careful risk assessment for the individual patient to avoid over- as well as under treatment. Since the benefit of chemotherapy is relatively large in HER2/neu positive and tumors characterized by absence of HER2/neu and estrogen receptor expression (basal type), compared to HER2/neu negative and estrogen receptor positive tumors (luminal type), the most challenging treatment decision concerns luminal tumors for which classical clinical factors like grading, tumor size or lymph node involvement do not provide a clear answer to the question whether to use chemotherapy or not. Newer molecular tools like a 21 gene assay, a genomic grade index assay and others have been developed to address this medical need.

Treatment guidelines are usually developed by renowned experts in the field. In Europe the St Gallen guidelines from the year 2009 recommend chemotherapy to patients with HER2 positive breast cancer as well as to patients with HER2 negative and ER negative disease. Uncertainty about the usefulness of chemotherapy arises in patients with HER2 negative and ER positive disease. In order to make a balanced treatment decision for the individual the likelihood of cancer recurrence is used as the most useful criteria. Clinical criteria like lymph node status, tumor grading, tumor size and others are helpful since they provide information about the risk of recurrence. More recently, multigene assays have been shown to provide information superior or additional to the standard clinical risk factors. It is generally recognized, that proliferation markers seem to provide the dominant prognostic information. Prominent examples of those predictors are the Mammaprint test from Agendia, the Relapse Score from Veridex and the Genomic Grade Index, developed at the institute Jules Bordet and licensed to Ipsogen. All of these assays are based on determination of the expression levels of at least 70 genes and all have been developed for RNA not heavily degraded by formalin fixation and paraffin embedding, but isolated from fresh tissue (shipped in RNALater™). Another prominent multigene assay is the Recurrence Score test of Genomic Health Inc. The test determines the expression level of 16 cancer related genes and 5 reference genes after RNA extraction from formalin fixed and paraffin embedded tissue samples.

However, the current tools suffer from a lack of clinical validity and utility in the most important clinical risk group, i.e. those breast cancer patients of intermediate risk of recurrence based on standard clinical parameter. Therefore, better tools are needed to optimize treatment decisions based on patient prognosis. For the clinical utility of avoiding chemotherapy, a test with a high sensitivity and high negative predictive value is needed, in order not to undertreat a patient that eventually develops a distant metastasis after surgery. In regard to the continuing need for materials and methods useful in making clinical decisions on adjuvant therapy, the present invention fulfills the need for advanced methods for the prognosis of breast cancer on the basis of readily accessible clinical and experimental data.

Definitions

Unless defined otherwise, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

The term “cancer” is not limited to any stage, grade, histomorphological feature, aggressivity, or malignancy of an affected tissue or cell aggregation.

The term “predicting an outcome” of a disease, as used herein, is meant to include both a prediction of an outcome of a patient undergoing a given therapy and a prognosis of a patient who is not treated. The term “predicting an outcome” may, in particular, relate to the risk of a patient developing metastasis, local recurrence or death.

The term “prediction”, as used herein, relates to an individual assessment of the malignancy of a tumor, or to the expected survival rate (OAS, overall survival or DFS, disease free survival) of a patient, if the tumor is treated with a given therapy. In contrast thereto, the term “prognosis” relates to an individual assessment of the malignancy of a tumor, or to the expected survival rate (OAS, overall survival or DFS, disease free survival) of a patient, if the tumor remains untreated.

An “outcome” within the meaning of the present invention is a defined condition attained in the course of the disease. This disease outcome may e.g. be a clinical condition such as “recurrence of disease”, “development of metastasis”, “development of nodal metastasis”, development of distant metastasis”, “survival”, “death”, “tumor remission rate”, a disease stage or grade or the like.

A “risk” is understood to be a number related to the probability of a subject or a patient to develop or arrive at a certain disease outcome. The term “risk” in the context of the present invention is not meant to carry any positive or negative connotation with regard to a patient's wellbeing but merely refers to a probability or likelihood of an occurrence or development of a given condition.

The term “clinical data” relates to the entirety of available data and information concerning the health status of a patient including, but not limited to, age, sex, weight, menopausal/hormonal status, etiopathology data, anamnesis data, data obtained by in vitro diagnostic methods such as histopathology, blood or urine tests, data obtained by imaging methods, such as x-ray, computed tomography, MRI, PET, spect, ultrasound, electrophysiological data, genetic analysis, gene expression analysis, biopsy evaluation, intraoperative findings.

The term “node positive”, “diagnosed as node positive”, “node involvement” or “lymph node involvement” means a patient having previously been diagnosed with lymph node metastasis. It shall encompass both draining lymph node, near lymph node, and distant lymph node metastasis. This previous diagnosis itself shall not form part of the inventive method. Rather it is a precondition for selecting patients whose samples may be used for one embodiment of the present invention. This previous diagnosis may have been arrived at by any suitable method known in the art, including, but not limited to lymph node removal and pathological analysis, biopsy analysis, in-vitro analysis of biomarkers indicative for metastasis, imaging methods (e.g. computed tomography, X-ray, magnetic resonance imaging, ultrasound), and intraoperative findings.

In the context of the present invention a “biological sample” is a sample which is derived from or has been in contact with a biological organism. Examples for biological samples are: cells, tissue, body fluids, lavage fluid, smear samples, biopsy specimens, blood, urine, saliva, sputum, plasma, serum, cell culture supernatant, and others.

A “tumor sample” is a biological sample containing tumor cells, whether intact or degraded. The sample may be of any biological tissue or fluid. Such samples include, but are not limited to, sputum, blood, serum, plasma, blood cells (e.g., white cells), tissue, core or fine needle biopsy samples, cell-containing body fluids, urine, peritoneal fluid, and pleural fluid, liquor cerebrospinalis, tear fluid, or cells isolated therefrom. This may also include sections of tissues such as frozen or fixed sections taken for histological purposes or microdissected cells or extracellular parts thereof. A tumor sample to be analyzed can be tissue material from a neoplastic lesion taken by aspiration or punctuation, excision or by any other surgical method leading to biopsy or resected cellular material. Such comprises tumor cells or tumor cell fragments obtained from the patient. The cells may be found in a cell “smear” collected, for example, by a nipple aspiration, ductal lavage, fine needle biopsy or from provoked or spontaneous nipple discharge. In another embodiment, the sample is a body fluid. Such fluids include, for example, blood fluids, serum, plasma, lymph, ascitic fluids, gynecologic fluids, or urine but not limited to these fluids.

A “gene” is a set of segments of nucleic acid that contains the information necessary to produce a functional RNA product. A “gene product” is a biological molecule produced through transcription or expression of a gene, e.g. an mRNA, cDNA or the translated protein.

An “mRNA” is the transcribed product of a gene and shall have the ordinary meaning understood by a person skilled in the art. A “molecule derived from an mRNA” is a molecule which is chemically or enzymatically obtained from an mRNA template, such as cDNA.

The term “expression level” refers to a determined level of gene expression. This may be a determined level of gene expression as an absolute value or compared to a reference gene (e.g. a housekeeping gene), to the average of two or more reference genes, or to a computed average expression value (e.g. in DNA chip analysis) or to another informative gene without the use of a reference sample. The expression level of a gene may be measured directly, e.g. by obtaining a signal wherein the signal strength is correlated to the amount of mRNA transcripts of that gene or it may be obtained indirectly at a protein level, e.g. by immunohistochemistry, CISH, ELISA or RIA methods. The expression level may also be obtained by way of a competitive reaction to a reference sample. An expression value which is determined by measuring some physical parameter in an assay, e.g. fluorescence emission, may be assigned a numerical value which may be used for further processing of information.

A “reference pattern of expression levels”, within the meaning of the invention shall be understood as being any pattern of expression levels that can be used for the comparison to another pattern of expression levels. In a preferred embodiment of the invention, a reference pattern of expression levels is, e.g., an average pattern of expression levels observed in a group of healthy individuals, diseased individuals, or diseased individuals having received a particular type of therapy, serving as a reference group, or individuals with good or bad outcome.

The term “mathematically combining expression levels”, within the meaning of the invention shall be understood as deriving a numeric value from a determined expression level of a gene and applying an algorithm to one or more of such numeric values to obtain a combined numerical value or combined score.

An “algorithm” is a process that performs some sequence of operations to produce information.

A “score” is a numeric value that was derived by mathematically combining expression levels using an algorithm. It may also be derived from expression levels and other information, e.g. clinical data. A score may be related to the outcome of a patient's disease.

A “discriminant function” is a function of a set of variables used to classify an object or event. A discriminant function thus allows classification of a patient, sample or event into a category or a plurality of categories according to data or parameters available from said patient, sample or event. Such classification is a standard instrument of statistical analysis well known to the skilled person. E.g. a patient may be classified as “high risk” or “low risk”, “high probability of metastasis” or “low probability of metastasis”, “in need of treatment” or “not in need of treatment” according to data obtained from said patient, sample or event. Classification is not limited to “high vs. low”, but may be performed into a plurality of categories, grading or the like. Classification shall also be understood in a wider sense as a discriminating score, where e.g. a higher score represents a higher likelihood of distant metastasis, e.g. the (overall) risk of a distant metastasis. Examples for discriminant functions which allow a classification include, but are not limited to functions defined by support vector machines (SVM), k-nearest neighbors (kNN), (naive) Bayes models, linear regression models or piecewise defined functions such as, for example, in subgroup discovery, in decision trees, in logical analysis of data (LAD) and the like. In a wider sense, continuous score values of mathematical methods or algorithms, such as correlation coefficients, projections, support vector machine scores, other similarity-based methods, combinations of these and the like are examples for illustrative purpose.

The term “therapy modality”, “therapy mode”, “regimen” as well as “therapy regimen” refers to a timely sequential or simultaneous administration of anti-tumor, and/or anti vascular, and/or immune stimulating, and/or blood cell proliferative agents, and/or radiation therapy, and/or hyperthermia, and/or hypothermia for cancer therapy. The administration of these can be performed in an adjuvant and/or neoadjuvant mode. The composition of such “protocol” may vary in the dose of the single agent, timeframe of application and frequency of administration within a defined therapy window. Currently various combinations of various drugs and/or physical methods, and various schedules are under investigation.

The term “cytotoxic chemotherapy” refers to various treatment modalities affecting cell proliferation and/or survival. The treatment may include administration of alkylating agents, antimetabolites, anthracyclines, plant alkaloids, topoisomerase inhibitors, and other antitumor agents, including monoclonal antibodies and kinase inhibitors. In particular, the cytotoxic treatment may relate to a taxane treatment. Taxanes are plant alkaloids which block cell division by preventing microtubule function. The prototype taxane is the natural product paclitaxel, originally known as Taxol and first derived from the bark of the Pacific Yew tree. Docetaxel is a semi-synthetic analogue of paclitaxel. Taxanes enhance stability of microtubules, preventing the separation of chromosomes during anaphase.

The term “endocrine treatment” or “hormonal treatment” (sometimes also referred to as “anti-hormonal treatment”) denotes a treatment which targets hormone signaling, e.g. hormone inhibition, hormone receptor inhibition, use of hormone receptor agonists or antagonists, use of scavenger- or orphan receptors, use of hormone derivatives and interference with hormone production. Particular examples are tamoxifene therapy which modulates signaling of the estrogen receptor, or aromatase treatment which interferes with steroid hormone production.

Tamoxifen is an orally active selective estrogen receptor modulator (SERM) that is used in the treatment of breast cancer and is currently the world's largest selling drug for that purpose. Tamoxifen is sold under the trade names Nolvadex, Istubal, and Valodex. However, the drug, even before its patent expiration, was and still is widely referred to by its generic name “tamoxifen.” Tamoxifen and Tamoxifen derivatives competitively bind to estrogen receptors on tumors and other tissue targets, producing a nuclear complex that decreases RNA synthesis and inhibits estrogen effects.

Steroid receptors are intracellular receptors (typically cytoplasmic) that perform signal transduction for steroid hormones. Examples include type I Receptors, in particular sex hormone receptors, e.g. androgen receptor, estrogen receptor, progesterone receptor; Glucocorticoid receptor, mineralocorticoid receptor; and type II Receptors, e.g. vitamin A receptor, vitamin D receptor, retinoid receptor, thyroid hormone receptor.

The term “hybridization-based method”, as used herein, refers to methods imparting a process of combining complementary, single-stranded nucleic acids or nucleotide analogues into a single double stranded molecule. Nucleotides or nucleotide analogues will bind to their complement under normal conditions, so two perfectly complementary strands will bind to each other readily. In bioanalytics, very often labeled, single stranded probes are used in order to find complementary target sequences. If such sequences exist in the sample, the probes will hybridize to said sequences which can then be detected due to the label. Other hybridization based methods comprise microarray and/or biochip methods. Therein, probes are immobilized on a solid phase, which is then exposed to a sample. If complementary nucleic acids exist in the sample, these will hybridize to the probes and can thus be detected. These approaches are also known as “array based methods”. Yet another hybridization based method is PCR, which is described below. When it comes to the determination of expression levels, hybridization based methods may for example be used to determine the amount of mRNA for a given gene.

An oligonucleotide capable of specifically binding sequences a gene or fragments thereof relates to an oligonucleotide which specifically hybridizes to a gene or gene product, such as the gene's mRNA or cDNA or to a fragment thereof. To specifically detect the gene or gene product, it is not necessary to detect the entire gene sequence. A fragment of about 20-150 bases will contain enough sequence specific information to allow specific hybridization.

The term “a PCR based method” as used herein refers to methods comprising a polymerase chain reaction (PCR). This is a method of exponentially amplifying nucleic acids, e.g. DNA by enzymatic replication in vitro. As PCR is an in vitro technique, it can be performed without restrictions on the form of DNA, and it can be extensively modified to perform a wide array of genetic manipulations. When it comes to the determination of expression levels, a PCR based method may for example be used to detect the presence of a given mRNA by (1) reverse transcription of the complete mRNA pool (the so called transcriptome) into cDNA with help of a reverse transcriptase enzyme, and (2) detecting the presence of a given cDNA with help of respective primers. This approach is commonly known as reverse transcriptase PCR (rtPCR).

Moreover, PCR-based methods comprise e.g. real time PCR, and, particularly suited for the analysis of expression levels, kinetic or quantitative PCR (qPCR).

The term “Quantitative PCR” (qPCR)” refers to any type of a PCR method which allows the quantification of the template in a sample. Quantitative real-time PCR comprise different techniques of performance or product detection as for example the TaqMan technique or the LightCycler technique. The TaqMan technique, for examples, uses a dual-labelled fluorogenic probe. The TaqMan real-time PCR measures accumulation of a product via the fluorophore during the exponential stages of the PCR, rather than at the end point as in conventional PCR. The exponential increase of the product is used to determine the threshold cycle, CT, i.e. the number of PCR cycles at which a significant exponential increase in fluorescence is detected, and which is directly correlated with the number of copies of DNA template present in the reaction. The set up of the reaction is very similar to a conventional PCR, but is carried out in a real-time thermal cycler that allows measurement of fluorescent molecules in the PCR tubes. Different from regular PCR, in TaqMan real-time PCR a probe is added to the reaction, i.e., a single-stranded oligonucleotide complementary to a segment of 20-60 nucleotides within the DNA template and located between the two primers. A fluorescent reporter or fluorophore (e.g., 6-carboxyfluorescein, acronym: FAM, or tetrachlorofluorescin, acronym: TET) and quencher (e.g., tetramethylrhodamine, acronym: TAMRA, of dihydrocyclopyrroloindole tripeptide ‘black hole quencher’, acronym: BHQ) are covalently attached to the 5′ and 3′ ends of the probe, respectively[2]. The close proximity between fluorophore and quencher attached to the probe inhibits fluorescence from the fluorophore. During PCR, as DNA synthesis commences, the 5′ to 3′ exonuclease activity of the Taq polymerase degrades that proportion of the probe that has annealed to the template. Degradation of the probe releases the fluorophore from it and breaks the close proximity to the quencher, thus relieving the quenching effect and allowing fluorescence of the fluorophore. Hence, fluorescence detected in the real-time PCR thermal cycler is directly proportional to the fluorophore released and the amount of DNA template present in the PCR.

By “array” or “matrix” an arrangement of addressable locations or “addresses” on a device is meant. The locations can be arranged in two dimensional arrays, three dimensional arrays, or other matrix formats. The number of locations can range from several to at least hundreds of thousands. Most importantly, each location represents a totally independent reaction site. Arrays include but are not limited to nucleic acid arrays, protein arrays and antibody arrays. A “nucleic acid array” refers to an array containing nucleic acid probes, such as oligonucleotides, nucleotide analogues, polynucleotides, polymers of nucleotide analogues, morpholinos or larger portions of genes. The nucleic acid and/or analogue on the array is preferably single stranded. Arrays wherein the probes are oligonucleotides are referred to as “oligonucleotide arrays” or “oligonucleotide chips.” A “microarray,” herein also refers to a “biochip” or “biological chip”, an array of regions having a density of discrete regions of at least about 100/cm2, and preferably at least about 1000/cm2.

“Primer pairs” and “probes”, within the meaning of the invention, shall have the ordinary meaning of this term which is well known to the person skilled in the art of molecular biology. In a preferred embodiment of the invention “primer pairs” and “probes”, shall be understood as being polynucleotide molecules having a sequence identical, complementary, homologous, or homologous to the complement of regions of a target polynucleotide which is to be detected or quantified. In yet another embodiment, nucleotide analogues are also comprised for usage as primers and/or probes. Probe technologies used for kinetic or real time PCR applications could be e.g. TaqMan® systems obtainable at Applied Biosystems, extension probes such as Scorpion® Primers, Dual Hybridisation Probes, Amplifluor® obtainable at Chemicon International, Inc, or Minor Groove Binders.

“Individually labeled probes”, within the meaning of the invention, shall be understood as being molecular probes comprising a polynucleotide, oligonucleotide or nucleotide analogue and a label, helpful in the detection or quantification of the probe. Preferred labels are fluorescent molecules, luminescent molecules, radioactive molecules, enzymatic molecules and/or quenching molecules.

“Arrayed probes”, within the meaning of the invention, shall be understood as being a collection of immobilized probes, preferably in an orderly arrangement. In a preferred embodiment of the invention, the individual “arrayed probes” can be identified by their respective position on the solid support, e.g., on a “chip”.

When used in reference to a single-stranded nucleic acid sequence, the term “substantially homologous” refers to any probe that can hybridize (i.e., it is the complement of) the single-stranded nucleic acid sequence under conditions of low stringency as described above.

SUMMARY OF THE INVENTION

In general terms, the present invention provides a method to assess the risk of recurrence of a node negative or positive, estrogen receptor positive and HER2/NEU negative breast cancer patient, in particular patients receiving endocrine therapy, for example when treated with tamoxifen. Estrogen receptor status is generally determined using immunohistochemistry, HER2/NEU (ERBB2) status is generally determined using immunohistochemistry and fluorescence in situ hybridization. However, estrogen receptor status and HER2/NEU (ERBB2) status may, for the purposes of the invention, be determined by any suitable method, e.g. immunohistochemistry, fluorescence in situ hybridization (FISH), or RNA expression analysis.

The present invention relates to a method for predicting an outcome of breast cancer in an estrogen receptor positive and HER2 negative tumor of a breast cancer patient, said method comprising:

(a) determining in a tumor sample from said patient the RNA expression levels of at least 2 of the following 9 genes: UBE2C, BIRC5, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP

(b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient. In one embodiment at least 3, 4, 5 or 6 genes are selected.

In a further embodiment of the invention the method comprises:

(a) determining in a tumor sample from said patient the RNA expression levels of the following 8 genes: UBE2C, RACGAP1, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP

(b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient.

In a further embodiment the method of the invention comprises:

(a) determining in a tumor sample from said patient the RNA expression levels of the following 8 genes: UBE2C, BIRC5, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP;

(b) mathematically combining expression level values for the genes of the said set which values were determined in the tumor sample to yield a combined score, wherein said combined score is indicative of a prognosis of said patient.

In yet another embodiment of the invention

BIRC5 may be replaced by UBE2C or TOP2A or RACGAP1 or AURKA or NEK2 or E2F8 or PCNA or CYBRD1 or DCN or ADRA2A or SQLE or CXCL12 or EPHX2 or ASPH or PRSS16 or EGFR or CCND1 or TRIM29 or DHCR7 or PIP or TFAP2B or WNT5A or APOD or PTPRT with the proviso that after a replacement 8 different genes are selected; and

UBE2C may be replaced by BIRC5 or RACGAP1 or TOP2A or AURKA or NEK2 or E2F8 or PCNA or CYBRD1 or ADRA2A or DCN or SQLE or CCND1 or ASPH or CXCL12 or PIP or PRSS16 or EGFR or DHCR7 or EPHX2 or TRIM29 with the proviso that after a replacement 8 different genes are selected; and

DHCR7 may be replaced by AURKA, BIRC5, UBE2C or by any other gene that may replace BIRC5 or UBE2C with the proviso that after a replacement 8 different genes are selected; and

STC2 may be replaced by INPP4B or IL6ST or SEC14L2 or MAPT or CHPT1 or ABAT or SCUBE2 or ESR1 or RBBP8 or PGR or PTPRT or HSPA2 or PTGER3 with the proviso that after a replacement 8 different genes are selected; and

AZGP1 may be replaced by PIP or EPHX2 or PLAT or SEC14L2 or SCUBE2 or PGR with the proviso that after a replacement 8 different genes are selected; and

RBBP8 may be replaced by CELSR2 or PGR or STC2 or ABAT or IL6ST with the proviso that after a replacement 8 different genes are selected; and

IL6ST may be replaced by INPP4B or STC2 or MAPT or SCUBE2 or ABAT or PGR or SEC14L2 or ESR1 or GJA1 or MGP or EPHX2 or RBBP8 or PTPRT or PLAT with the proviso that after a replacement 8 different genes are selected; and

MGP may be replaced by APOD or IL6ST or EGFR with the proviso that after a replacement 8 different genes are selected.

According to an aspect of the invention there is provided a method as described above, wherein said combined score is indicative of benefit from cytotoxic chemotherapy.

Using the method of the invention before a patient receives endocrine therapy allows a prediction of the efficacy of endocrine therapy.

Table 2 below shows whether the overexpression of each of the above marker genes is indicative of a good outcome or a bad outcome in a patient receiving endocrine therapy. The skilled person can thus construct a mathematical combination i.e. an algorithm taking into account the effect of a given genes. For example a summation or weighted summation of genes whose overexpression is indicative of a good outcome results in an algorithm wherein a high risk score is indicative of a good outcome. The validity of the algorithm may be examined by analyzing tumor samples of patients with a clinical record, wherein e.g. the score for good outcome patients and bad outcome patients may be determined separately and compared. The skilled person, a biostatistician, will know to apply further mathematical methods, such as discriminate functions to obtain optimized algorithms. Algorithms may be optimized e.g. for sensitivity or specificity. Algorithms may be adapted to the particular analytical platform used to measure gene expression of marker genes, such as quantitiative PCR.

According to an aspect of the invention there is provided a method as described above, wherein said endocrine therapy comprises tamoxifen or an aromatase inhibitor.

According to an aspect of the invention there is provided a method as described above, wherein a risk of developing recurrence is predicted.

According to an aspect of the invention there is provided a method as described above, wherein said expression level is determined as a non-protein expression level.

According to an aspect of the invention there is provided a method as described above, wherein said expression level is determined as an RNA expression level.

According to an aspect of the invention there is provided a method as described above, wherein said expression level is determined by at least one of

    • a PCR based method,
    • a microarray based method, and
    • a hybridization based method.

According to an aspect of the invention there is provided a method as described above, wherein said determination of expression levels is in a formalin-fixed paraffin embedded tumor sample or in a fresh-frozen tumor sample.

According to an aspect of the invention there is provided a method as described above, wherein the expression level of said at least on marker gene is determined as a pattern of expression relative to at least one reference gene or to a computed average expression value.

According to an aspect of the invention there is provided a method as described above, wherein said step of mathematically combining comprises a step of applying an algorithm to values representative of an expression level of a given gene.

According to an aspect of the invention there is provided a method as described above, wherein said algorithm is a linear combination of said values representative of an expression level of a given gene.

According to an aspect of the invention there is provided a method as described above, wherein a value for a representative of an expression level of a given gene is multiplied with a coefficient.

According to an aspect of the invention there is provided a method as described above, wherein one, two or more thresholds are determined for said combined score and discriminated into high and low risk, high, intermediate and low risk, or more risk groups by applying the threshold on the combined score.

According to an aspect of the invention there is provided a method as described above, wherein a high combined score is indicative of benefit from a more aggressive therapy, e.g. cytotoxic chemotherapy. The skilled person understands that a “high score” in this regard relates to a reference value or cutoff value. The skilled person further understands that depending on the particular algorithm used to obtain the combined score, also a “low” score below a cut off or reference value can be indicative of benefit from a more aggressive therapy, e.g. cytotoxic chemotherapy. This is the case when genes having a positive correlation with high risk of metastasis factor into the algorithm with a positive coefficient, such that an overall high score indicates high expression of genes having a positive correlation with high risk.

According to an aspect of the invention there is provided a method as described above, wherein an information regarding nodal status of the patient is processed in the step of mathematically combining expression level values for the genes to yield a combined score.

According to an aspect of the invention there is provided a method as described above, wherein said information regarding nodal status is a numerical value≦0 if said nodal status is negative and said information is a numerical value>0 if said nodal status positive or unknown. In exemplary embodiments of the invention a negative nodal status is assigned the value 0, an unknown nodal status is assigned the value 0.5 and a positive nodal status is assigned the value 1. Other values may be chosen to reflect a different weighting of the nodal status within an algorithm.

The invention further relates to a kit for performing a method as described above, said kit comprising a set of oligonucleotides capable of specifically binding sequences or to seqences of fragments of the genes in a combination of genes, wherein

(i) said combination comprises at least the 8 genes UBE2C, BIRC5, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP; or

(ii) said combination comprises at least the 10 genes BIRC5, AURKA, PVALB, NMU, STC2, RBBP8, PTGER3, CXCL12, CDH1, and PIP; or

(iii) said combination comprises at least the 9 genes BIRC5, DHCR7, RACGAP1, PVALB, STC2, IL6ST, PTGER3, CXCL12, and ABAT; or

(iv) said combination comprises at least the 9 genes DHCR7, RACGAP1, NMU, AZGP1, RBBP8, IL6ST, and MGP;

The invention further relates to the use of a kit for performing a method of any of claims 1 to 17, said kit comprising a set of oligonucleotides capable of specifically binding sequences or to sequences of fragments of the genes in a combination of genes, wherein

(i) said combination comprises at least the 8 genes UBE2C, BIRC5, DHCR7, STC2, AZGP1, RBBP8, IL6ST, and MGP; or

(ii) said combination comprises at least the 10 genes BIRC5, AURKA, PVALB, NMU, STC2, RBBP8, PTGER3, CXCL12, CDH1, and PIP; or

(iii) said combination comprises at least the 9 genes BIRC5, DHCR7, RACGAP1, PVALB, STC2, IL6ST, PTGER3, CXCL12, and ABAT; or

(iv) said combination comprises at least the 9 genes DHCR7, RACGAP1, NMU, AZGP1, RBBP8, IL6ST, and MGP;19. A computer program product capable of processing values representative of an expression level of the genes AKR1C3, MAP4 and SPP1 by mathematically combining said values to yield a combined score, wherein said combined score is indicative of benefit from cytotoxic chemotherapy of said patient.

The invention further relates to a computer program product capable of processing values representative of an expression level of a combination of genes mathematically combining said values to yield a combined score, wherein said combined score is indicative of efficacy or benefit from endocrine therapy of said patient, according to the above methods.

Said computer program product may be stored on a data carrier or implemented on a diagnostic system capable of outputting values representative of an expression level of a given gene, such as a real time PCR system.

If the computer program product is stored on a data carrier or running on a computer, operating personal can input the expression values obtained for the expression level of the respective genes. The computer program product can then apply an algorithm to produce a combined score indicative of benefit from cytotoxic chemotherapy for a given patient.

The methods of the present invention have the advantage of providing a reliable prediction of an outcome of disease based on the use of only a small number of genes. The methods of the present invention have been found to be especially suited for analyzing the response to endocrine treatment, e.g. by tamoxifen, of patients with tumors classified as ESR1 positive and ERBB2 negative.

DETAILED DESCRIPTION OF THE INVENTION

The invention is explained in conjunction with exemplary embodiments and the attached figures:

FIG. 1 shows a Forrest Plot of the adjusted hazard unit ratio with 95% confidence intervall of the T5 score in the combined cohort, as well as the individual treatment arms of the ABCSG06 and 08 studies, using distant metastasis as endpoint.

FIG. 2 shows a Kaplan Meier Analysis of ER+, HER−, N0-3 patients from the combined ABCSG06 and 08 cohorts, stratified as high or low risk according to T5 Score value.

Herein disclosed are unique combinations of marker genes which can be combined into an algorithm for the here presented new predictive test. Technically, the method of the invention can be practiced using two technologies: 1.) Isolation of total RNA from fresh or fixed tumor tissue and 2.) Kinetic RT-PCR of the isolated nucleic acids. Alternatively, it is contemplated to measure expression levels using alternative technologies, e.g by microarray or by measurement at a protein level.

The methods of the invention are based on quantitative determination of RNA species isolated from the tumor in order to obtain expression values and subsequent bioinformatic analysis of said determined expression values. RNA species might be isolated from any type of tumor sample, e.g. biopsy samples, smear samples, resected tumor material, fresh frozen tumor tissue or from paraffin embedded and formalin fixed tumor tissue. First, RNA levels of genes coding for specific combinations of the genes UBE2C, BIRC5, DHCR7, RACGAP1, AURKA, PVALB, NMU, STC2, AZGP1, RBBP8, IL6ST, MGP, PTGER3, CXCL12, ABAT, CDH1, and PIP or specific combinations thereof, as indicated, are determined. Based on these expression values a prognostic score is calculated by a mathematical combination, e.g. according to formulas T5 T1, T4, or T5b (see below). A high score value indicates a high risk for development of distant metastasis, a low score value indicates a low risk of distant metastasis. Consequently, a high score also indicates that the patient is a high risk patient who will benefit from a more aggressive therapy, e.g. cytotoxic chemotherapy.

The present examples are based on identification of prognostic genes using tumors of patients homogeneously treated in the adjuvant setting with tamoxifen. Furthermore, identification of relevant genes has been restricted to tumors classified as ESR1 positive and ERBB2 negative based on RNA expression levels. In addition, genes allowing separation of intermediate risk, e.g. grade 2 tumors were considered for algorithm development. Finally, a platform transfer from Affymetrix HG_U133a arrays to quantitative real time PCR, as well as a sample type transfer from fresh frozen tissue to FFPE tissue was performed to ensure robust algorithm performance, independent from platform and tissue type. As a result, determination of the expression level of RNA species from the primary tumor and the subsequent complex and multivariate analysis as described above provides a superior method for prediction of the likelihood of disease recurrence in patients diagnosed with lymph node negative or positive early breast cancer, when treated with tamoxifen only in the adjuvant setting. Thus the test relies on fewer genes than those of the competitors but provides superior information regarding high sensitivity and negative predictive value, in particular for tumors considered to exhibit an intermediate risk of recurrence based on standard clinical factors.

The total RNA was extracted with a Siemens, silica bead-based and fully automated isolation method for RNA from one 10 μm whole FFPE tissue section on a Hamilton MICROLAB STARlet liquid handling robot (17). The robot, buffers and chemicals were part of a Siemens VERSANT® kPCR Molecular System (Siemens Healthcare Diagnostics, Tarrytown, N.Y.; not commercially available in the USA). Briefly, 150 μl FFPE buffer (Buffer FFPE, research reagent, Siemens Healthcare Diagnostics) were added to each section and incubated for 30 minutes at 80° C. with shaking to melt the paraffin. After cooling down, proteinase K was added and incubated for 30 minutes at 65° C. After lysis, residual tissue debris was removed from the lysis fluid by a 15 minutes incubation step at 65° C. with 40 μl silica-coated iron oxide beads. The beads with surface-bound tissue debris were separated with a magnet and the lysates were transferred to a standard 2 ml deep well-plate (96 wells). There, the total RNA and DNA was bound to 40 μl unused beads and incubated at room temperature. Chaotropic conditions were produced by the addition of 600 μl lysis buffer. Then, the beads were magnetically separated and the supernatants were discarded. Afterwards, the surface-bound nucleic acids were washed three times followed by magnetization, aspiration and disposal of supernatants. Afterwards, the nucleic acids were eluted by incubation of the beads with 100 μl elution buffer for 10 minutes at 70° C. with shaking. Finally, the beads were separated and the supernatant incubated with 12 μl DNase I Mix (2 μL DNase I (RNase free); 10 μl 10× DNase I buffer; Ambion/Applied Biosystems, Darmstadt, Germany) to remove contaminating DNA. After incubation for 30 minutes at 37° C., the DNA-free total RNA solution was aliquoted and stored at −80° C. or directly used for mRNA expression analysis by reverse transcription kinetic PCR (RTkPCR). All the samples were analyzed with one-step RT-kPCR for the gene expression of up to three reference genes, (RPL37A, CALM2, OAZ1) and up to 16 target genes in an ABI PRISM® 7900HT (Applied Biosystems, Darmstadt, Germany). The SuperScript® III Platinum® One-Step Quantitative RT-PCR System with ROX (6-carboxy-X-rhodamine) (Invitrogen, Karlsruhe, Germany) was used according to the manufacturer's instructions. Respective probes and primers are shown in table 1. The PCR conditions were as follows: 30 minutes at 50° C., 2 minutes at 95° C. followed by 40 cycles of 15 seconds at 95° C. and 30 seconds at 60° C. All the PCR assays were performed in triplicate. As surrogate marker for RNA yield, the housekeeper gene, RPL37A cycle threshold (Ct) value was used as described elsewhere (17). The relative gene expression levels of the target genes were calculated by the delta-Ct method using the formula:


20−(Ct(target)−mean(Ct(reference genes))).

A platform transfer from Affymetrix HG_U133a arrays (fresh frozen tissue) to quantitative real time PCR (FFPE tissue) was calculated as follows. Material from 158 patients was measured using both platforms to yield paired samples. Delta-Ct values were calculated from the PCR data. Log 2-Expressions were calculated from the Affymetrix data by applying a lower bound (setting all values below the lower bound to the lower bound) and then calculating the logarithm of base 2. The application of a lower bound reduces the effect of increased relative measurement noise for low expressed genes/samples; a lower bound of 20 was used, lower bounds between 0.1 and 200 also work well. A HG_U133a probe set was selected for each PCR-measured gene by maximizing the Pearson correlation coefficient between the delta-Ct value (from PCR) and the log 2-expression (from Affymetrix). Other correlation measures will also work well, e.g. the Spearman correlation coefficient. In most cases the best-correlating probe set belonged to the intended gene, for the remaining cases the PCR-gene was removed for further processing. Those genes showing a bad correlation between platforms were also removed, where a threshold on the Pearson correlation coefficient of 0.7 was used (values of between 0.5 and 0.8) also work well. The platform transformation was finalized by calculating unsupervised z-transformations for both platforms and combining them; a single PCR-delta-Ct value then is transformed to the Affymetrix scale by the following steps: (i) apply affine linear transformation where coefficients were determined by z-transformation of PCR data, (ii) apply inverse affine linear transformation where coefficients were determined by z-transformation of Affymetrix data, (iii) invert log 2, i.e. calculate exponential with respect to base 2. Alternatives to the two-fold z-transformations are linear or higher order regression, robust regression or principal component based methods, which will also work well.

The sequences of the primers and probes were as follows:

TABLE 1
Primer and probe sequences for the respective genes:
SeqSeq
geneprobeIDforward primerID
ABATTCGCCCTAAGAGGCTCTTCCTC1GGCAACTTGAGGTCTGACTTTTG2
ADRA2ATTGTCCTTTCCCCCCTCCGTGC4CCCCAAGAGCTGTTAGGTATCAA5
APODCATCAGCTCTCAACTCCTGGTTTAACA7ACTCACTAATGGAAAACGGAAAGATC8
ASPHTGGGAGGAAGGCAAGGTGCTCATC10TGTGCCAACGAGACCAAGAC11
AURKACCGTCAGCCTGTGCTAGGCAT13AATCTGGAGGCAAGGTTCGA14
BIRC5AGCCAGATGACGACCCCATAGAGGAACA16CCCAGTGTTTCTTCTGCTTCAAG17
CELSR2ACTGACTTTCCTTCTGGAGCAGGTGGC19TCCAAGCATGTATTCCAGACTTGT20
CHPT1CCACGGCCACCGAAGAGGCAC22CGCTCGTGCTCATCTCCTACT23
CXCL12CCACAGCAGGGTTTCAGGTTCC25GCCACTACCCCCTCCTGAA26
CYBRD1AGGGCATCGCCATCATCGTC28GTCACCGGCTTCGTCTTCA29
DCNTCTTTTCAGCAACCCGGTCCA31AAGGCTTCTTATTCGGGTGTGA32
DHCR7TGAGCGCCCACCCTCTCGA34GGGCTCTGCTTCCCGATT35
E2F8CAGGATACCTAATCCCTCTCACGCAG37AAATGTCTCCGCAACCTTGTTC38
EPHX2TGAAGCGGGAGGACTTTTTGTAAA40CGATGAGAGTGTTTTATCCATGCA41
ESR1ATGCCCTTTTGCCGATGCA43GCCAAATTGTGTTTGATGGATTAA44
GJA1TGCACAGCCTTTTGATTTCCCCGAT46CGGGAAGCACCATCTCTAACTC47
HSPA2CAAGTCAGCAAACACGCAAAA49CATGCACGAACTAATCAAAAATGC50
IL6STCAAGCTCCACCTTCCAAAGGACCT52CCCTGAATCCATAAAGGCATACC53
INPP4BTCCGAGCGCTGGATTGCATGAG55GCACCAGTTACACAAGGACTTCTTT56
MAPTAGACTATTTGCACACTGCCGCCT58GTGGCTCAAAGGATAATATCAAACAC59
MGPCCTTCATATCCCCTCAGCAGAGATGG61CCTTCATTAACAGGAGAAATGCAA62
NEK2TCCTGAACAAATGAATCGCATGTCCTACAA64ATTTGTTGGCACACCTTATTACATGT65
PCNAAAATACTAAAATGCGCCGGCAATGA67GGGCGTGAACCTCACCAGTA68
PGRTTGATAGAAACGCTGTGAGCTCGA70AGCTCATCAAGGCAATTGGTTT71
PIPTGCATGGTGGTTAAAACTTACCTCA73TGCTTGCAGTTCAAACAGAATTG74
PLATCAGAAAGTGGCCATGCCACCCTG76TGGGAAGACATGAATGCACACTA77
PRSS16CACTGCCGGTCACCCACACCA79CTGAGGAGCACAGAACCTCAACT80
PTGER3TCGGTCTGCTGGTCTCCGCTCC82CTGATTGAAGATCATTTTCAACATCA83
PTPRTTTGGCTTCTGGACACCCTCACA85GAGTTGTGGCCTCTACCATTGC86
RACGAP1ACTGAGAATCTCCACCCGGCGCA88TCGCCAACTGGATAAATTGGA89
RBBP8ACCGATTCCGCTACATTCCACCCAAC91AGAAATTGGCTTCCTGCTCAAG92
SCUBE2CTAGAGGGTTCCAGGTCCCATACGTGACATA94TGTGGATTCAGTTCAAGTCCAATG95
SEC14L2TGGGAGGCATGCAACGCGTG97AGGTCTTACTAAGCAGTCCCATCTCT98
SQLETATGCGTCTCCCAAAAGAAGAACACCTCG100GCAAGCTTCCTTCCTCCTTCA101
TFAP2BCAACACCACCACTAACAGGCACACGTC103GGCATGGACAAGATGTTCTTGA104
TOP2ACAGATCAGGACCAAGATGGTTCCCACAT106CATTGAAGACGCTTCGTTATGG107
TRIM29TGCTGTCTCACTACCGGCCATTCTACG109TGGAAATCTGGCAAGCAGACT110
UBE2CTGAACACACATGCTGCCGAGCTCTG112CTTCTAGGAGAACCCAACATTGATAGT113
WNT5ATATTCACATCCCCTCAGTTGCAGTGAATTG115CTGTGGCTCTTAATTTATTGCATAATG116
STC2TCTCACCTTGACCCTCAGCCAAG118ACATTTGACAAATTTCCCTTAGGATT119
AZGP1CACCAGCCACCAGGCCCCAG121TCCTGGACCGGCAAGATC122
CALM2TCGCGTCTCGGAAACCGGTAGC124GAGCGAGCTGAGTGGTTGTG125
CDH1CCTGCCAATCCCGATGAAATTGGAAAT127TGAGTGTCCCCCGGTATCTTC128
NMUACCCTGCTGACCTTCTTCCATTCCGT130AGAAATTGGCTTCCTGCTCAAG131
OAZ1TGCTTCCACAAGAACCGCGAGGA133CGAGCCGACCATGTCTTCAT134
PVALBAAGTTCTTCCAAATGGTCGGCC136CCGACTCCTTCGACCACAA137
RPL37ATGGCTGGCGGTGCCTGGA139TGTGGTTCCTGCATGAAGACA140
Seq
genereverse primerID
ABATGGTCAGCTCACAAGTGGTGTGA3
ADRA2ATCAATGACATGATCTCAACCAGAA6
APODTCACCTTCGATTTGATTCACAGTT9
ASPHTCGTGCTCAAAGGAGTCATCA12
AURKATCTGGATTTGCCTCCTGTGAA15
BIRC5CAACCGGACGAATGCTTTTT18
CELSR2TGCCCACAGCCTCTTTTTCT21
CHPT1CCCAGTGCACATAAAAGGTATGTC24
CXCL12TCACCTTGCCAACAGTTCTGAT27
CYBRD1CAGGTCCACGGCAGTCTGT30
DCNTGGATGGCTGTATCTCCCAGTA33
DHCR7AGTCATAGGGCAAGCAGAAAATTC36
E2F8CTGCCCCCAGGGATGAG39
EPHX2GCTGAGGCTGGGCTCTTCT42
ESR1GACAAAACCGAGTCACATCAGTAATAG45
GJA1TTCATGTCCAGCAGCTAGTTTTTT48
HSPA2ACATTATTCGAGGTTTCTCTTTAATGC51
IL6STCAGCTTCGTTTTTCCCTACTTTTT54
INPP4BTCTCTATGCGGCATCCTTCTC57
MAPTACCTTGCTCAGGTCAACTGGTT60
MGPATTGAGCTCGTGGACAGGCTTA63
NEK2AAGCAGCCCAATGACCAGATa66
PCNACTTCGGCCCTTAGTGTAATGATATC69
PGRACAAGATCATGCAAGTTATCAAGAAGTT72
PIPCACCTTGTAGAGGGATGCTGCTA75
PLATGGAGGTTGGGCTTTAGCTGAA78
PRSS16CGAACTCGGTACATGTCTGATACAA81
PTGER3GACGGCCATTCAGCTTATGG84
PTPRTGAGCGGGAACCTTGGGATAG87
RACGAP1GAATGTGCGGAATCTGTTTGAG90
RBBP8AAAACCAACTTCCCAAAAATTCTCT93
SCUBE2CCATCTCGAACTATGTCTTCAATGAGT96
SEC14L2CGACCGGCACCTGAACTC99
SQLECCTTTAGCAGTTTTCTCCATAGTTTTATATC102
TFAP2BCCTCCTTGTCGCCAGTTTTACT105
TOP2ACCAGTTGTGATGGATAAAATTAATCAG108
TRIM29CAATCCCGTTGCCTTTGTTG111
UBE2CGTTTCTTGCAGGTACTTCTTAAAAGCT114
WNT5ATTAGTGCTTTTTGCTTTCAAGATCTT117
STC2CCAGGACGCAGCTTTACCAA120
AZGP1TAGGCCAGGCACTTCAGTTTC123
CALM2AGTCAGTTGGTCAGCCATGCT126
CDH1TCAGCCGCTTTCAGATTTTCA129
NMUAAAACCAACTTCCCAAAAATTCTCT132
OAZ1AAGCCCAAAAAGCTGAAGGTT135
PVALBCATCATCCGCACTCTTTTTCTTC138
RPL37AGTGACAGCGGAAGTGGTATTGTAC141

Table 2, below, lists the genes used in the methods of the invention and in the particular embodiments T5, T1, T4, and T5b. Table 2 also shows whether overexpression of a given gene is indicative of good or bad outcome under Tamoxifen therapy. Table 2 lists the function of the gene, the compartment localization within the cell and the cellular processes it is involved in.

TABLE 2
List of genes of algorithms T5, T1, T4, and T5b:
High
GeneNameExpressionFunctionComponentProcess
UBE2Cubiquitin-BadATPcytosolcell
conjugatingOutcomebindingdivision
enzyme E2C
BIRC5baculoviralBadRan GTPasecytosolcell cycle
IAP repeat-Outcomebinding
containing 5
DHCR77-Bad7-endoplasmaticregulation
dehydrocholesterolOutcomedehydrocholesterolreticulumof cell
reductasereductasemembraneproliferation
activity
RACGAP1Rac GTPaseBadGTPasecytoplasmcell cycle
activatingOutcomeactivator
protein 1activity
AURKAauroraBadATPcentrosomemitotic
kinase AOutcomebindingcell cycle
PVALBparvalbuminBadcalcium
Outcomeion
binding
NMUneuromedin UBadreceptorextracellularsignal
Outcomebindingregiontransduction
STC2stanniocalcin 2Goodhormoneextracellularcell
Outcomeactivityregionsurface
receptor
linked
signal
transduction
AZGP1alpha-2-Goodproteinextracellularnegative
glycoprotein 1Outcometransmembraneregionregulation
transporterof cell
activityproliferation
RBBP8retinoblastomaGoodproteinnucleuscell cycle
bindingOutcomebindingcheckpoint
protein 8
IL6STinterleukinGoodreceptorextracellularsignal
6 signalOutcomeactivityregiontransduction
transducer
MGPmatrix GlaGoodextracellularextracellularcell
proteinOutcomematrixregiondifferentiation
structural
constituent
PTGER3prostaglandin EGoodligand-plasmasignal
receptor 3Outcomedependentmembranetransduction
receptor
activity
CXCL12chemokineGoodchemokineextracellularsignal
(C-XCOutcomeactivityregiontransduction
motif)
ligand 12
ABAT4-Goodtransferasemitochondriongamma-
aminobutyrateOutcomeactivityaminobutyric
aminotransferaseacid
catabolic
process
CDH1cadherin 1Goodcellplasmahomophilic
Outcomeadhesionmembranecell
moleculeadhesion
binding
PIPprolactin-Goodactinextracellular
inducedOutcomebindinregion
protein
CALM2Reference
Gene
OAZ1Reference
Gene
RPL37AReference
Gene

Table 3, below, shows the combinations of genes used for each algorithm.

TABLE 3
Combination of genes for the respective algorithms:
GeneAlgo_T1Algo_T4Algo_T5Algo_T5b
UBE2CX
BIRC5XXX
DHCR7XXX
RACGAP1XX
AURKAX
PVALBXX
NMUXX
STC2XXX
AZGP1XX
RBBP8XXX
IL6STXXX
MGPXX
PTGER3XX
CXCL12XX
ABATX
CDH1X
PIPX

Table 4, below, shows Affy probeset ID and TagMan design ID mapping of the marker genes of the present invention.

TABLE 4
Gene symbol, Affy probeset ID and TaqMan design ID mapping:
GeneDesign IDProbeset ID
UBE2CR65202954_at
BIRC5SC089202095_s_at
DHCR7CAGMC334201791_s_at
RACGAP1R125-2222077_s_at
AURKACAGMC336204092_s_at
PVALBCAGMC339205336_at
NMUCAGMC331206023_at
STC2R52203438_at
AZGP1CAGMC372209309_at
RBBP8CAGMC347203344_s_at
IL6STCAGMC312212196_at
MGPCAGMC383202291_s_at
PTGER3CAGMC315213933_at
CXCL12CAGMC342209687_at
ABATCAGMC338209460_at
CDH1CAGMC335201131_s_at

Table 5, below, shows full names, Entrez GeneID, gene bank accession number and chromosomal location of the marker genes of the present invention

OfficialEntrezAccesion
SymbolOfficial Full NameGeneIDNumberLocation
UBE2Cubiquitin-11065U7337920q13.12
conjugating enzyme
E2C
BIRC5baculoviral IAP332U7528517q25
repeat-containing 5
DHCR77-1717AF03454411q13.4
dehydrocholesterol
reductase
STC2staniocalcin 28614AB0126645q35.2
RBBP8retinoblastoma5932AF04343118q11.2
binding protein 8
IL6STinterleukin 63572M572305q11
signal transducer
MGPmatrix Gla protein4256M5854912p12.3
AZGP1alpha-2-563BC00530611q22.1
glycoprotein 1,
zinc-binding
RACGAP1Rac GTPase29127NM_01327712q13
activating protein 1
AURKAaurora kinase A6790BC00128020q13
PVALBparvalbumin5816NM_00285422q13.1
NMUneuromedin U10874X760294q12
PTGER3prostaglandin E5733X838631p31.2
receptor 3 (subtype
EP3)
CXCL12chemokine6387L3603310q11.1
(C—X—C
motif) ligand 12
(stromal cell-
derived factor 1)
ABAT4-aminobutyrat18L3296116p13.2
aminotransferase
CDH1cadherin 1, type 1,999L0859916q22.1
E-cadherin
(epithelial)
PIPprolactin-induced5304NMM_0026527q32-qter
protein

EXAMPLE ALGORITHM T5

Algorithm T5 is a committee of four members where each member is a linear combination of two genes. The mathematical formulas for T5 are shown below; the notation is the same as for T1. T5 can be calculated from gene expression data only.

riskMember1=0.434039[0.3010.567]*(0.939*BIRC5-3.831) -0.491845[-0.714-0.270]*(0.707*RBBP8-0.934) riskMember2=0.488785[0.3020.675]*(0.794*UBE2C-1.416) -0.374702[-0.570-0.179]*(0.814*IL6ST-5.034) riskMember3=-0.39169[-0.541-0.242]*(0.674*AZGP1-0.777) +0.44229[0.2560.628]*(0.891*DHCR7-4.378) riskMember4=-0.377752[-0.543-0.212]* (0.485*MGP+4.330) -0.177669[-0.267-0.088]*(0.826*STC2-3.630) risk=riskMember1+riskMember2+riskMember3+riskMember4

Coefficients on the left of each line were calculated as COX proportional hazards regression coefficients, the numbers in squared brackets denote 95% confidence bounds for these coefficients. In other words, instead of multiplying the term (0.939*BIRC5−3.831) with 0.434039, it may be multiplied with any coefficient between 0.301 and 0.567 and still give a predictive result with in the 95% confidence bounds. Terms in round brackets on the right of each line denote a platform transfer from PCR to Affymetrix: The variables PVALB, CDH1, . . . denote PCR-based expressions normalized by the reference genes (delta-Ct values), the whole term within round brackets corresponds to the logarithm (base 2) of Affymetrix microarray expression values of corresponding probe sets.

Performance of the algorithm T5 was tested in Tamoxifen or Anastrozole treated patients with no more than 3 positive lymph nodes and ER+, HER2− tumors, who participated in the randomized clinical trials ABCSG06 (n=332) or ABCSG08 (n=1244). As shown in FIG. 1, Cox regression analysis reveals, that the T5 score has a significant association with the development of distant metastasis in all cohorts tested.

Kaplan Meier analysis was performed, after classifying the patients of the combined ABCSG cohorts using a predefined cut off for T5 score. Patients with a low risk of development of a distant metastasis had a T5 score≦−9.3, while patients with a high risk of development of a distant metastasis had a T5 score above −9.3. As shown in FIG. 2, a highly significant separation of both risk groups is observed.

Importantly, the T5 score was evaluated and compared against “Adjuvant! Online”, an online tool to aid in therapy selection based on entry of clinical parameters such as tumor size, tumor grade and nodal status. When the T5 score was tested by bivariate Cox regression against the Adjuvant!Online Relapse Risk score, both scores remained a significant association with the development of distant metastasis. Bivariate Cox regression using dichotomized data, which were stratified according to T5 (cut off −9.3) respectively to Adjuvant!Online (cut off 8), again yielded highly significant and independent associations with time to metastasis as clinical endpoint.

TABLE 6
Bivariate Cox regression von T5 und Adjuvant!Online
Hazard
Variableratio95% CI*P
Adjuvant!Online2.361.58-3.54<0.0001
Gene-expression2.621.71-4.01<0.0001
signature (risk group)
Adjuvant!Online (score)1.041.02-1.06<0.0001
Gene-expression1.351.21-1.49<0.0001
signature (risk group)
with HR = Hazard Ratio,
95% CI = 95% Confidence Interval,
p = P value.

Exemplary Kaplan Meyer Curves are shown in FIG. 1 wherein High=High Risk Group, Low=Low Risk Group according to a predefined cut off

A high value of the T5 score indicates an increased risk of occurrence of distant metastasis in a given time period.

This has been shown to be the case for patients having been treated with tamoxifen and also for patients having been treated with aromatase inhibitors.

EXAMPLE ALGORITHM T1

Algorithm T1 is a committee of three members where each member is a linear combination of up to four variables. In general variables may be gene expressions or clinical variables. In T1 the only non-gene variable is the nodal status coded 0, if patient is lymph-node negative and 1, if patient is lymph-node-positive. The mathematical formulas for T1 are shown below.

riskMember1=+0.193935[0.1080.280]*(0.792*PVALB-2.189) -0.240252[-0.400-0.080]*(0.859*CDH1-2.900) -0.270069[-0.385-0.155]*(0.821*STC2-3.529) +1.2053[0.5341.877]*nodalStatus riskMember2=-0.25051[-0.437-0.064]*(0.558*CXCL12+0.324) -0.421992[-0.687-0.157]*(0.715*RBBP8-1.063) +0.148497[0.0290.268]*(1.823*NMU-12.563) +0.293563[0.1080.479]*(0.989*BIRC5-4.536) riskMember3=+0.308391[0.0740.543]*(0.812*AURKA-2.656) -0.225358[-0.395-0.055]*(0.637*PTGER3+0.492) -0.116312[-0.202-0.031]*(0.724*PIP+0.985) risk=+riskMember1+riskMember2+riskMember3

Coefficients on the left of each line were calculated as COX proportional hazards regression coefficients, the numbers in squared brackets denote 95% confidence bounds for these coefficients. Terms in round brackets on the right of each line denote a platform transfer from PCR to Affymetrix: The variables PVALB, CDH1, . . . denote PCR-based expressions normalized by the reference genes, the whole term within round brackets corresponds to the logarithm (base 2) of Affymetrix microarray expression values of corresponding probe sets.

EXAMPLE ALGORITHM T4

Algorithm T4 is a linear combination of motifs. The top 10 genes of several analyses of Affymetrix datasets and PCR data were clustered to motifs. Genes not belonging to a cluster were used as single gene-motifs. COX proportional hazards regression coefficients were found in a multivariate analysis.

In general motifs may be single gene expressions or mean gene expressions of correlated genes. The mathematical formulas for T4 are shown below.


prolif=((0.84 [0.697 . . . 0.977]*RACGAP1−2.174)+(0.85 [0.713 . . . 0.988]*DHCR7−3.808)+(0.94 [0.786 . . . 1.089]*BIRC5−3.734))/3


motiv2=((0.83 [0.693 . . . 0.96]*IL6ST−5.295)+(1.11 [0.930 . . . 1.288]*ABAT−7.019)+(0.84 [0.701 . . . 0.972]*STC2−3.857))/3


ptger3=(PTGER3*0.57 [0.475 . . . 0.659]+1.436)


cxcl12=(CXCL12*0.53 [0.446 . . . 0.618]−0.847)


pvalb=(PVALB*0.67 [0.558 . . . 0.774]−0.466)

Factors and offsets for each gene denote a platform transfer from PCR to Affymetrix: The variables RACGAP1, DHCR7, . . . denote PCR-based expressions normalized by CALM2 and PPIA, the whole term within round brackets corresponds to the logarithm (base 2) of Affymetrix microarray expression values of corresponding probe sets.

The numbers in squared brackets denote 95% confidence bounds for these factors.

As the algorithm performed even better in combination with a clinical variable the nodal status was added. In T4 the nodal status is coded 0, if patient is lymph-node negative and 1, if patient is lymph-node-positive. With this, algorithm T4 is:

risk=-0.32[-0.510-0.137]*motiv2+0.65[0.4110.886]*prolif-0.24[-0.398-0.08]*ptger3-0.05[-0.2250.131]*cxcl12+0.09[0.0190.154]*pvalb+nodalStatus

Coefficients of the risk were calculated as COX proportional hazards regression coefficients, the numbers in squared brackets denote 95% confidence bounds for these coefficients.

Algorithm T5b is a committee of two members where each member is a linear combination of four genes. The mathematical formulas for T5b are shown below, the notation is the same as for T1 and T5. In T5b a non-gene variable is the nodal status coded 0, if patient is lymph-node negative and 1, if patient is lymph-node-positive and 0.5 if the lymph-node status is unknown. T5b is defined by:

riskMember1=0.359536[0.1530.566]*(0.891*DHCR7-4.378) -0.288119[-0.463-0.113]*(0.485*MGP+4.330) +0.257341[0.1120.403]*(1.118*NMU-5.128) -0.337663[-0.499-0.176]*(0.674*AZGP1-0.777) riskMember2=-0.374940[-0.611-0.139]*(0.717*RBBP8-0.934) -0.387371[-0.597-0.178]*(0.814*IL6ST-5.034) +0.800745[0.5511.051]*(0.860*RACGAP1-2.518) +0.770650[0.3231.219]*Nodalstatus risk=riskMember1+riskMember2

The skilled person understands that these algorithms represent particular examples and that based on the information regarding association of gene expression with outcome as given in table 2 alternative algorithms can be established using routine skills.

Algorithm Simplification by Employing Subsets of Genes

“Example algorithm T5” is a committee predictor consisting of 4 members with 2 genes of interest each. Each member is an independent and self-contained predictor of distant recurrence, each additional member contributes to robustness and predictive power of the algorithm to predict time to metastasis, time to death or likelihood of survival for a breast cancer patient. The equation below shows the “Example Algorithm T5”; for ease of reading the number of digits after the decimal point has been truncated to 2; the range in square brackets lists the estimated range of the coefficients (mean+/−3 standard deviations).

T5Algorithm: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 +0.38[0.150.61]*UBE2C-0.30[-0.55-0.06]*IL6ST -0.28[-0.43-0.12]*AZGP1+0.42[0.160.68]*DHCR7 -0.18[-0.31-0.06]*MGP-0.13[-0.25-0.02]*STC2 c-indices:trainSet=0.724,

Gene names in the algorithm denote the difference of the mRNA expression of the gene compared to one or more housekeeping genes as described above.

Analysing a cohort different from the finding cohort (234 tumor samples) it was surprising to learn that some simplifications of the “original T5 Algorithm” still yielded a diagnostic performance not significantly inferior to the original T5 algorithm. The most straightforward simplification was reducing the committee predictor to one member only. Examples for the performance of the “one-member committees” are shown below:

member1only:+0.41[0.21..0.61]*BIRC5-0.33[-0.57..-0.09]*RBBP8 c-indices:trainSet=0.653,independentCohort=0.681 member2only:+0.38[0.15..0.61]*UBE2C-0.30[-0.55..-0.06]*IL6ST c-indices:trainSet=0.664,independentCohort=0.696 member3only:-0.28[-0.43..-0.12]*AZGP1+0.42[0.16..0.68]*DHCR7 c-indices:trainSet=0.666,independentCohort=0.601 member4only:-0.18[-0.31..-0.06]*MGP-0.13[-0.25..-0.02]*STC2 c-indices:trainSet=0.668,independentCohort=0.681

The performance of the one member committees as shown in an independent cohort of 234 samples is notably reduced compared to the performance of the full algorithm. Still, using a committee consisting of fewer members allows for a simpler, less costly estimate of the risk of breast cancer recurrence or breast cancer death that might be acceptable for certain diagnostic purposes.

Gradually combining more than one but less than four members to a new prognostic committee predictor algorithm, frequently leads to a small but significant increase in the diagnostic performance compared to a one-member committee. It was surprising to learn that there were marked improvements by some combination of committee members while other combinations yielded next to no improvement. Initially, the hypothesis was that a combination of members representing similar biological motives as reflected by the employed genes yielded a smaller improvement than combining members reflecting distinctly different biological motives. Still, this was not the case. No rule could be identified to foretell the combination of some genes to generate an algorithm exibiting more prognostic power than another combination of genes. Promising combinations could only be selected based on experimental data.

Identified combinations of combined committee members to yield simplified yet powerful algorithms are shown below.

members1and2only: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 +0.38[0.150.61]*UBE2C-0.30[-0.55-0.06]*IL6ST c-indices:trainSet=0.675,independentCohort=0.712 members1and3only: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 -0.28[-0.43-0.12]*AZGP1+0.42[0.160.68]*DHCR7 c-indices:trainSet=0.697,independentCohort=0.688 members1and4only: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 -0.18[-0.31-0.06]*MGP-0.13[-0.25-0.02]*STC2 c-indices:trainSet=0.705,independentCohort=0.679 members2and3only: +0.38[0.150.61]*UBE2C-0.30[-0.55-0.06]*IL6ST -0.28[-0.43-0.12]*AZGP1+0.42[0.160.68]*DHCR7 c-indices:trainSet=0.698,independentCohort=0.670 members1,2and3only: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 +0.38[0.150.61]*UBE2C-0.30[-0.55-0.06]*IL6ST -0.28[-0.43-0.12]*AZGP1+0.42[0.160.68]*DHCR7 c-indices:trainSet=0.701,independentCohort=0.715

Not omitting complete committee members but a single gene or genes from different committee members is also possible but requires a retraining of the entire algorithm. Still, it can also be advantageous to perform. The performance of simplified algorithms generated by omitting entire members or individual genes is largely identical.

Algorithm Variants by Gene Replacement

Described algorithms, such as “Example algorithm T5”, above can be also be modified by replacing one or more genes by one or more other genes. The purpose of such modifications is to replace genes difficult to measure on a specific platform by a gene more straightforward to assay on this platform. While such transfer may not necessarily yield an improved performance compared to a starting algorithm, it can yield the clue to implanting the prognostic algorithm to a particular diagnostic platform. In general, replacing one gene by another gene while preserving the diagnostic power of the predictive algorithm can be best accomplished by replacing one gene by a co-expressed gene with a high correlation (shown e.g. by the Pearson correlation coefficient). Still, one has to keep in mind that the mRNA expression of two genes highly correlative on one platform may appear quite independent from each other when assessed on another platform. Accordingly, such an apparently easy replacement when reduced to practice experimentally, may yield disappointingly poor results as well as surprising strong results, always depending on the imponderabilia of the platform employed. By repeating this procedure one can replace several genes.

The efficiency of such an approach can be demonstrated by evaluating the predictive performance of the T5 algorithm score and its variants on the validation cohorts. The following table shows the c-index with respect to endpoint distant recurrence in two validation cohorts.

TABLE 7
VariantValidation Study AValidation Study B
original algorithm T5c-index = 0.718c-index = 0.686
omission of BIRC5 (settingc-index = 0.672c-index = 0.643
expression to some
constant)
replacing BIRC5 by UBE2Cc-index = 0.707c-index = 0.678
(no adjustment of the
coefficient)

One can see that omission of one of the T5 genes, here shown for BIRC5 for example, notably reduces the predictive performance. Replacing it with another gene yields about the same performance.

A better method of replacing a gene is to re-train the algorithm. Since T5 consists of four independent committee members one has to re-train only the member that contains the replaced gene. The following equations demonstrate replacements of genes of the T5 algorithm shown above trained in a cohort of 234 breast cancer patients. Only one member is shown below, for c-index calculation the remaining members were used unchanged from the original T5 Algorithm. The range in square brackets lists the estimated range of the coefficients: mean+/−3 standard deviations.

Member1ofT5: Originalmember1: +0.41[0.210.61]*BIRC5-0.33[-0.57-0.09]*RBBP8 c-indices:trainSet=0.724,independentCohort=0.705 replaceBIRC5byTOP2Ainmember1: +0.47[0.240.69]*TOP2A-0.34[-0.58-0.10]*RBBP8 c-indices:trainSet=0.734,independentCohort=0.694 replaceBIRC5byRACGAP1inmember1: +0.69[0.371.00]*RACGAP1-0.33[-0.57-0.09]*RBBP8 c-indices:trainSet=0.736,independentCohort=0.743 replaceRBBP8byCELSR2inmember1: +0.38[0.190.57]*BIRC5-0.18[-0.410.05]*CELSR2 c-indices:trainSet=0.726,independentCohort=0.680 replaceRBBP8byPGRinmember1: +0.35[0.150.54]*BIRC5-0.09[-0.230.05]*PGR c-indices:trainSet=0.727,independentCohort=0.731 Member2ofT5: Originalmember2: +0.38[0.150.61]*UBE2C-0.30[-0.55-0.06]*IL6ST c-indices:trainSet=0.724,independentCohort=0.725 replaceUBE2CbyRACGAP1inmember2: +0.65[0.330.96]*RACGAP1-0.38[-0.62-0.13]*IL6ST c-indices:trainSet=0.735,independentCohort=0.718 replaceUBE2CbyTOP2Ainmember2: +0.42[0.200.65]*TOP2A-0.38[-0.62-0.13]*IL6ST c-indices:trainSet=0.734,independentCohort=0.700 replaceIL6STbyINPP4Binmember2: +0.40[0.170.62]*UBE2C-0.25[-0.550.05]*INPP4B c-indices:trainSet=0.725,independentCohort=0.686 replaceIL6STbyMAPTinmember2: +0.45[0.220.69]*UBE2C-0.14[-0.280.01]*MAPT c-indices:trainSet=0.727,independentCohort=0.711 Member3ofT5: Originalmember3: -0.28[-0.43-0.12]*AZGP1+0.42[0.160.68]*DHCR7 c-indices:trainSet=0.724,independentCohort=0.705 replaceAZGP1byPIPinmember3: -0.10[-0.18-0.02]*PIP+0.43[0.160.70]*DHCR7 c-indices:trainSet=0.725,independentCohort=0.692 replaceAZGP1byEPHX2inmember3: -0.23[-0.43-0.02]*EPHX2+0.37[0.100.64]*DHCR7 c-indices:trainSet=0.719,independentCohort=0.698 replaceAZGP1byPLATinmember3: -0.23[-0.40-0.06]*PLAT+0.43[0.180.68]*DHCR7 c-indices:trainSet=0.712,independentCohort=0.715 replaceDHCR7byAURKAinmember3: -0.23[-0.39-0.06]*AZGP1+0.34[0.100.58]*AURKA c-indices:trainSet=0.716,independentCohort=0.733 Member4ofT5: Originalmember4: -0.18[-0.31-0.06]*MGP-0.13[-0.25-0.02]*STC2 c-indices:trainSet=0.724,independentCohort=0.705 replaceMGPbyAPODinmember4: -0.16[-0.30-0.03]*APOD-0.14[-0.26-0.03]*STC2 c-indices:trainSet=0.717,independentCohort=0.679 replaceMGPbyEGFRinmember4: -0.21[-0.37-0.05]*EGFR-0.14[-0.26-0.03]*STC2 c-indices:trainSet=0.715,independentCohort=0.708 replaceSTC2byINPP4Binmember4: -0.18[-0.30-0.05]*MGP-0.22[-0.530.08]*INPP4B c-indices:trainSet=0.719,independentCohort=0.693 replaceSTC2bySEC14L2inmember4: -0.18[-0.31-0.06]*MGP-0.27[-0.49-0.06]*SEC14L2 c-indices:trainSet=0.718,independentCohort=0.681

One can see that replacements of single genes experimentally identified for a quantification with kinetic PCR normally affect the predictive performance of the T5 algorithm, assessed by the c-index only insignificantly.

The following table (Tab. 8) shows potential replacement gene candidates for the genes of T5 algorithm. Each gene candidate is shown in one table cell: The gene name is followed by the bracketed absolute Pearson correlation coefficient of the expression of the original gene in the T5 Algorithm and the replacement candidate, and the HG-U133A probe set ID.

TABLE 8
BIRC5RBBP8UBE2CIL6STAZGP1DHCR7MGPSTC2
UBE2C (0.775),CELSR2BIRC5 (0.775),INPP4BPIP (0.530),AURKA (0.345),APOD (0.368),INPP4B
202954_at(0.548),202095_s_at(0.477),206509_at204092_s_at201525_at(0.500),
204029_at205376_at205376_at
TOP2A (0.757),PGR (0.392),RACGAP1STC2 (0.450),EPHX2 (0.369),BIRC5 (0.323),IL6ST (0.327),IL6ST (0.450),
201292_at208305_at(0.756),203438_at209368_at202095_s_at212196_at212196_at
222077_s_at
RACGAP1STC2 (0.361),TOP2A (0.753),MAPT (0.440),PLAT (0.366),UBE2C (0.315),EGFR (0.308),SEC14L2
(0.704),203438_at201292_at206401_s_at201860_s_at202954_at201983_s_at(0.417),
222077_s_at204541_at
AURKA (0.681),ABAT (0.317),AURKA (0.694),SCUBE2SEC14L2MAPT (0.414),
204092_s_at209459_s_at204092_s_at(0.418),(0.351),206401_s_at
219197_s_at204541_at
NEK2 (0.680),IL6ST (0.311),NEK2 (0.684),ABAT (0.389),SCUBE2CHPT1 (0.410),
204026_s_at212196_at204026_s_at209459_s_at(0.331),221675_s_at
E2F8 (0.640),E2F8 (0.652),PGR (0.377),219197_s_atABAT (0.409),
219990_at219990_at208305_atPGR (0.302),209459_s_at
208305_at
PCNA (0.544),PCNA (0.589),SEC14L2SCUBE2
201202_at201202_at(0.356),(0.406),
CYBRD1CYBRD1204541_at219197_s_at
(0.462),(0.486),ESR1 (0.353),ESR1 (0.394),
217889_s_at217889_s_at205225_at205225_at
DCN (0.439),ADRA2AGJA1 (0.335),RBBP8 (0.361),
209335_at(0.391),201667_at203344_s_at
209869_at
ADRA2ADCN (0.384),MGP (0.327),PGR (0.347),
(0.416),209335_at202291_s_at208305_at
209869_at
SQLE (0.415),SQLE (0.369),EPHX2 (0.313),PTPRT
209218_at209218_at209368_at(0.343),
205948_at
CXCL12CCND1 (0.347),RBBP8 (0.311),HSPA2 (0.317),
(0.388),208712_at203344_s_at211538_s_at
209687_at
EPHX2 (0.362),ASPH (0.344),PTPRT (0.303),PTGER3
209368_at210896_s_at205948_at(0.314),
210832_x_at
ASPH (0.352),CXCL12PLAT (0.301),
210896_s_at(0.342),201860_s_at
209687_at
PRSS16PIP (0.328),
(0.352),206509_at
208165_s_at
EGFR (0.346),PRSS16
201983_s_at(0.326),
208165_s_at
CCND1 (0.331),EGFR (0.320),
208712_at201983_s_at
TRIM29DHCR7 (0.315),
(0.325),201791_s_at
202504_at
DHCR7 (0.323),EPHX2 (0.315),
201791_s_at209368_at
PIP (0.308),TRIM29
206509_at(0.311),
TFAP2B202504_at
(0.306),
214451_at
WNT5A (0.303),
205990_s_at
APOD (0.301),
201525_at
PTPRT (0.301),
205948_at

The following table (Tab. 9) lists qRT-PCR primer and probe sequences used for the table above.

TABLE 9
geneprobeforward primerreverse primer
ABATTCGCCCTAAGAGGCTCTTCCTCGGCAACTTGAGGTCTGACTTTTGGGTCAGCTCACAAGTGGTGTGA
ADRA2ATTGTCCTTTCCCCCCTCCGTGCCCCCAAGAGCTGTTAGGTATCAATCAATGACATGATCTCAACCAGAA
APODCATCAGCTCTCAACTCCTGGTTTAACAACTCACTAATGGAAAACGGAAAGATCTCACCTTCGATTTGATTCACAGTT
ASPHTGGGAGGAAGGCAAGGTGCTCATCTGTGCCAACGAGACCAAGACTCGTGCTCAAAGGAGTCATCA
AURKACCGTCAGCCTGTGCTAGGCATAATCTGGAGGCAAGGTTCGATCTGGATTTGCCTCCTGTGAA
BIRC5AGCCAGATGACGACCCCATAGAGGAACACCCAGTGTTTCTTCTGCTTCAAGCAACCGGACGAATGCTTTTT
CCND1
CELSR2ACTGACTTTCCTTCTGGAGCAGGTGGCTCCAAGCATGTATTCCAGACTTGTTGCCCACAGCCTCTTTTTCT
CHPT1CCACGGCCACCGAAGAGGCACCGCTCGTGCTCATCTCCTACTCCCAGTGCACATAAAAGGTATGTC
CXCL12CCACAGCAGGGTTTCAGGTTCCGCCACTACCCCCTCCTGAATCACCTTGCCAACAGTTCTGAT
CYBRD1AGGGCATCGCCATCATCGTCGTCACCGGCTTCGTCTTCACAGGTCCACGGCAGTCTGT
DCNTCTTTTCAGCAACCCGGTCCAAAGGCTTCTTATTCGGGTGTGATGGATGGCTGTATCTCCCAGTA
DHCR7TGAGCGCCCACCCTCTCGAGGGCTCTGCTTCCCGATTAGTCATAGGGCAAGCAGAAAATTC
E2F8CAGGATACCTAATCCCTCTCACGCAGAAATGTCTCCGCAACCTTGTTCCTGCCCCCAGGGATGAG
EGFR
EPHX2TGAAGCGGGAGGACTTTTTGTAAACGATGAGAGTGTTTTATCCATGCAGCTGAGGCTGGGCTCTTCT
ESR1ATGCCCTTTTGCCGATGCAGCCAAATTGTGTTTGATGGATTAAGACAAAACCGAGTCACATCAGTAATAG
GJA1TGCACAGCCTTTTGATTTCCCCGATCGGGAAGCACCATCTCTAACTCTTCATGTCCAGCAGCTAGTTTTTT
HSPA2CAAGTCAGCAAACACGCAAAACATGCACGAACTAATCAAAAATGCACATTATTCGAGGTTTCTCTTTAATGC
IL6STCAAGCTCCACCTTCCAAAGGACCTCCCTGAATCCATAAAGGCATACCCAGCTTCGTTTTTCCCTACTTTTT
INPP4BTCCGAGCGCTGGATTGCATGAGGCACCAGTTACACAAGGACTTCTTTTCTCTATGCGGCATCCTTCTC
MAPTAGACTATTTGCACACTGCCGCCTGTGGCTCAAAGGATAATATCAAACACACCTTGCTCAGGTCAACTGGTT
MGPCCTTCATATCCCCTCAGCAGAGATGGCCTTCATTAACAGGAGAAATGCAAATTGAGCTCGTGGACAGGCTTA
NEK2TCCTGAACAAATGAATCGCATGTCCTACAAATTTGTTGGCACACCTTATTACATGTAAGCAGCCCAATGACCAGATa
PCNAAAATACTAAAATGCGCCGGCAATGAGGGCGTGAACCTCACCAGTACTTCGGCCCTTAGTGTAATGATATC
PGRTTGATAGAAACGCTGTGAGCTCGAAGCTCATCAAGGCAATTGGTTTACAAGATCATGCAAGTTATCAAGAAGTT
PIPTGCATGGTGGTTAAAACTTACCTCATGCTTGCAGTTCAAACAGAATTGCACCTTGTAGAGGGATGCTGCTA
PLATCAGAAAGTGGCCATGCCACCCTGTGGGAAGACATGAATGCACACTAGGAGGTTGGGCTTTAGCTGAA
PRSS16CACTGCCGGTCACCCACACCACTGAGGAGCACAGAACCTCAACTCGAACTCGGTACATGTCTGATACAA
PTGER3TCGGTCTGCTGGTCTCCGCTCCCTGATTGAAGATCATTTTCAACATCAGACGGCCATTCAGCTTATGG
PTPRTTTGGCTTCTGGACACCCTCACAGAGTTGTGGCCTCTACCATTGCGAGCGGGAACCTTGGGATAG
RACGAP1ACTGAGAATCTCCACCCGGCGCATCGCCAACTGGATAAATTGGAGAATGTGCGGAATCTGTTTGAG
RBBP8ACCGATTCCGCTACATTCCACCCAACAGAAATTGGCTTCCTGCTCAAGAAAACCAACTTCCCAAAAATTCTCT
SCUBE2CTAGAGGGTTCCAGGTCCCATACGTGACATATGTGGATTCAGTTCAAGTCCAATGCCATCTCGAACTATGTCTTCAATGAGT
SEC14L2TGGGAGGCATGCAACGCGTGAGGTCTTACTAAGCAGTCCCATCTCTCGACCGGCACCTGAACTC
SQLETATGCGTCTCCCAAAAGAAGAACACCTCGGCAAGCTTCCTTCCTCCTTCACCTTTAGCAGTTTTCTCCATAGTTTTATATC
STC2TCTCACCTTGACCCTCAGCCAAGACATTTGACAAATTTCCCTTAGGATTCCAGGACGCAGCTTTACCAA
TFAP2BCAACACCACCACTAACAGGCACACGTCGGCATGGACAAGATGTTCTTGACCTCCTTGTCGCCAGTTTTACT
TOP2ACAGATCAGGACCAAGATGGTTCCCACATCATTGAAGACGCTTCGTTATGGCCAGTTGTGATGGATAAAATTAATCAG
TRIM29TGCTGTCTCACTACCGGCCATTCTACGTGGAAATCTGGCAAGCAGACTCAATCCCGTTGCCTTTGTTG
UBE2CTGAACACACATGCTGCCGAGCTCTGCTTCTAGGAGAACCCAACATTGATAGTGTTTCTTGCAGGTACTTCTTAAAAGCT
WNT5ATATTCACATCCCCTCAGTTGCAGTGAATTGCTGTGGCTCTTAATTTATTGCATAATGTTAGTGCTTTTTGCTTTCAAGATCTT

A second alternative for unsupervised selection of possible gene replacement candidates is based on Affymetrix data only. This has the advantage that it can be done solely based on already published data (e.g. from www.ncbi.nlm.nih.gov/geo/). The following table (Tab. 10) lists HG-U133a probe set replacement candidates for the probe sets used in algorithms T1-T5. This is based on training data of these algorithms. The column header contains the gene name and the probe set ID in bold. Then, the 10 best-correlated probe sets are listed, where each table cell contains the probe set ID, the correlation coefficient in brackets and the gene name.

TABLE 10
UBE2CBIRC5DHCR7RACGAP1AURKAPVALBNMUSTC2
202954_at202095_s_at201791_s_at222077_s_at204092_s_at205336_at206023_at203438_at
210052_s_at202954_at201790_s_at218039_at208079_s_at208683_at205347_s_at203439_s_at
(0.82) TPX2(0.82) UBE2C(0.66) DHCR7(0.79) NUSAP1(0.89) STK6(−0.33) CAPN2(0.45) TMSL8(0.88) STC2
202095_s_at218039_at202218_s_at214710_s_at202954_at219682_s_at203764_at212496_s_at
(0.82) BIRC5(0.81) NUSAP1(0.48) FADS2(0.78) CCNB1(0.80) UBE2C(0.30) TBX3(0.45) DLG7(0.52) JMJD2B
218009_s_at218009_s_at202580_x_at203764_at210052_s_at218704_at203554_x_at219440_at
(0.82) PRC1(0.79) PRC1(0.47) FOXM1(0.77) DLG7(0.77) TPX2(0.30) FLJ20315(0.44) PTTG1(0.52) RAI2
203554_x_at202705_at208944_at204026_s_at202095_s_at204962_s_at215867_x_at
(0.82) PTTG1(0.78) CCNB2(−0.46)(0.77) ZWINT(0.77) BIRC5(0.44) CENPA(0.51) CA12
TGFBR2
208079_s_at204962_s_at202954_at218009_s_at203554_x_at204825_at214164_x_at
(0.81) STK6(0.78) CENPA(0.46) UBE2C(0.76) PRC1(0.76) PTTG1(0.43) MELK(0.50) CA12
202705_at203554_x_at209541_at204641_at218009_s_at209714_s_at204541_at
(0.81) CCNB2(0.78) PTTG1(−0.45) IGF1(0.76) NEK2(0.75) PRC1(0.41) CDKN3(0.50) SEC14L2
218039_at208079_s_at201059_at204444_at201292_at219918_s_at203963_at
(0.81) NUSAP1(0.78) STK6(0.45) CTTN(0.75) KIF11(0.73) TOP2A(0.41) ASPM(0.50) CA12
202870_s_at210052_s_at200795_at202705_at214710_s_at207828_s_at212495_at
(0.80) CDC20(0.77) TPX2(−0.45)(0.75) CCNB2(0.73) CCNB1(0.41) CENPF(0.50) JMJD2B
SPARCL1
204092_s_at202580_x_at218009_s_at203362_s_at204962_s_at202705_at208614_s_at
(0.80) STK6(0.77) FOXM1(0.45) PRC1(0.75) MAD2L1(0.73) CENPA(0.41) CCNB2(0.49) FLNB
209408_at204092_s_at218542_at202954_at218039_at219787_s_at213933_at
(0.80) KIF2C(0.77) STK6(0.45) C10orf3(0.75) UBE2C(0.73) NUSAP1(0.40) ECT2(0.49) PTGER3
AZGP1RBBP8IL6STMGPPTGER3CXCL12ABATCDH1
209309_at203344_s_at212196_at202291_s_at213933_at209687_at209460_at201131_s_at
217014_s_at36499_at212195_at201288_at210375_at204955_at209459_s_at201130_s_at
(0.92) AZGP1(0.49) CELSR2(0.85) IL6ST(0.46)(0.74) PTGER3(0.81) SRPX(0.92) ABAT(0.57) CDH1
ARHGDIB
206509_at204029_at204864_s_at219768_at210831_s_at209335_at206527_at221597_s_at
(0.52) PIP(0.45) CELSR2(0.75) IL6ST(0.42) VTCN1(0.74) PTGER3(0.81) DCN(0.63) ABAT(0.40) HSPC171
204541_at208305_at211000_s_at202849_x_at210374_x_at211896_s_at213392_at203350_at
(0.46) SEC14L2(0.45) PGR(0.68) IL6ST(−0.41) GRK6(0.73) PTGER3(0.81) DCN(0.54)(0.38) AP1G1
MGC35048
200670_at205380_at214077_x_at205382_s_at210832_x_at201893_x_at221666_s_at209163_at
(0.45) XBP1(0.43) PDZK1(0.61) MEIS4(0.40) DF(0.73) PTGER3(0.81) DCN(0.49) PYCARD(0.36) CYB561
209368_at203303_at204863_s_at200099_s_at210834_s_at203666_at218016_s_at210239_at
(0.45) EPHX2(0.41) TCTE1L(0.58) IL6ST(0.39) RPS3A(0.55) PTGER3(0.80) CXCL12(0.48) POLR3E(0.35) IRX5
218627_at205280_at202089_s_at221591_s_at210833_at211813_x_at214440_at200942_s_at
(−0.43)(0.38) GLRB(0.57)(−0.37) FAM64A(0.55) PTGER3(0.80) DCN(0.46) NAT1(0.34) HSBP1
FLJ11259SLC39A6
202286_s_at205279_s_at210735_s_at214629_x_at203438_at208747_s_at204981_at209157_at
(0.43)(0.38) GLRB(0.56) CA12(0.37) RTN4(0.49) STC2(0.79) C1S(0.45) SLC22A18(0.34) DNAJA2
TACSTD2
213832_at203685_at200648_s_at200748_s_at203439_s_at203131_at212195_at210715_s_at
(0.42) —(0.38) BCL2(0.52) GLUL(0.37) FTH1(0.46) STC2(0.78) PDGFRA(0.45) IL6ST(0.33) SPINT2
204288_s_at203304_at214552_s_at209408_at212195_at202994_s_at204497_at203219_s_at
(0.41) SORBS2(−0.38) BAMBI(0.52) RABEP1(−0.37) KIF2C(0.41) IL6ST(0.78) FBLN1(0.45) ADCY9(0.33) APRT
202376_at205862_at219197_s_at218726_at217764_s_at208944_at215867_x_at218074_at
(0.41) SERPINA3(0.36) GREB1(0.51) SCUBE2(−0.36)(0.40) RAB31(0.78) TGFBR2(0.45) CA12(0.33) FAM96B
DKFZp762E1312

After selection of a gene or a probe set one has to define a mathematical mapping between the expression values of the gene to replace and those of the new gene. There are several alternatives which are discussed here based on the example “replace delta-Ct values of BIRC5 by RACGAP1”. In the training data the joint distribution of expressions looks like in FIG. 3.

The Pearson correlation coefficient is 0.73.

One approach is to create a mapping function from RACGAP1 to BIRC5 by regression. Linear regression is the first choice and yields in this example


BIRC5=1.22*RACGAP1−2.85.

Using this equation one can easily replace the BIRC5 variable in e.g. algorithm T5 by the right hand side. In other examples robust regression, polynomial regression or univariate nonlinear pre-transformations may be adequate.

The regression method assumes measurement noise on BIRC5, but no noise on RACGAP1. Therefore the mapping is not symmetric with respect to exchangeability of the two variables. A symmetric mapping approach would be based on two univariate z-transformations.


z=(BIRC5−mean(BIRC5))/std(BIRC5) and


z=(RACGAP1−mean(RACGAP1))/std(RACGAP1)


z=(BIRC5−8.09)/1.29=(RACGAP1−8.95)/0.77


BIRC5=1.67*RACGAP1+−6.89

Again, in other examples, other transformations may be adequate: normalization by median and/or mad, nonlinear mappings, or others.