WO2010076322A1 - Prediction of response to taxane/anthracycline-containing chemotherapy in breast cancer - Google Patents

Abstract

A method for predicting a response to and/or benefit from chemotherapy in a patient suffering from cancer comprising the steps of (i) classifying a tumor into at least two classes, (ii) determining in a tumor sample the expression of at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class, (iii) depending on said gene expression, predicting said response and/or benefit; wherein said at least one marker gene comprises a gene selected from the group consisting of TMSL8, ABCC1, EGFR, MVP, ACOX2, HER2/NEU, MYH11, TOB1, AKR1C1, ERBB4, NFKB1A, TOP2A, AKR1C3, ESR1, OLFM1, TOP2B, ALCAM, FRAP1, PGR, TP53, BCL2, GADD45A, PRKAB1, TUBA1A, C16orf45, HIF1A, PTPRC, TUBB, CA12, 1GKC, RACGAP1, UBE2C, CD14, 1KBKB, S100A7, VEGFA, CD247, KRT5, SEPT8, YBX1, CD3D, MAPK3, SLC2A1, CDKN1A, MAPT, SLC7A8, CHPT1, MLPH, SPON1, CXCL13, MMP1, STAT1, CXCL9, MMP7, STC2, DCN, MUC1, STMN1 and combinations thereof.

Description

Title

Prediction of response to taxane/anthracycline-containing chemotherapy in breast cancer

Field of the invention

The present invention relates to methods for prediction of the therapeutic success of cancer therapy.

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. In breast cancer a multitude of treatment options are available which can be applied in addition to the routinely performed surgical removal of the tumour and subsequent radiation of the tumor bed.

The development of malignant tumors is controlled by accumulating genetic abnormalities on the one hand and by epigenetic and host factors on the other hand. Therefore, malignant tumors constitute a complex micro-ecosystem that dependents on the interplay between tumor cells, stromal cells and host inflammatory cells. Several studies have shown that the presence of a lymphocytic infiltrate in cancer tissue is associated with an improved clinical outcome. From animal experiments, there is evidence that the immune system participates in the elimination of tumor cells and the control of tumor growth. Recently, it has been suggested that immunological mechanisms may also be involved in the response to cytotoxic chemotherapy, and that the presence of a low level immunological response might trigger the effects of existing conventional chemotherapy approaches. However, despite this accumulating evidence it has been somewhat difficult to translate the results from basic tumor immunology into clinically relevant parameters that can be used in the context of clinical trials. Neoadjuvant chemotherapy of early breast cancer leads to high clinical response rates of 70-90%. However, in the majority of clinical responders, the pathological assessment of the tumor after surgery reveals the presence of residual tumor cell foci. A complete absence of residual invasive tumor, the so- called pathological complete response (pCR) is observed only in 10-25% of patients. The pCR is a surrogate marker for disease-free survival and a strong indicator of benefit from chemotherapy .

As the neoadjuvant chemotherapy constitutes an in vivo chemoresistance test, it is an excellent basis for the analysis of predictive biological factors in pretherapeutic core biopsies, in order to identify those patients that would benefit most from chemotherapy.

For patients with a low probability of response, other therapeutic approaches might be considered. Accepted parameters linked to response to neoadjuvant chemotherapy are hormone receptor status as well as tumor grade. The present invention is based on the hypothesis that the presence of an inflammatory lymphocyte-mediated response to tumor cells may predict the response to neoadjuvant chemotherapy.

Chemotherapy may be applied postoperative, i.e. in the adjuvant setting or preoperative, that is in the neoadjuvant setting in which patients receive several cycles of drug treatment over a limited period of time, before remaining tumor cells are removed by surgery. Neoadjuvant chemotherapy is used for patients with large tumors and locally advanced breast cancer. Primary goal is a reduction of tumor size in order to increase the possibility of breast-conserving treatment .

Accurate prediction of the response of a breast cancer patient to neoadjuvant chemotherapy could help to select the most efficient and appropriate drug for breast cancer treatment in the patient, providing a means of individualized patient care. Thus, there is a need in the art for reliable methods of predicting the response of breast cancer patients to neoadjuvant chemotherapy.

Progress has been made in the treatment of breast cancer with cytotoxic chemotherapy. However, resistance against different treatment modalities is a major reason for therapy failure and bad prognosis of advanced disease. Following neo-adjuvant primary chemotherapy, for example, 20-50% of patients do not respond to therapy, i.e. they demonstrate either no or minor histopathological changes (NC) of the neoplastic disease or even a tumor growth after being submitted to said given mode of treatment.

In contrast, only 15 - 25% of patients achieve pathological complete remission (pCR) which means a complete eradication of invasive tumor cells as assessed by histopathological examination, and which is associated with prolonged survival.

For patients with a low probability of response, other therapeutic approaches might be considered. Accepted parameters linked to response to neoadjuvant chemotherapy are hormone receptor status as well as tumor grade.

Thus, there is a substantial number of patients who suffer from side effects of neoadjuvant chemotherapy without clinical benefit of the same. Up to now, there is no reliable marker or diagnostic test for prediction of response to cytotoxic chemotherapy. This means that, to date, the only way to find out whether or not a breast cancer patient is responsive to neo-adjuvant primary chemotherapy, is to administer such therapy to the patient, e.g. by an "in vivo" chemoresistance test by two cycles of chemotherapy. In case the patient does not respond to the therapy the physician may then decide which second line therapy (e.g., different chemotherapy, endocrine therapy, anti-angiogenesis therapy, radiation therapy or surgery) is to be applied.

Definitions

The term "response", as used herein, relates to the reaction of an individual under a defined therapy. Reactions as used in this document can for example be beneficial or adverse. Possible reactions include prolongation or shortening of time to local and/or distant recurrence, prolongation or shortening of time to death, prolongation or shortening of disease progression, and prolongation or shortening of time to metastasis in an adjuvant or neoadjuvant setting. In a neoadjuvant setting, reactions to therapy additionally include the shrinkage, growth, or absence of change of the primary tumor within a given time frame and is usually measured as a quantification of change, usually given as a percentage, e.g. diameter or volume, or as a class as, for example, defined by WHO.

The term "pathological complete response" (pCR) , as used herein, relates to a complete disappearance or absence of invasive tumor cells in the breast and/or lymph nodes as assessed by a histopathological examination of the surgical specimen following neoadjuvant chemotherapy.

The term "tissue response" (TR) , as used herein, relates to at least limited response with residual invasive tumor <0.5cm as assessed by a histopathological examination of the surgical specimen following neoadjuvant chemotherapy. "No tissue response" was defined as no changes or limited cellular response (sclerosis, resorption, inflammation, zytopathic changes) in the tumor.

The term "prognosis", as used herein, relates to an individual assessment of the malignancy of a tumor, or to the expected response if there is no drug therapy. In contrast thereto, the term "prediction" relates to an individual assessment of the malignancy of a tumor, or to the expected response if the therapy contains a drug in comparison to the malignancy or response without this drug.

The term "prognosis under therapy" relates to an individual assessment of the malignancy of a tumor, or to the expected response if there is any drug therapy without considering malignancy or response without this drug.

The term "response marker" relates to a marker which can be used to predict the pathological and/or clinical response and/or clinical outcome of a patient towards a given treatment .

The term "therapy modality", "therapy mode", "regimen" as well as "therapy" refers to a timely sequential or simultaneous administration of anti-tumor, and/or anti vascular, and/or anti stroma, and/or immune stimulating or suppressive, 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 each of the single agents, 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. A "taxane/anthracycline-containing chemotherapy" is a therapy modality comprising the administration of taxane and/or anthracycline and therapeutically effective derivates thereof . The term "neoadjuvant chemotherapy" relates to a preoperative therapy regimen consisting of a panel of hormonal, chemotherapeutic and/or antibody agents, which is aimed to shrink the primary tumor, thereby rendering local therapy

(surgery or radiotherapy) less destructive or more effective, enabling breast conserving surgery and evaluation of responsiveness of tumor sensitivity towards specific agents in vivo .

The terms "sample", as used herein, refer to a sample obtained from a patient. 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), circulating cells (e.g. stem cells or endothelial cells in the blood, tissue, core or fine needle biopsy samples, cell-containing body fluids, free floating nucleic acids, urine, stool, peritoneal fluid, and pleural fluid, liquor cerebrospinalis, tear fluid, or cells there from. Biological samples 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" is a sample containing tumor material e.g. tissue material from a neoplastic lesion taken by aspiration or puncture, excision or by any other surgical method leading to biopsy or resected cellular material, including preserved material such as fresh frozen material, formalin fixed material, paraffin embedded material and the like. Such a biological sample may comprise cells obtained from a 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, gynecological fluids, or urine but not limited to these fluids. The term "marker" or "biomarker" refers to a biological molecule, e.g., a nucleic acid, peptide, protein, hormone, etc., whose presence or concentration can be detected and correlated with a known condition, such as a disease state.

The term "marker gene", as used herein, refers to a differentially expressed gene whose expression pattern may be utilized as part of a predictive, prognostic or diagnostic process in healthy conditions, premalignant disease status, malignant neoplasia or cancer evaluation, or which, alternatively, may be used in methods for identifying compounds useful for the treatment or prevention of malignant neoplasia and head and neck, colon or breast cancer in particular. A marker gene may also have the characteristics of a target gene.

The term "expression level" refers, e.g., to a determined level of gene expression. The term "pattern of expression levels" refers to a determined level of gene expression compared either to a reference gene (e.g. housekeeper or inversely regulated genes) or to a computed average expression value (e.g. in DNA-chip analyses) . A pattern is not limited to the comparison of two genes but is more related to multiple comparisons of genes to reference genes or samples. A certain "pattern of expression levels" may also result and be determined by comparison and measurement of several genes disclosed hereafter and display the relative abundance of these transcripts to each other.

The term "determining the expression level of a gene/protein on a non-protein basis" relates to methods which are not restricted to the secondary gene translation products, i.e proteins, but on other levels of the gene expression, like the mRNA, premRNA and genomic DNA structures.

Alternatively, a differentially expressed gene disclosed herein may be used in methods for identifying reagents and compounds and uses of these reagents and compounds for the treatment of cancer as well as methods of treatment. The differential regulation of the gene is not limited to a specific cancer cell type or clone, but rather displays the interplay of cancer cells, muscle cells, stromal cells, connective tissue cells, other epithelial cells, endothelial cells of blood vessels as well as cells of the immune system (e.g. lymphocytes, macrophages, killer cells) .

The terms "modulated" or "modulation" or "regulated" or "regulation" and "differentially regulated" or

"differentially expressed" as used herein refer to both upregulation (i.e., activation or stimulation (e.g., by agonizing or potentiating) and down regulation [i.e., inhibition or suppression (e.g., by antagonizing, decreasing or inhibiting) ] .

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 or diseased individuals, serving as a reference group.

"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 and /or morpholinos are also comprised for usage as primers and/or probes. "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".

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. The regions in a microarray have typical dimensions, e.g., diameters, in the range of between about 10-250 μm, and are separated from other regions in the array by about the same distance. A "protein array" refers to an array containing polypeptide probes or protein probes which can be in native form or denatured. An "antibody array" refers to an array containing antibodies which include but are not limited to monoclonal antibodies (e.g. from a mouse), chimeric antibodies, humanized antibodies or phage antibodies and single chain antibodies as well as fragments from antibodies .

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 "minor groove binder'', acronym: MGB) 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 (Hence its name: Taq polymerase + PacMan) . 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.

The term "immunohistochemistry" or IHC refers to the process of localizing proteins in cells of a tissue section exploiting the principle of antibodies binding specifically to antigens in biological tissues. Immunohistochemical staining is widely used in the diagnosis and treatment of cancer. Specific molecular markers are characteristic of particular cancer types. IHC is also widely used in basic research to understand the distribution and localization of biomarkers in different parts of a tissue. "Prediction of recurrence" or "prediction of therapeutic success" does refer to the methods described in this invention. Wherein a tumor specimen is analyzed for its gene expression and furthermore classified based on correlation of the expression pattern to known ones from reference samples. This classification may either result in the statement that such given tumor will develop recurrence or will not achieve a pathological complete response or a tissue response following neoadjuvant chemotherapy and therefore is considered as a "non-responding" tumor to the given therapy, or may result in a classification as a tumor with a prolonged disease free post therapy time or as tumor that will achieve a pathological complete response or a tissue response.

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 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 above. 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.

The term "determining the protein level", as used herein, refers to methods which allow the quantitative and/or qualitative determination of one or more proteins in a sample. These methods include, among others, protein purification, including ultracentrifugation, precipitation and chromatography, as well as protein analysis and determination, including immunohistochemistry, immunofluorescence, ELISA (enzyme linked immuno assay) , RIA (radioimmunoassay) or the use of protein microarrays, two- hybrid screening, blotting methods including western blot, one- and two dimensional gelelectrophoresis, isoelectric focusing as well as methods being based on mass spectrometry like MALDI-TOF and the like.

The term "nucleic acid molecule" is intended to indicate any single- or double stranded nucleic acid molecule comprising DNA (cDNA and/or genomic DNA) , RNA (preferably mRNA) , PNA, LNA and/or Morpholino, or fractions, derivatives, fragments or analogues thereof.

Object of the invention

It is one object of the present invention to provide a method which allows for a better prediction a response of a patient suffering from or at risk of developing a neoplastic disease towards at least one given mode of treatment.

It is another object of the present invention to avoid unnecessary neoadjuvant chemotherapy in patients suffering from a neoplastic disease.

It is another object of the present invention to offer a more robust and specific diagnostic assay system than conventional immunohistochemistry for clinical routine fixed tissue samples that better helps the physician to select individualized treatment modalities.

In a more preferred embodiment the disclosed method can be used to select a suitable therapy for a neoplastic disease, particularly breast cancers.

It is another object of the present invention to detect new targets for newly available targeted drugs, or to determine drugs yet to be developed.

Summary of the invention

Before the invention is described in detail, it is to be understood that this invention is not limited to the particular component parts of the devices described or process steps of the methods described as such devices and methods may vary. It is also to be understood that the terminology used herein is for purposes of describing particular embodiments only, and is not intended to be limiting. It must be noted that, as used in the specification and the appended claims, the singular forms "a, " "an" and "the" include singular and/or plural referents unless the context clearly dictates otherwise. It is moreover to be understood that, in case parameter ranges are given which are delimited by numeric values, the ranges are deemed to include these limitation values.

The above problems are solved by methods and means provided by the invention.

The invention relates to a method for predicting a response to and/or benefit from chemotherapy in a patient suffering from cancer comprising the steps of a) classifying a tumor into at least two classes said at least two classes being selected from the group consisting of a a Her 2/neu negative, ESR negative (basal / triple negative) class of tumors, and a Her 2/neu negative, ESR positive (luminal class) class of tumors,

b) determining in a tumor sample the expression of at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class, and

c) depending on said gene expression, predicting said response and/or benefit;

wherein said at least one marker gene comprises a plurality of genes for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR positive (luminal class) tumors said plurality of genes comprising the genes CD3D, CXCL9, UBE2C, and, optionally ESRl; or

wherein said at least one marker gene comprises a plurality of genes for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR negative (basal or triple negative class) tumors, said plurality of genes comprising the genes STMNl, HER2/NEU, and NFKBIA. According to a further aspect of the invention the plurality of genes is used for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR positive (luminal class) tumors further comprising the gene ESRl.

In more general terms, the invention provides a method for predicting a response to and/or benefit from chemotherapy in a patient suffering from cancer comprising the steps of a) classifying a tumor as belonging to at least one class,

b) determining in a tumor sample the expression of at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class; and

c) depending on said gene expression, predicting said response and/or benefit;

wherein said at least one marker gene comprises a gene selected from the group consisting of TMSL8, ABCCl, EGFR, MVP, ACOX2 , HER2/NEU, MYHIl, TOBl, AKRlCl, ERBB4, NFKBIA, TOP2A, AKR1C3, ESRl, OLFMl, TOP2B, ALCAM, FRAPl, PGR, TP53, BCL2, GADD45A, PRKABl, TUBAlA, C16orf45, HIFlA, PTPRC, TUBB, CA12, IGKC, RACGAPl, UBE2C, CD14, IKBKB, S100A7, VEGFA, CD247, KRT5, SEPT8, YBXl, CD3D, MAPK3, SLC2A1, CDKNlA, MAPT, SLC7A8, CHPTl, MLPH, SPONl, CXCL13, MMPl, STATl, CXCL9, MMP7, STC2, DCN, MUCl, STMNl and combinations thereof.

The methods of the invention particularly suited for predicting a response to cytotoxic chemotherapy, preferably taxane/anthracycline-containing chemotherapy, preferably in the neodajuvant mode.

According to an aspect of the more general embodiment of the invention said tumor is classified into HER2/NEU positive or negative, Luminal and Basal / triple negative classes. According to an aspect of the invention, said at least one marker gene for predicting a response to and/or benefit from chemotherapy in Her 2/neu positive tumors is selected from the group consisting of ERBB4, CHPTl, BCL2, MLPH, SPONl and combinations thereof.

According to an aspect of the invention said classification is performed by determining in a tumor sample the expression of at least one gene indicative for each class as described in this disclosure and depending on said gene expression, classifying the tumor.

According to an aspect of the invention said gene expression is determined on a RNA level by a PCR based method and/or a microarray based method. Gene expression may further be determined at a protein level or non-protein level, by any suitable method, e.g. hybridization based methods or array based methods .

According to a further aspect of the invention for Her2/neu positive tumors said at least one marker gene is selected from the group consisting of ERBB4, CHPTl, BCL2 MLPH, and the combinations of CHPT1/ERBB4, and CHPTl /SPONl .

According to a further aspect of the invention for luminal tumors said at least one marker gene is selected from the group consisting of CXCL9, MUCl, IGKC, CD3Z, and the combinations of CD3D/MUC1, FRAPl /MUCl, ACOX2/CD3D, ACOX2/CD3Z, and AKR1C3/EGFR.

According to a further aspect of the invention for of basal / triple negative tumors said at least one marker gene is selected from the group consisting of TMSL8, ERBB2 (HER2/NEU), MUCl and the combinations of STMNl, HER2/NEU/STMN1, HER2/NEU/TMSL8 , HER2/NEU/NFKBIA.

According to a further aspect of the invention the expression level of no more than five marker genes are determined in a given class, preferably no more than 4, 3, 2, or 1 marker genes. A low number of genes is preferred, as it reduces the amount of measurements needed to obtain a predictive result. Preferred embodiments of the invention allow a predictive determination to be made using just 5, 4, 3, 2, or even 1 marker gene (s) .

According to a further aspect of the invention the expression level of said 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. The expression level of said at least on marker gene may be determined relative to a combination of several reference genes. Preferred reference genes are RPL37A CALM2, and OAZl.

According to a further aspect of the invention, the gene TMSL8 has been determined as a new marker which is predictive for pCR in all Tumors (tables 4 and 5) , especially in ESRl negative tumors (table 8), in triple-negative / basal tumors (table 12) and in Her2/neu positive tumors (table 14) .

According to a further aspect of the invention, the expression levels of a plurality of marker genes are mathematically combined to give a score indicative of a response to and/or benefit from chemotherapy. This mathematical combination may include, but is not limited to summation, weighted summation, correlation coefficients, discriminant functions, and statistical functions. Gene expression values of marker genes may be used relative values normalized to one or more reference genes.

The invention further provides a kit for performing the method of any of the preceding claims comprising at least one probe specific for a gene or gene product for each at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class. The invention further provides a use of the kit described above for performing the methods according to the invention

The following genes and gene combinations are especially predictive for Her2/neu positive tumors:

Most robustly selected single genes (logistic regression model) : ERBB4 CHPTl BCL2 MLPH

Most robustly selected gene pairs (bivariate logistic regression model) : CHPT1/ERBB4 CHPT1/SPON1

The following genes and gene combinations are especially predictive for luminal tumors:

Most robustly selected single genes (logistic regression model) : CD3D

CXCL9

MUCl

IGKC

CD3Z

Most robustly selected gene pairs (bivariate logistic regression model) :

CD3D/MUC1

FRAPl /MUCl ACOX2/CD3D

ACOX2/CD3Z

AKR1C3/EGFR The following genes and gene combinations are especially predictive for basal / triple negative tumors:

Most robustly selected single genes (logistic regression model) :

TMSL8

HER2/NEU

MUCl

STMNl

Most robustly selected gene pairs (bivariate logistic regression model) :

HER2/NEU/STMN1

HER2/NEU/TMSL8 HER2/NEU/NFKBIA

According to a preferred embodiment of the invention, the combination of genes comprising CD3D, CXCL9, and UBE2C are used for the prediction of response to chemotherapy in luminal tumors. This combination of marker genes allows for a particularly reliable response to chemotherapy.

According to a preferred embodiment of the invention, the combination of genes comprising CD3D, CXCL9, ESRl, and UBE2C are used for the predicition of response to chemotherapy in luminal tumors. This combination of marker genes allows for a particularly reliable response to chemotherapy.

According to a preferred embodiment of the invention, the combination of genes comprising STMNl, HER2/NEU, NFKBIA are used for the prediction of response to chemotherapy in basal / triple negative tumors. This combination of marker genes allows for a particularly reliable response to chemotherapy. Description of the invention

Additional details, features, characteristics and advantages of the object of the invention are disclosed in the following description of the respective figures and examples, which, in an exemplary fashion, show preferred embodiments of the present invention. However, these drawings should by no means be understood as to limit the scope of the invention.

Fig. 1: schematically shows the basic classification of the finding cohort in molecular subgroups.

Fig. 2: schematically shows a block diagram of an exemplary embodiment of the inventive method including exemplary cutoff values for classifying tumors according to the basic classification shown in figure 1.

Fig.3: shows a Receiver Operator Characteristics Curve (ROC) for the algorithm NLRS for luminal tumors in a training cohort (top panel) and a validation cohort (bottom panel) .

Fig. 4: shows the sensitivity and specificity for an exemplary cutoff value of -3 for the algorithm NLRS for luminal tumors in a training cohort (top panel) and a validation cohort (bottom panel) .

Fig. 5: shows Receiver Operator Characteristics (ROC) for the algorithm NTRS for triple negative (basal) tumors in a training cohort (top panel) and a validation cohort (bottom panel) .

Fig. 6: shows the sensitivity and specificity for an exemplary cutoff value of -0.2 for the algorithm NTRS for triple negative (basal) tumors.

Fig. 7: shows a decision tree for the algorithm C_NLRS for luminal tumors . Fig. 8 to 13 show additional data regarding the performance of exemplary best model algorithms in different tumor classes. Top panels each show the area under curve (AUC) of an R.O.C. curve (left scale) and the Bayesian information criterion (BIC, right scale) relative of the number of genes used to predict pCR. Middle panels show the probability of pCR relative to the selected cutoff value. Bottom panels each show the AUC.

An embodiment of the invention is based upon a classification of tumor samples according to the diagram shown in Figure 1 : The tumor of the patient is classified according to Her2/neu (also referred to as ERBB2) status into Her2/neu positive or negative tumors and Her2/neu negative tumors are further classified into estrogen receptor (also referred to as "ER" or "ESR) negative tumors (so called "triple negative" or "basal" class of tumors) or Her2/neu negative ER positive tumors (so called "luminal" class of tumors) . It is further based on the finding that the degree of infiltration of a tumor with lymphocytes is an indication of the response of the tumor to therapy.

For each of these classes (Her2/neu positive, basal / triple negative and luminal class) , the inventors have identified genes which are differentially expressed in patients which are responsive to chemotherapy vs. nonresponsive patients as assessed by pathological complete response (pCR) or non- response. Determining expression status of one of these genes (univariate classifier) or a plurality of these genes (multivariate classifier) thus allows prediction of a response to chemotherapy.

The following genes were identified as predictive and measured by quantitative Polymerase Chain Reaction (PCR) : TMSL8, ABCCl, EGFR, MVP, ACOX2, HER2/NEU, MYHIl, TOBl,

AKRlCl, ERBB4, NFKBIA, TOP2A, AKR1C3, ESRl, OLFMl, TOP2B, ALCAM, FRAPl, PGR, TP53, BCL2, GADD45A, PRKABl, TUBAlA, C16orf45, HIFlA, PTPRC, TUBB, CA12, IGKC, RACGAPl, UBE2C, CD14, IKBKB, S100A7, VEGFA, CD247, KRT5, SEPT8, YBXl, CD3D, MAPK3, SLC2A1, CDKNlA, MAPT, SLC7A8, CHPTl, MLPH, SPONl, CXCL13, MMPl, STATl, CXCL9, MMP7, STC2, DCN, MUCl, STMNl, and, as reference point expression of housekeeping genes CALM 2, OAZl, RPL37A was measured..

Classifier Training was performed as follows:

For each of the groups, identify informative genes (or combinations of genes) using logistic regression analysis. - Single extreme values (e.g. outliers) can adversely impact feature selection => Repeat discovery for various subsets of training data to assess robustness. Random selection of m samples (out of n original samples) with putting back was used for training. - In each discovery step, best genes (significance of regression coefficient less than a predetermined cutoff value, e.g. 5%) are selected.

- To all of these genes, other genes are combined to achieve even better risk scores (partially greedy approach) . - Quality of a combination was assessed by max p value of all regression coefficients (excluding p value of constant) .

- Summarize how often each gene/gene combination was selected.

The genes or gene combinations identified by classifier training were then validated in different patient cohorts.

TRAINING COHORT - HISTOLOGICAL EVALUATION OF GEPARDUO CASES As a first step, the percentage of mononuclear cells in the tumor stroma (stromal lymphocytes (str-Ly) ) as well as the percentage of mononuclear cells directly infiltrating the tumor cell nests (intratumoral lymphocytes (iTu-Ly) ) in 218 breast cancer core biopsies of the GeparDuo cohort were investigated (Table 1) . Many pCR cases showed a similar histology with a prominent mononuclear infiltrate in the tumor stroma and/or in the tumor cell nests. Based on experience from diagnostic histopathology, a prominent inflammatory infiltrate is quite unusual in the majority of breast cancer cases. For comparison, a typical case without such an infiltrate did not have a pCR.

A significant correlation between str-Ly and iTu-Ly was observed (Pearson correlation coefficient 0.61, p<0.0005) . A pathological complete response was observed primarily in those tumors with increased lymphocyte infiltrate. The overall pCR rate was 12.8% and could be increased to 18.8 % in patients receiving the more efficient chemotherapy AC/Doc. High grade tumors as a subgroup had a pCR rate of 26%. The pCR rate for the cases with increased iTu-Ly (>10%) was 31%. To combine both lymphocyte parameters, the subgroup of lymphocyte-predominant breast cancer (LPBC) was defined as those cases with more than 60% either stromal or intratumoral lymphocytes. This subgroup of LPBC tumors had a pCR rate of 42.7%.

To support the hypothesis in the context of other known clinicopathological factors, a multivariate analysis including the new immunological parameter iTu-Ly was performed together with established clinicopathological parameters that are related to pCR in the GeparDuo study. These parameters were hormone receptor status, grading and therapy arm. As shown in table 1, the percentage of iTu-Ly was a significant independent parameter (p=0.013) with an odds ratio for pCR of 1.03 (95% CI 1.01-1.06) per 1% increase in lymphocyte infiltrate. Other already published parameters were hormone receptor status, grade and therapy arm. This multivariate analysis shows that the combination of different parameters could be used to define subgroups with a high response rate, and that adding immunological parameters is useful for the better definition of this subgroup. Table 1: Definitions for histopathological evaluation

Parameter Definition Comments lntratumoral lymphocytes Percentage of tumor cell Only those mononuclear cells (iTu-Ly) '•^/- nests with intraepithelial that are within the epithelium mononuclear cells. of the invasive tumor cell nests are evaluated. Any infiltrate of intraductal carcinoma is not included. The infiltrate must consist of mononuclear cells, any granulocyte infiltrate in the area of tumor necrosis is not included.

Stromal lymphocytes Percentage of tumor Only tumor stroma of the (str-Ly) "^>'- stroma with mononuclear invasive carcinoma is inflammatory cells. included, stromal infiltrate adjacent to intraductal carcinoma is not included. Furthermore, any inflammatory infiltrate around the normal breast tissue adjacent to the tumor is not included.

Lymphocyte-predominant Those carcinomas with Although LPBC is used as a breast cancer (LPBC) either more than 60% subgroup of carcinomas for intratumoral lymphocytes this evaluation, it should be or more than 60% stromal noted that the data suggests lymphocytes. that the response to The designation indicates chemotherapy is dependent that in those tumors the on the lymphocytic infiltrate as lymphocytes are the a continuous parameter, as predominant host cells seen in the logistic regression within the as well as in comparison of microecosystem of the subgroups with different tumor. percentages of lymphocytes. Therefore LPBC should be used as a working category to indicate an increased odds ratio for pathological complete response rather than a separate tumor entity.

No lymphocyte infiltrate No detectable lymphocytes in tumor cell nests and tumor stromal.

Focal infiltrate Detectable lymphocytes, but less than 60% of str- or iTu-L VALIDATION COHORT - TUMOR BIOPSIES FROM GEPARTRIO TRIAL The association between str-Ly and iTu-Ly and pathological complete response was validated using 840 cases with available tissue from a separate independent clinical study, the GeparTrio trial. For the evaluation of the H&E sections from this clinical trial, a completely blinded setup was used. The evaluating pathologists did not have any access to clinical data at the time of analysis. The database was opened for analysis only after completion of the histological reevaluation .

The presence of stromal and intratumoral lymphocytes was a strong predictor of pCR in univariate (p<0.0005) and multivariate logistic regression (p<0.0005) . Other independent parameters were patient age (p=0.023), and negative hormone receptor status (p<0.0005), while grading was of borderline significance (p=0.064) . The stromal lymphocytes were significantly correlated with iTu-Ly (Pearson correlation coefficient 0.80, p<0.0005) .

Table 2 Validation cohort (GeparTrio) - Factors associated with a pathological complete response in the GeparTrio cohort in univariate and multivariate analysis. Results of univariate and multivariate logistic regression are shown. The parameter str-Ly is not included in multivariate analysis as it is correlated with iTu-Ly. In a separate multivariate analysis the parameter str-Ly is significant as well (OR 1.02 (1.01- 1.02), p<0.0005, data not shown)

Intratumoral 1.03 (1.02-1.04) <0.0005 1.02 (1.01-1.03) <0.0005 lymphocytes (iTu- Ly) (%) Stromal 1.02 (1.03-1.03) <0.0005 lymphocytes (str-Ly)

Age group 1.50 (1.04-2.15) 0.028 1.71 (1.08-2.71 ) 0.023

< 50 years ≥ 50 years

Tumor type 2.40 (1.22-4.71 ) 0.011 1.22 (0.51-2.89) 0.652

Ducta I/other

Lobular Tumor grade 2.91 (1.94-4.35) <0.0005 1.58 (0.97-2.58) 0.064

G3

G1 or 2 ER/PR Status 6.00 (3.97-9.08) <0.0005 3.71 (2.25-6-11 ) <0.0005

ER-/PR- ER+ and/or PR+

Table 2 Validation cohort

In this first analysis, the lymphocytic infiltrate was evaluated as a continuous parameter. To visualize the odds ratio for comparison with other predictive parameters, an evaluation of grouped iTu-Ly and str-Ly as well as known predictive parameters was performed. The odds ratio for pCR increases with the extent of iTu-Ly and str-Ly, with a maximal OR of 13.39 (95% CI 6.1-29.37, p<0.0005) for tumors with more than 60% of iTu-Ly in tumor cell nests. Both parameters were combined in the subgroup of lymphocyte- predominant breast cancer (LPBC) as those cases with more than 60% of either iTu-Ly or str-Ly. This subgroup consisted of 12% of all samples and had a pCR rate of 40%, which is very similar to the result obtained in the GeparDuo cohort. In contrast, those 16.4% of cases without any infiltrating lymphocytes had a pCR rate of only 7.2%. The OR for immunological parameters in tumor tissue was higher than for the established predictors grade and hormone receptor status.

EXPRESSION OF LYMPHOCYTE MRNAS AND CHEMOKINES IN TUMOR TISSUE CORRELATES WITH LYMPHOCYTE INFILTRATE AND PATHOLOGICAL COMPLETE REMISSION

As a next step, in a subset of 136 cases from the GeparTrio cohort the type of leukocyte infiltration and what molecular factors might participate in the constitution of the inflammatory infiltrate was investigated by quantitative PCR. As a parameter for general leukocyte infiltration, the pan- leukocyte marker PTPRC (CD45)was used, T cell infiltration was assessed by analysis of CD3D and CD247 (CD3z) , B-cells were analysed by investigation of IGKC. CXCL9 and CXCL13 were measured as known chemoattractants for T and B lymphocytes, respectively. A hierarchical cluster analysis and a heat map of the expression data showed a co-regulation of the lymphocyte markers and an association of all of those markers with the achievement of a pCR and the presence of a lymphocyte infiltration. This indicates that the infiltration consisted of both, T and B cells. Moreover, the relative mRNA expression level of the lymphocyte markers significantly increased with the proportion of tumor infiltrating cells. The expression levels of the B and T cell markers were 2- to 12-fold higher in samples from patients achieving pCR in comparison with those who did not achieve pCR (Figure x) . Finally, logistic regression analysis showed a significant association between the T cell markers CD3D, CXCL9 and CD247 whereas the B cell markers did not.

The inventors show by using two large independent cohorts of samples from neo-adjuvant clinical trials that it is possible to identify a distinct inflammatory subgroup of tumors by standard H&E histopathological analysis of pretherapeutic core biopsies. This subgroup of tumors is characterized by a lymphocytic infiltrate in the tumor tissue and a particular strong response to cytotoxic chemotherapy. This tumor subtype may be called "lymphocyte predominant breast cancer" (LPBC) . This description should indicate that it is a tumor type in which lymphocytes and/or plasma cells are the predominant cell type in the tumor stroma and that the inflammatory cells show a tight association to the tumor cells. While primary analysis was performed solely by careful morphological analysis of standard H&E sections, molecular methods were used to validate the link between T and B cell infiltrate, H&E morphology and pathological response. In addition, evidence is provided that chemoattractant mediators for B and T cells are produced in the tumor tissue and are linked to chemotherapy response.

The presence of lymphocytic infiltrates in some cases of breast cancer has been recognized for a long time and has been used to define the subtype of medullary breast cancer (MBC) . In addition to a lymphocyte infiltrate, MBCs are defined by a syncytial growth pattern as well as circumscribed margins. However, this subtype has been shown to be associated with a particular high inter-observer variability on histopathological evaluation in several studies. It should be emphasized that the subtype of LPBC is clearly distinct from MBC. Only 1.4% of the cases in the GeparDuo study and 0.3% of the cases in the GeparTrio study met the criteria for MBC. In contrast an increased intratumoral lymphocyte infiltrate (>10%) was observed in 51% of cases in the GeparTrio study, and 12% of cases were LPBC. Therefore, the lymphocyte infiltrate is observed in a much larger subset of cases than the MBC group.

To our knowledge, this is the first time that a strong association between lymphocytic infiltrate and chemotherapy response is described in a large set of more than 1000 samples from two clinical breast cancer studies. There are results from several previous investigations that point to a contribution of lymphocytes to chemotherapy response. Gianni et al . have investigated gene expression in 89 tumors treated with neoadjuvant chemotherapy, of which 11 had a pCR. Their gene expression analysis also showed that immune-related genes, such as CD3, are increased in tumors with good response to chemotherapy. For the humoral immune system, a recent microarray-based analysis has shown that a B cell typical gene signature had an independent prognostic role in node-negative breast cancer. Several studies have suggested possible mechanisms of tumor-immune interaction in the response to chemotherapy. Apetoh et al . have shown that a polymorphism of the toll-like receptor-4 is an independent prognostic factor in response to chemotherapy. Ladoire et al . have shown an association of therapy response and a reduction of Foxp3 immunosuppressive T cells in 56 tumors treated with neoadjuvant chemotherapy. These regulatory T cells are a key contributor to the maintenance of immune tolerance. The presence of FoxP3 regulatory T cells is increased in blood and tumor tissue of cancer patients and has been linked to poor prognosis in breast cancer. Interestingly, a decrease in regulatory T cells has also been found to be associated with response to Herceptin therapy, suggesting a contribution of immune-mechanisms to different types of anti-tumor therapy. The chemokine CXCL9 is involved in the regulation of tumor growth and metastasis in animal models .

These investigations have led to the hypothesis that the pretreatment host response may enhance the ability of chemotherapy to eliminate cancer cells. This hypothesis is strongly supported by the present analysis.

The discovery that lymphocyte infiltrates associated with increased response to chemotherapy is interesting in the light of other studies that have shown that parameters that are relevant for immune system function are also involved in response to chemotherapy. It may be speculated that the destruction of tumor cells by chemotherapeutic agents may release tumor-associated antigens. This may trigger an immune response directed against the tumor cells which will be particularly strong in those cases where a sensitization of the immune system against some tumor antigens is present before the onset of chemotherapy. Therefore, the chemotherapy may act as a functional immunotherapy in those tumor types and the combination of chemotherapeutic destruction of tumor cells as well as increased immune response may lead to a pathological complete remission. At present, it is not clear if this hypothesis may be the basis for further therapeutic approaches that may use a combination of stimulation of immune responses with classical chemotherapy to improve the rates of pathological complete remission in neoadjuvant chemotherapy .

As a conclusion, the inventors established and independently validated that the presence of a mononuclear infiltrate in tumor stroma as well as within the tumor cells nests is associated with an increased response to neo-adjuvant chemotherapy in univariate and multivariate analysis. This might be the basis for new therapeutic approaches of the combination of conventional chemotherapy with immune therapy, to use the synergies between both types of therapy.

In the diagnostic setting, iTu-Ly and str-Ly are promising additional parameters for routine diagnostic reporting in combination with grading and hormone receptor status. The analysis of the inflammatory infiltrate in histopathological analysis of breast cancer core biopsies gives useful information to oncologists to identify the subgroup of patients with an increased chance of response to chemotherapy.

Exemplary results of finding for specific genes in the GeparTrio cohort are shown in the table 3 below.

Single genes and two-gene combinations m ERBB2, luminal and basal tumors

Table 3: Single genes and gene combinations predictive in various tumor classes.

In this table 3 "class" designates the respective tumor class; "Objective" designates whether the algorithm was obtained with respect to pathological complete response or tissue response; "Gene" indicates the name of the marker gene used; "model" indicates the algorithm used to obtain the score which indicates the probability of achieving the objective in the respective sample, p value indicates the p value of the respective gene and AUC indicates the "area under curve" for the respective receiver operator curve associated with the respective algorithm given under "model".

Briefly, for generation of an algorithm predictive of e.g. pCR in luminal samples, training was performed in GeparTrio luminal samples by using uni-, bi- and trivariate logistic regression. Receiver operator characteristics were assessed for the top ranked gene combinations. The models with highest area under the curve (AUC) represent preferred marker genes or combinations of marker genes. It was observed that the logistic regression models showed the effect that CD3D could be switched for another T cellular immune response marker CXCL9 without sacrificing overall classifier performance. Therefore, a T cellular immune metagene (TIMG) can be constructed using the first principal component of a principal component analysis (PCA) involving CD3D and CXCL9 in order to improve robustness of algorithms. A positive coefficient or score indicates that increased expression of a gene is associated with a high probability of pCR, whereas a negative coefficient indicates an inverse association of the gene expression value with the probability of pCR. For a gene associated with a high probability of pCR higher scores therefore indicate a higher likelihood of achieving a pCR. Models were constructed by using expression values of the respective genes by quantitative PCR using a univariate or multivariate logistic regression, as described in more detail elsewhere herein.

For luminal tumors, proliferation- and immune-metagenes were constructed. These "metagenes" are further combinations of marker genes which may be linked by an algorithm. The algorithm Immunmetagene (IMG) can be described as follows: 0.309827 * STATl + 0.355327 * PTPRC + 0.434047 * CD3D + 0.495569 * CD3Z + 0.586314 * CXCL9

The algorithm Reduced Immunmetagene (redIMG) can be described as follows: 0.492791 * CD3D + 0.568132 * CD3Z + 0.659078 * CXCL9

The algorithm Proliferation metagene (PMG) can be described as follows: 0.510917 * UBE2C + 0.439843 * RACGAPl + 0.554379 * TOP2A + 0.488023 * STMNl.

The following algorithms were obtained using the above described metagenes:

Table 4 : Exemplary algorithms

The following AUC values were obtained with the above models

Table 5: AUC values for the gene combinations/algorithms of table 4.

Methods

STUDY POPULATION AND HISTOPATHOLOGICAL EXAMINATION As a training cohort for histopathological examination of lymphocyte infiltrations a set of 218 paraffin-embedded pretherapeutic breast cancer core biopsies from the multicenter neoadjuvant Phase III GeparDuo trial performed by the German Breast Group (GBG) , Neu-Isenburg, Germany was used. Histopathological analysis of the lymphocyte infiltrate was performed on H&E sections by two dedicated breast cancer pathologists who were unaware of the clinical and chemotherapy response data. In addition, the grading was reevaluated in the GeparDuo cohort according to current guidelines. All other parameters were extracted from the study database. As a validation cohort for histopathological evaluation of lymphocyte infiltration, a total of 840 samples from the GeparTrio study (NCT00544765) were used. Samples were collected with written informed consent of the patients and stored in the GBG tumor bank at the Institute of Pathology, Charite Hospital, Berlin, Germany. Ethic committee approval was obtained for all centers participating in the clinical studies. In addition, the translational research projects were approved by the institutional review board of the Charite hospital (EA1/139/05) . The clinical data of the 218 GeparDuo samples as well as the 840 GeparTrio samples is shown in table 3, the distribution of clinical parameters was generally comparable to the data reported for the complete GeparDuo and GeparTrio study cohorts. The pCR rate was 12.8% in the GeparDuo cohort and 17% in the GeparTrio cohort.

RNA EXTRACTION

RNA was isolated from 136 FFPE tissue sections (thickness lOμm) of core biopsies from the GeparTrio cohort using a Siemens Healthcare Diagnostics proprietary fully automated method based on silica-coated magnetic beads (research reagent, Siemens Healthcare Diagnostics, Tarrytown, NY, USA) . Total RNA and DNA was specifically bound to silica-coated iron oxide beads, which were magnetically separated using a liquid handling robot. Following elution of nucleic acids from the beads treatment with DNase I (Ambion/Applied Biosystems, Darmstadt, Germany) for 30 min at 37°C was performed in order to remove contaminating DNA.

GENE EXPRESSION ANALYSIS USING KINETIC POLYMERASE CHAIN REACTION

Relative expression of CD3D, CD247 (CD3z) , CD45 (PTPRC), IGKC, CXCL9 and CXCL13 as well as RPL37A used for normalization was assessed by one-step kinetic reverse transcription PCR (kPCR) using the Superscript III Platinum One-Step Quantitative RT-PCR System with ROX (Invitrogen, Karlsruhe, Germany) according to manufacturer's instructions in an ABI PRISM 7900HT (Applied Biosystems, Darmstadt, Germany) . Relative expression levels of genes of interest (GOI) were calculated as ΔCt (cycle threshold) values (ΔCt = 20 - [Ct gene of interest - CtRPL37A] ) . ΔCt values positively correlate with relative gene expression. All PCR assays were performed in triplicate. STATISTICAL EVALUATION

Statistical analysis was performed using SPSS version 15.0 (SPSS Inc. Chicago, Illinois, USA) as well as GraphPad Prism 4 (GraphPad software, La Jolla, California, USA) . The probability of pCR as a function of inflammatory parameters was determined by univariate logistic regression analysis. For multivariate analysis, a multivariate logistic regression was performed including all those clinical parameters that had been determined in the clinical studies GeparDuo and GeparTrio to be predictive of pCR together with the inflammatory parameters.

Table 6 Clinicopathological parameters of the training (GeparDuo) and the validation (GeparTrio) cohort.

Characteristic GeparDuo GeparTrio cohort cohort

No (%) No (%)

No. of samples 218 840

Age group

< 50 years 95 (43.6) 388 (46.2)

> 50 years 123(56.4) 452 (53.8)

Tumor type

Ductal 166(76.1) 641 (76.3)

Lobular 41 (18.8) 119(14.2)

Other 11 (5.0) 74 (8.8) missing 6 (0.7)

Tumor grade

G1 19(8.7) 34(4)

G2 127(58.3) 421 (50.1)

G3 72 (33.0) 278(33.1) missing 0(0) 107(12.7)

ER/PR Status

ER-/PR- 47(21.6) 219(26)

ER+ and/or PR+ 147 (67.4) 512(61) mssing 24(11) 109(13)

Clinical tumor stage

T1 14(6.4)

T2 172(78.9)

T3 32(14.7)

T4 0(0)

Clinical nodal stage cNO 142(65.1) 338 (40.2) cN+ 76 (34.9) 418(49.8) missing 0(0) 84(10.0)

Pathological response pCR 28(12.8) 143(17) nopCR 190(87.2) 697 (83)

Table 6: clinical parameters of study cohorts

In a further approach, further preferred algorithms were obtained to predict response to chemotherapy in luminal or basal / triple negative tumors

According to a preferred embodiment of the invention, the combination of genes comprising CD3D, CXCL9, ESRl, and UBE2C are used for the prediction of response to chemotherapy in luminal tumors . The expression values for these genes may be linked in the algorithm NLRS, wherein ::

NLRS = 0.631604* ( 0.52661*CD3D + 0.850107*CXCL9) - 0.747566*ESRl

+ 0.575971*UBE2C

- 10.710975

Wherein CD3D, CXCL9 and UBE2C represent the expression values for the respective genes obtained as described below, and wherein a value of NLRS above a predetermined cutoff value in the range of -8 to 0, preferably -4 to - 2, more preferably at -3 represents a higher likelihood of a breast cancer patient having a luminal tumor responding to chemotherapy. In the example shown in Fig. 4, a cutoff of -3 was selected for high sensitivity.

According to a preferred embodiment of the invention, the combination of genes comprising CD3D, CXCL9, and UBE2C are used for the prediction of response to chemotherapy in luminal tumors.

The expression values for these genes may be linked in the algorithm c NLRS, wherein :

If UBE2C < 15: no pCR;

If UBE2C > 15 AND (CD3D+CXCL9) > 19: pCR;

If UBE2C > 15 AND (CD3D+CXCL9) < no 19: pCR;

wherein UBE2C, CD3D, and CXCL9 represent the expression values for the respective genes obtained as described below and "no pCR" represents a higher likelihood of the patient having no response to chemotherapy and "pCR" represents a higher likelihood of the patient having a response to chemotherapy, measured as pathological complete response.

According to a preferred embodiment of the invention, the combination of genes comprising STMNl, NFKBIA and HER2/NEU are used for the prediction of response to chemotherapy in basal / triple negative tumors.

The expression values for these genes may be linked in the algorithm NTRS, wherein ::

NTRS = corr ( [STMNl, NFKBIA, HER2/NEU] , ref) where ref = [0.546797, 0.607374, -1.154171]

Wherein STMNl, NFKBIA and HER2/NEU represent the expression values for the respective genes obtained as described below, and wherein a value of NTRS above a predetermined cutoff value in the range of -1 to 1, preferably -0.4 to 0.4, more preferably at -0.2 represents a higher likelihood of a breast cancer patient having a basal / triple negative tumor responding to chemotherapy (example shown in figure 6) .

Expression of the genes shown was assessed by one-step RT- kPCR using the Superscript® III Platinum® One-Step Quantitative RT-PCR System with ROX (Invitrogen, Karlsruhe, Germany) according to manufacturer's instructions in 384-well plates in an ABI PRISM® 7900HT (Applied Biosystems, Darmstadt, Germany) with 30 min at 50⁰C, 2 min at 95°C followed by 40 cycles of 15 sec at 95°C and 30 sec at 60⁰C. In each assay 0.5 μl sample RNA was included. Primer and probe sequences were selected on the basis of empirical rules and with the help of the Primer Express software (V2.0.0, Applied Biosystems, Darmstadt, Germany) . PCR assays were optimized for fragmented RNA from FFPE tissue. For each primer/probe set the efficacy, linearity and reproducibility of the PCR assay was determined.

All PCR assays were performed in duplicate in the GeparTrio training cohort and in triplicate in a further validation cohort. The PCR assays were performed blinded to the clinical outcome data. Means of the Ct values for each gene were calculated. If all duplicates or triplicates of a gene in a specific sample had no PCR signal the Ct value was set as 40 and was censored. If at least on duplicate or triplicate had a Ct value below 40 and at least one duplicate or triplicate had no PCR signal the Ct value for the well without signal was set as 40 and the mean of the duplicates or triplicates was calculated. Relative expression levels of genes of interest (GOI) were calculated as ΔCt values (ΔCt = 20 - [CtGOI - Ct (mean of RPL37A, CALM2, OAZl)]) . In case the Ct value of a gene-of-interest was censored the respective ΔCt value was censored as well. ΔCt values positively correlate with relative gene expression. Assuming an amplification efficacy of 100% increase of one unit corresponds to a doubling of the amount of mRNA. ΔCt values ranged from 4 to

25.

To assure accuracy of the assays a standardized reference RNA

(Stratagene qPCR Human Reference Total RNA, Agilent Technologies, Waldbronn, Germany) was tested for each gene in parallel to the FFPE samples. For exclusion of contamination no-template-controls were assessed in parallel as well.

Data analysis approach for the development of the NLRS and NTRS-algorithm

The 38 candidate genes as described above from different functional groups were selected from the literature, from microarray gene expression data from fresh-frozen tissue (Modlich et al, , Rody et al, Rody et al) or from other own proof-of-principle studies (Denkert et al . , JCO, Schmidt et al Cancer Res) and were measured in 300 samples from the GeparTrio training cohort by RT-kPCR. Only samples with a mean of Ct values of the normalization genes RPL37A, Calm2 and OAZl below 33.4 indicating sufficient RNA were included in this study (n=266) . Raw Ct values of the genes of interest were normalized to the mean Ct of the three normalization genes by Delta Ct(GOI, HK) = 20 - (Ct(GO) - mean CT(HKs)) . The minus sign is to facilitate a straight-forward interpretation (higher values indicate higher expression) , the arbitrary number of 20 was added solely to ensure positivity of the values. These values (Delta Ct values) were used for all subsequent calculations. If the expression of a gene of interest was so low that no signal could be picked up before the last amplification cycle, this partial information was conserved when computing relative expression values; this lead to censored (one-sided) expression values ("Expression of gene is at most...") . Calculations of classifiers and the prediction of response classes used this partial information whenever possible, e.g. when computing score values and comparing them with a threshold.

As a first step in the development of the response scores was the classification of the samples in the three following molecular subgroups using pre-specified cutoff values for ESRl (Cutoff value: 16) and HER2/NEU (cutoff value: 19.5) mRNA expression: "Her2" (HER2/NEU-positive) , "Luminal" (HER2/NEU negative, ESRl positive) and "triple-negative" (HER2/NEU negative, ESRl negative) . The "Her2" subgroup was excluded from this study. For luminal and triple-negative tumors, separate algorithms were designed.

For generation of a luminal algorithm predictive of pCR training was performed in GeparTrio luminal samples by using uni-, bi- and trivariate logistic regression. Since some values (e.g. outliers) can adversely impact feature selection, feature selection was bootstrapped for various subsets of training data to assess robustness of feature choice. Feature selection was performed by a forward selection scheme where in each step, new genes were combined with an existing set of genes and all those combinations were kept where the regression coefficients of all genes used were significantly different from zero (2 x alpha < 0.05) . This selection process was performed 1000 times for a combination of up to three features as the power of the available data did not suggest more complex algorithms. Finally, gene combinations were ranked by their respective robustness (frequency of significance in the bootstrapping process) . Receiver operator characteristics were assessed for the five top ranked gene triplets. The model with highest area under the curve (AUC) was a combination of the relative expression values of CD3D, ESRl and UBE2C. We observed that the logistic regression models showed the effect that CD3D could be switched for another T cellular immune response marker CXCL9 without sacrificing overall classifier performance. Therefore, a T cellular immune metagene (TIMG) was constructed using the first principal component of a principal component analysis (PCA) involving CD3D and CXCL9 in order to improve robustness of the algorithm. TIMG was calculated by TIMG = 0.526610 x CD3D + 0.850107 x CXCL9. The final predictive model was generated by trivariate logistic regression including the TIMG and the neoadjuvant luminal response score (NLRS) was calculated as follows: NLRS = - 10.710975 + 0.631604 x TIMG - 0.747566 x ESRl + 0.575971 x UBE2C. A positive coefficient indicates that increased expression of a gene is associated with a high probability of pCR, whereas a negative coefficient indicates an inverse association of the gene expression value with the probability of pCR. NLRS ranged between -8.5 and 1.0 in the GeparTrio training cohort, and higher scores indicate a higher likelihood of achieving a pCR.

For generation of a predictive algorithm for triple negative tumors primarily the same approach was used as described for the luminal algorithm. However, logistic regression did not result in models with satisfactory performance. Therefore, we used correlation clusters which were based on the discovery of a reference profile in a set of at least three genes (a smaller number of genes does not allow such a thing) . If the correlation of a given sample to the reference profile is large (close to 1), the patient is likely to achieve a pCR. If the correlation is negative (close to -1), she is likely not to achieve a pCR. The training and feature selection of this model involved a constraint non-linear optimization which is not in the scope of this publication. The approach is closely related to k-means clustering with two clusters: Given a set of candidate genes, the idea is to find two reference profiles (centroids) that are characteristic for each cluster, usually the vector of the class means of the gene expressions. Unknown samples are classified such that distance to each centroid is computed, and classification is then performed by comparison of these distances. Usually, the unknown sample is classified into the class whose centroid is nearest.

The approach used here follows the same general idea but uses Pearson correlation as the distance measure. This correlation coefficient is a number between -1 and 1 where -1 denotes perfect anti-correlation, and 1 denotes perfect correlation. It turns out that the problem of finding centroids for the two classes in this case is more difficult. If one observed that the coefficient is invariant to translation and scaling, the difference of the distance measures to each controid reduced to the difference of covariance. Using the linear properties of the covariance, it can then be shown that not two separate profiles need to be determined, but only their difference for which nonlinear constraints are imposed (zero mean and unit variance) .

Given training data, this single reference profile is determined as the parameter set fulfilling the constraints while minimizing square sum of the residuals (1-corr (ref, sample) ) ^A2 for pCRs, (1+corr (ref, sample) ) ^A2 for non-pCRs . Since we lose two degrees of freedom to the constraints, this approach is useful only when using sets of at least three genes .

The best combination was assessed by determining the AUC for all possible combinations of three genes.

The final algorithm correlates the expression values of STMNl, NFKBIA and HER2/NEU to a reference profile in the following manner: neoadjuvant triple-negative response score (NTRS) = corr ( [STMNl, NFKBIA, HER2/NEU] , ref) where ref = [0.546797, 0.607374, -1.154171] . A positive value indicates a positive association of expression level with the achievement of pCR whereas a negative value indicates a negative association.

Selection of cutoff values for classification of an unknown patient sample into a "pCR" and "no pCR" group

In order to select appropriate cutoff values for NLRS and NTRS to classify a patient into a "pCR" or "no pCR" group a graph was generated plotting each potential cutoff value against sensitivity and specificity of the classifier (Figures 4&6) . To achieve the clinical demand of a sensitivity of at least 80% in luminal tumors and of at least 90% in triple-negative tumors, "-3" was selected as cutoff value for NLRS and "-0.2" for NTRS.

Progress has been made in the treatment of breast cancer with cytotoxic chemotherapy. However, resistance against different treatment modalities is a major reason for therapy failure and bad prognosis of advanced disease. Following neo-adjuvant primary chemotherapy, for example, about 20-50% of patients do not respond to therapy and only 15- 25% of patients achieve pathological complete remission (pCR) which means a complete eradication of invasive tumor cells as assessed by histopathological examination and which is associated with prolonged survival. Thus, there is a substantial number of patients who suffer from side effects without clinical benefit of therapy. Up to now, there is no reliable diagnostic test for prediction of response to cytotoxic chemotherapy .

It is common practice to measure estrogen receptor (ESRl), progesterone receptor (PGR) and Her-2/neu status by immunohistochemistry and/or fluorescence in situ hybridization as well as to assess tumor grade by histopathology at diagnosis of breast cancer. Combining these markers with clinical response after 2 cycles of neo-adjuvant chemotherapy (in-vivo chemoresistance test) it is possible to select a patient group in which the pCR rate will be up to 50%. Using this approach, patients still get 2 cycles of chemotherapy and there is still a substantial number of patients who do not benefit from chemotherapy and need other therapies .

We envisaged the combinatorial expression analysis of several genes in tumor tissue obtained before start of therapy. The result of the analysis allows prediction of response or resistance to chemotherapy and can avoid toxic and expensive chemotherapy without a clinical benefit.

Measurement of the markers for the algorithm can be performed on mRNA level using RT-kPCR or gene expression array platforms such as for example Affymetrix, Illumina or Planar Wave Guide or on protein level by, for example, immunological techniques such as immunohistochemistry . The combined marker genes can be used in breast cancer for prediction of response to a taxane/anthracycline-containing chemotherapy in the adjuvant as well as in the neo-adjuvant setting. The combined marker genes may be useful for prediction of taxane/anthracycline-response also in other cancer types.

On the basis of a set of gene expression data of 58 candidate genes assessed in 300 formalin-fixed paraffin-embedded breast cancer samples as well as a set of genome-wide gene expression microarray data from 56 fresh-frozen breast cancer samples obtained before start of a taxane/anthracycline combination chemotherapy we identified genes expression of which was associated with pCR or no pCR as well as of genes associated with tissue response (TR) or no tissue response. The expression values of the genes can be used in combination with each other to predict response to taxane/anthracycline chemotherapy .

The advantage of the here presented biomarker test is that prediction of therapy response is possible by a molecular test prior to start of chemotherapy. The use of an "in-vivo chemoresistance test" by 2 cycles of chemotherapy is not necessary. Moreover, the combined assessment of several genes in an algorithm helps to overcome one main issue: This approach allows the resolution of the fact that there might be not one, but multiple reasons for a given response behaviour which is the case in a heterogeneous disease such as breast cancer. This situation cannot satisfactorily be resolved using single markers.

In summary, we identified highly informative genes predictive of response to taxane/anthracycline containing neo-adjuvant cytotoxic chemotherapy in a set of expression data of 58 genes from 300 fresh-frozen breast cancer samples as well as in a set of Affymetrix U133A microarray expression data from 56 fresh-frozen breast cancer samples. Tissue samples were obtained by core needle biopsies from patients with breast cancer (T4/T>2 cm, NO-3, MO) before start of neo-adjuvant chemotherapy with 4 or 6 cycles of docetaxel (75 mg/m² ) , doxorubin (50 mg/m²) and cyclophosphamide (500 mg/m²) (TAC) . Pathological response was assessed in each patient following completion of therapy using the tissue preparation from surgery. The project pursued two aims: (1) Identification of patients achieving a pCR (no invasive tumor left in the breast or lymph nodes), (2) Identification of patients achieving tissue response (TR; at least limited response with residual invasive tumor <0.5cm) . Regarding chemotherapy response patients were discretized in two groups: pCR vs. no pCR or tissue response vs. no tissue response. For finding in FFPE tissue, only samples with sufficient RNA yield and available response data were included in the study (n=256) . Relative expression of each gene was computed as Delta CT value to housekeeping gene RPL37A.

The following genes were measured by quantitative PCR: TMSL8, ABCCl, EGFR, MVP, ACOX2 , HER2/NEU, MYHIl, TOBl,

AKRlCl, ERBB4, NFKBIA, TOP2A, AKR1C3, ESRl, OLFMl, TOP2B, ALCAM, FRAPl, PGR, TP53, BCL2, GADD45A, PRKABl, TUBAlA, C16orf45, HIFlA, PTPRC, TUBB, CA12, IGKC, RACGAPl, UBE2C, CD14, IKBKB, S100A7, VEGFA, CD247, KRT5, SEPT8, YBXl, CD3D, MAPK3, SLC2A1, CDKNlA, MAPT, SLC7A8, CHPTl, MLPH, SPONl,

CXCL13, MMPl, STATl, CXCL9, MMP7, STC2, DCN, MUCl, STMNl, and house keeping genes OAZl, CALM2, RPL37A.

For identification of informative genes samples were divided in three molecular subgroups according to ESRl and HER2/NEU mRNA expression: Luminal (HER2/NEU neg.;ESRl pos.; n=143) , Basal / triple negative (HER2/NEU neg., ESRl neg.; n=64) and HER2 (HER2/NEU pos.; n=49) . Training was performed by using uni- and bivariate logistic regression. Since single extreme values (e.g. outliers) can adversely impact feature selection discovery was repeated for various subsets of training data to assess robustness. Random selection of m samples (out of n original samples) with putting back was used for training. In each discovery step, best genes (significance of regression coefficient less than some cutoff value, e.g. 5%) are selected.

To all of these genes, other genes were combined to achieve even better risk scores (partially greedy approach) . Quality of a combination was assessed by max p value of all regression coefficients (excluding p value of constant) . Best single genes and best gene pairs were assessed for robust selection in two independent runs of 1000 resampling cohorts. Results for both pCR and TR for all three groups were finally summarized. Receiver operator characteristics were assessed for the top ranked genes/gene combinations.

Moreover, informative genes predictive of response to taxane/anthracycline-containing neo-adjuvant cytotoxic chemotherapy were also identified in fresh-frozen breast cancer samples profiled by Affymetrix U133A microarrays. Again, samples were divided in three molecular subgroups according to ESRl and HER2/NEU mRNA expression: Luminal (HER2/NEU neg.;ESRl pos.), Basal / triple negative (HER2/NEU neg., ESRl neg.) and HER2 (HER2/NEU pos.) . Best significant informative genes for univariate separation of patients with pCR vs. patients without pCR were identified by standard t test statistics. Genuine multivariate classifiers can be built from that.

In an alternative approach, differential gene expression was determined in a cohort of 136 tumors (n=136) with regard to pCR vs. no pCR and tissueR response vs. no tissue response. These 136 tumor sample represent a subcohort of the above mentioned 300 tumor samples. In this independent approach, further differential gene expression with regard to pCR vs. no pCR and tissue response vs. no tissue response was determined in ER+ tumors (n=85) , ER- tumors (n=51); luminal tumors (ER+ and HER2-, n=72) and basal (triple negative) tumors (ER- und HER2-, n=35) and HER+ tumors (n=29) . The genes examined in this approach were ABCCl, ACOX2, AKR1C3,

BCL2, CD3D, CD3Z,

CHPTl, CXCL13, CXCL9, EGFR, HER2/NEU, ERBB4,

ESRl, FRAPl, IGKC, MAPK3, MAPT, MLPH, MMPl,

MUCl, MVP, NFKBIA, PGR, PTPRC, RACGAPl,

S100A7, SEPT8, SLC2A1, SPONl, STATl, STC2,

STMNl, TMSL8, TOP2A, TP53, TUBB3, UBE2C, and YBXl.

Genes expressed differentially with regard to pCR vs. no pCR in all tumors are shown in table 7, below.

Table 7: Differentially expressed genes, pCR, all tumors Genes expressed differentially with regard to tissue response vs. no tissue response in all tumors are shown in table 8, below.

Table i Differentially expressed genes, tissue response, all tumors Genes expressed differentially with regard to pCR vs. no pCR in ER+ tumors are shown in table 9, below.

Table 9: Differentially expressed genes, pCR, ER+ tumors

Genes expressed differentially with regard to tissue response vs. no tissue response in ER+ tumors are shown in table 10, below

Table 10: Differentially expressed genes, tissue response, ER+ tumors

Genes expressed differentially with regard to pCR vs. no pCR in ER- tumors are shown in table 11, below.

Table 11: pCR vs. no pCR in ER- tumors

Genes expressed differentially with regard to tissue response vs. no tissue response in ER- tumors are shown in table 12, below.

Table 12: tissue response vs. no tissue response in ER- tumors Genes expressed differentially with regard to pCR vs. no pCR in luminal tumors are shown in table 13, below.

Table 13: pCR vs. no pCR in luminal tumors

Genes expressed differentially with regard to tissue response vs. no tissue response in luminal tumors are shown in table 14, below.

Table 14: tissue response vs. no tissue response in luminal tumors

Genes expressed differentially with regard to pCR vs. no pCR in basal / triple negative tumors are shown in table 15, below.

Table 15 pCR vs. no pCR in basal / triple negative tumors

Genes expressed differentially with regard to tissue response vs. no tissue response in basal / triple negative tumors are shown in table 16, below.

Table 16: tissue response vs. no tissue response in basal / triple negative tumors Genes expressed differentially with regard to pCR vs. no pCR in HER+ tumors are shown in table 17, below.

Table 17: pCR vs. no pCR in HER+ tumors

Genes expressed differentially with regard to tissue response vs. no tissue response in HER+ tumors are shown in table 18, below.

Table 18 tissue response vs. no tissue response in

HER+ tumors

Table 16 below indicates the best models obtained in this alternative approach.

model P BIC AUC specificity threshold (70% /

(70% / 80% / 90%) 80% / 90%) all tumors pCR ~ TMSL8 6,90E-06 135,1 0,76 76%/64%/37% 0.24/0.2/0.13 pCR ~ NFKBIA + TMSL8 1 ,10E-06 132,7 0,79 73%/68%/61% 0.24/0.19/0.1

C pCR ~ MVP + PTPRC + 1 J0E-07 130,7 0,82 74%/67%/62% 0.22/0.18/0.1

TMSL8 5 pCR ~ MVP + PTPRC + 4,90E-08 130,3 0,84 77%/70%/55% 0.26/0.18/0.1

TMSL8 + UBE2C 1 pCR ~ HER2/NEU + ERBB4 2J0E-08 131 ,2 0,86 89%/84%/66% 0.38/0.31/0.1

+ MVP + PTPRC + TMSL8 3

ESR1 positive pCR ~ CD3D 0,00063 58,8 0,84 83%/83%/65% 0.17/0.16/0.1 pCR ~ CD3D + MUC1 0,00015 57,3 0,87 83%/81 %/79% 0.18/0.18/0.1 2 pCR ~ CD3D + TP53 + 5,50E-05 57 0,9 87%/83%/59% 0.22/0.15/0.0

UBE2C 3 pCR ~ ABCC1 + CD3D + 8,60E-06 55 0,94 92%/88%/81% 0.21/0.14/0.0

MUC1 + TP53 9 pCR ~ CD3D + MLPH + 4,00E-08 45,4 0,98 97%/97%/89% 0.46/0.33/0.1

MUC1 + SP0N1 + TP53 2

ESR1 negative pCR ~ STMN1 0,0021 66,4 0,74 66%/52%/52% 0.38/0.33/0.3 2 pCR ~ MAPK3 + TMSL8 0,0017 67 0,76 66%/62%/28% 0.36/0.35/0.2 pCR ~ IGKC + STAT1 + 0,00012 62,9 0,84 76%/76%/66% 0.53/0.5/0.28

TMSL8 pCR ~ BCL2 + IGKC + 2,80E-05 61 ,4 0,88 83%/79%/72% 0.56/0.42/0.3

STAT1 + TMSL8 4 pCR ~ BCL2 + HER2/NEU 3,50E-05 63,4 0,89 83%/76%/69% 0.54/0.48/0.3

+ IGKC + STAT1 + TMSL8 6

Luminal pCR ~ CD3D 0,00089 43,4 0,86 86%/86%/71% 0.16/0.16/0.0

Q

O pCR ~ FRAP1 + MUC1 0,00019 41 ,6 0,91 95%/78%/77% 0.29/0.07/0.0

D pCR ~ CD3D + FRAP1 + 2,60E-05 39,1 0,95 98%/86%/83% 0.44/0.1 1/0.0

MUC1 6 pCR ~ ESR1 + FRAP1 + 3,10E-06 36,3 0,98 100%/94%/89% 0.59/0.17/0.1

MUC1 + TP53 pCR ~ CXCL9 + FRAP1 + 9,40E-09 25,7 1 100%/100%/100 1/1/0.54

MUC1 + NFKBIA + STAT1 %

Basal pCR - NFKBIA 0,0032 45,5 0,8 86%/62%/48% 0.5/0.33/0.24 pCR ~ IGKC + STAT1 6,00E-04 42,9 0,85 76%/71 %/71% 0.51/0.29/0.2

Q

O pCR ~ IGKC + STAT1 + 6,90E-05 39,5 0,88 76%/76%/76% 0.52/0.49/0.4 TMSL8 5 pCR ~ MLPH + NFKBIA + 4,10E-05 39,4 0,94 95%/86%/81% 0.7/0.45/0.28

STMN1 + YBX1 pCR ~ BCL2 + IGKC + 1 ,60E-05 38,6 0,95 100%/95%/86% 0.73/0.58/0.3

PTPRC + SPON1 + TMSL8 6

Her2 pCR ~ TMSL8 0,0084 36,2 0,78 56%/56%/56% 0.29/0.27/0.2 pCR ~ AKR1 C3 + ERBB4 0,00087 32,4 0,9 89%/83%/67% 0.42/0.36/0.1

7 pCR ~ CD3D + CD3Z + 0,00019 30 0,93 89%/89%/72% 0.54/0.49/0.1

TMSL8 3 pCR ~ AKR1 C3 + CD3D + 2,30E-07 16,7 1 100%/100%/100 1/1/0.57

ERBB4 + TMSL8 % pCR ~ AKR1 C3 + BCL2 + 7,50E-07 20 1 100%/100%/100 1/1/0.47

ERBB4 + NFKBIA + SPON1 % all tumors

TissueResponse ~ CXCL9 6,60E-07 165,1 0,74 61%/58%/46% 0.36/0.32/0.2

7

TissueResponse ~ BCL2 + 2,00E-09 154,6 0,8 77%/70%/43% 0.46/0.36/0.1

UBE2C 8

TissueResponse ~ BCL2 + 1 ,60E-10 150,9 0,82 80%/77%/54% 0.53/0.41/0.2

CXCL9 + UBE2C

TissueResponse ~ BCL2 + 6,10E-11 150,8 0,84 81%/77%/63% 0.48/0.4/0.25

MVP + PTPRC + UBE2C

TissueResponse ~ AKR1 C3 3,40E-11 151 ,6 0,86 81%/80%/71% 0.51/0.45/0.3

+ BCL2 + CXCL9 + MVP +

UBE2C

ESR1 positive

TissueResponse ~ CD3D 8,10E-07 81 ,8 0,84 75%/73%/70% 0.29/0.26/0.2

TissueResponse ~ CD3D + 2,10E-07 79,8 0,88 87%/75%/70% 0.39/0.27/0.2

RACGAP 1 1

TissueResponse ~ CD3D + 2,10E-08 76,3 0,91 90%/84%/81% 0.45/0.36/0.2

MAPT + RACGAP 1 6

TissueResponse ~ ABCC1 7,80E-09 75,8 0,93 92%/90%/86% 0.57/0.41/0.2

+ CD3D + MAPT + 6

RACGAP 1

TissueResponse ~ ACOX2 1 ,90E-09 74,5 0,94 92%/90%/83% 0.48/0.4/0.21

+ CD3D + MAPT + MVP +

RACGAP 1

ESR1 negative

TissueResponse ~ S100A7 0,014 69,1 0,71 50%/45%/40% 0.52/0.48/0.4

TissueResponse ~ S100A7 0,0082 69,4 0,73 60%/40%/35% 0.52/0.47/0.4

+ UBE2C 5

TissueResponse ~ ABCC1 0,0036 69,4 0,8 65%/45%/45% 0.53/0.42/0.3

+ S100A7 + UBE2C 9

TissueResponse ~ BCL2 + 0,00028 65,6 0,85 75%/65%/65% 0.56/0.49/0.4

MAPK3 + MVP + STAT1 5

TissueResponse ~ BCL2 + 0,00021 66,7 0,86 80%/70%/60% 0.68/0.52/0.4

IGKC + MVP + PTPRC + 5

UBE2C Luminal

TissueResponse ~ CD3D 1 ,90E-06 62,2 0,86 77%/77%/70% 0.24/0.24/0.1 8

TissueResponse ~ CD3D + 4,60E-07 59,9 0,89 80%/80%/79% 0.29/0.28/0.2

UBE2C 1

TissueResponse ~ BCL2 + 2J0E-08 55,3 0,93 89%/89%/80% 0.41/0.39/0.2

CD3D + FRAP1

TissueResponse ~ BCL2 + 4,50E-09 52,9 0,96 98%/95%/80% 0.51/0.44/0.1

CD3D + FRAP1 + SPON1 7

TissueResponse ~ BCL2 + 3,10E-09 53,7 0,96 96%/89%/88% 0.63/0.35/0.2

CD3D + FRAP1 + TOP2A + 6

UBE2C

Basal

TissueResponse ~ S100A7 0,0072 47,7 0,77 53%/53%/47% 0.43/0.42/0.3 6

TissueResponse ~ MUC1 + 0,001 1 44,9 0,84 87%/53%/27% 0.64/0.51/0.1

S100A7 8

TissueResponse ~ MVP + 0,00029 43,1 0,89 87%/80%/67% 0.64/0.58/0.3

STAT1 + TUBB3 2

TissueResponse ~ MMP1 + 1 ,60E-06 33,2 0,97 100%/93%/87% 0.85/0.68/0.5

MVP + STAT1 + TUBB3 1

TissueResponse ~ AKR1 C3 3,90E-09 21 ,3 1 100%/100%/100 1/1/1

+ FRAP1 + MMP1 + MVP + %

STAT1

Her2

TissueResponse ~ CHPT1 0,0066 37,5 0,8 83%/75%/25% 0.64/0.6/0.22

TissueResponse ~ CHPT1 0,0018 35,6 0,86 83%/83%/42% 0.62/0.58/0.2

+ ERBB4 8

TissueResponse ~ CHPT1 0,0012 35,6 0,91 92%/92%/67% 0.65/0.63/0.4

+ ERBB4 + MVP

TissueResponse ~ ABCC1 0,00047 34,8 0,94 100%/100%/83 0.83/0.79/0.4

+ CHPT1 + SPON1 + % 6

TOP2A

TissueResponse ~ ABCC1 1 ,60E-05 28,4 0,98 100%/100%/92 1/0.96/0.44

+ CHPT1 + SPON1 + %

TOP2A + TUBB3

Table 16: Best model algprithms, alternative approach

In Table 16, p designates the significance from Omnibus-Test for logistic Model, AUC designates the Area under ROC-Curve BIC designates the Bayesian information criterion Specificity refers to the specificity for sensitivities of 70%, 80%, 90% respectively.

Threshold refers to threshold for fitted probability, to reach sensitivities of 70%, 80%, 90% respectively.

Additional data from the gene combination of table 16 is shown in an exemplary gene combination for each tumor class with regard to pCR in Fig. 8 (all tumors), Fig. 9 (ER+) tumors, Fig. 10 (ER- tumors), Fig. 11 (luminal tumors), Fig.

12 (basal/triple negative tumors), and Fig. 13 (Her2+ tumors) . The top panel of Figs. 8 to 13 shows the values for BIC and AUC as related to the number of genes used in the respective algorithm. The middle panel of Figs. 8 to 13 shows the fitted probabilities of the exemplary algorithm as indicated in the middle panel. The bottom panel of Figs. 8 to

13 shows the ROC curve of the exemplary algorithm as indicated in the middle panel.

In yet another approach, best significant informative genes for univariate separation of patients with pCR vs. patients without pCR were identified by standard t test statistics in the group of Her2 negative/ESRl negative (basal) patients. Genuine multivariate classifiers can be built from that. Receiver operator characteristics were calculated.

Table 17 shows 4 informative genes obtained through this approach.

Gene Symbol MW pCR non-pCR Fold TTEST Probe Set ID Gene Title change

SSH3 4,67 6,54 0,27 0,0008 219919_s_at slingshot homolog 3 (Drosophila)

TOPORS 6,30 7,25 0,52 0,0076 221979_at Topoisomeras e I binding, arginine/serine -rich

EN1 9,81 8,19 3,06 0,0132 220559_at engrailed homolog 1

DUSP6 5,14 6,69 0,34 0,0158 208893_s_at dual specificity phosphatase 6

The algorithms, correlation coefficients, cutoffs, factors and summands presented herein where obtained with the analytical platforms and instruments described herein. Transferring to a different instrument platform (e.g. from a TaqMan PCR system (present examples) to another PCR System, such as Stratagene PCR system, Lightcycler PCR System) or to a different system (e.g. from a real-time PCR based system to a microarray based system) will result in algorithms with different correlation coefficients, cutoffs, factors and summands . This is known to the skilled person. Knowledge of the genes used in an algorithm and the structure of an algorithm will allow the skilled person to routinely determine the different correlation coefficients, cutoffs, factors and summands applicable for the transferred system.

Supplemental table 1:

Sequences of oligonucleotides used as probes, forward primers ("primer, fo") , and reverse primers ("primer re") used for PCR experiments.

Supplemental Table 2: Gene names and accession numbers

References DeNardo DG, Coussens LM Inflammation and breast cancer. Balancing immune response: crosstalk between adaptive and innate immune cells during breast cancer progression. Breast Cancer Res. 2007 ; 9 (4) : 212.

Aaltomaa S, Lipponen P, Eskelinen M, Kosma VM, Marin S, Alhava E, Syrjanen K. Lymphocyte infiltrates as a prognostic variable in female breast cancer. Eur J Cancer. 1992;28A(4- 5) :859-64.

Zitvogel L, Kroemer G. The immune response against dying tumor cells: avoid disaster, achieve cure. Cell Death Differ, 2008 Jan;15 (1) :l-2.

Apetoh L, Ghiringhelli F, Tesniere A, Criollo A, Ortiz C, Lidereau R, Mariette C, Chaput N, Mira JP, Delaloge S, Andre F, Tursz T, Kroemer G, Zitvogel L. The interaction between HMGBl and TLR4 dictates the outcome of anticancer chemotherapy and radiotherapy. Immunol Rev. 2007 Dec; 220: 47- 59.

Apetoh L, Tesniere A, Ghiringhelli F, Kroemer G, Zitvogel L. Molecular interactions between dying tumor cells and the innate immune system determine the efficacy of conventional anticancer therapies. Cancer Res. 2008 Jun 1 ; 68 (11) : 4026-30. Zitvogel L, Apetoh L, Ghiringhelli F, Kroemer G. Immunological aspects of cancer chemotherapy. Nat Rev Immunol. 2008 Jan; 8 (1) : 59-73.

Smith IC, Heys SD, Hutcheon AW, Miller ID, Payne S, Gilbert FJ, Ah-See AK, Eremin O, Walker LG, Sarkar TK, Eggleton SP, Ogston KN. Neoadjuvant chemotherapy in breast cancer: significantly enhanced response with docetaxel. J Clin Oncol. 2002 Mar 15; 20 ( 6) : 1456-66.

Fisher B, Bryant J, Wolmark N, Mamounas E, Brown A, Fisher ER, Wickerham DL,

Begovic M, DeCillis A, Robidoux A, Margolese RG, Cruz AB Jr, Hoehn JL, Lees AW,

Dimitrov NV, Bear HD. Effect of preoperative chemotherapy on the outcome of women with operable breast cancer. J Clin Oncol. 1998 Aug; 16 (8) : 2672-85. von Minckwitz G, Raab G, Caputo A, Schutte M, Hilfrich J, Blohmer JU, Gerber B, Costa SD, Merkle E, Eidtmann H, Lampe D, Jackisch C, du Bois A, Kaufmann M. Doxorubicin with cyclophosphamide followed by docetaxel every 21 days compared with doxorubicin and docetaxel every 14 days as preoperative treatment in operable breast cancer: the GEPARDUO study of the German Breast Group. J Clin Oncol. 2005 Apr 20;23 (12) :2676-85.

Rakha EA, El-Sayed ME, Lee AH, Elston CW, Grainge MJ, Hodi Z, Blarney RW, Ellis 10. Prognostic significance of

Nottingham histologic grade in invasive breast carcinoma. J Clin Oncol. 2008 JuI 1 ; 26 (19) : 3153-8 von Minckwitz G, Blohmer JU, Raab G, Lδhr A, Gerber B, Heinrich G, Eidtmann H, Kaufmann M, Hilfrich J, Jackisch C, Zuna I, Costa SD; German Breast Group. In vivo chemosensitivity-adapted preoperative chemotherapy in patients with early-stage breast cancer: the GEPARTRIO pilot study. Ann Oncol. 2005 Jan; 16 (1) : 56-63. von Minckwitz G, Kummel S, Vogel P, Hanusch C, Eidtmann H, Hilfrich J, Gerber B, Huober J, Costa SD, Jackisch C, Loibl S, Mehta K, Kaufmann M; German Breast Group. Neoadjuvant vinorelbine-capecitabine versus docetaxel-doxorubicin- cyclophosphamide in early nonresponsive breast cancer: phase III randomized GeparTrio trial. J Natl Cancer Inst. 2008 Apr 16;100 (8) :542-51. von Minckwitz G, Kϋmmel S, Vogel P, Hanusch C, Eidtmann H, Hilfrich J, Gerber B, Huober J, Costa SD, Jackisch C, Loibl S, Mehta K, Kaufmann M; German Breast Group. Intensified neoadjuvant chemotherapy in early-responding breast cancer: phase III randomized GeparTrio study. J Natl Cancer Inst. 2008 Apr 16; 100 (8) : 552-62.

Gianni L, Zambetti M, Clark K, Baker J, Cronin M, Wu J, Mariani G, RodriguezJ, Carcangiu M, Watson D, Valagussa P, Rouzier R, Symmans WF, Ross JS, Hortobagyi GN, Pusztai L, Shak S. Gene expression profiles in paraffin-embedded core biopsy tissue predict response to chemotherapy in women with locally advanced breast cancer. J Clin Oncol. 2005 Oct 10;23 (29) :7265-77.

Schmidt M, Bδhm D, von Tδrne C, Steiner E, Puhl A, Pilch H, Lehr HA, Hengstler JG, Kδlbl H, Gehrmann M. The humoral immune system has a key prognostic impact in node-negative breast cancer. Cancer Res. 2008 JuI 1 ; 68 (13) : 5405-13.

Apetoh L, Ghiringhelli F, Tesniere A, Obeid M, Ortiz C, Criollo A, Mignot G, Maiuri MC, Ullrich E, Saulnier P, Yang H, Amigorena S, Ryffel B, Barrat FJ, Saftig P, Levi F, Lidereau R, Nogues C, Mira JP, Chompret A, Joulin V, Clavel- Chapelon F, Bourhis J, Andre F, Delaloge S, Tursz T, Kroemer G, Zitvogel L. Toll-like receptor 4-dependent contribution of the immune system to anticancer chemotherapy and radiotherapy. Nat Med. 2007 Sep; 13 ( 9) : 1050-9. Ladoire S, Arnould L, Apetoh L, Coudert B, Martin F,

Chauffert B, Fumoleau P, Ghiringhelli F. Pathologic complete response to neoadjuvant chemotherapy of breast carcinoma is associated with the disappearance of tumor-infiltrating foxp3+ regulatory T cells. Clin Cancer Res. 2008 Apr 15;14 (8) :2413-20..

Wolf AM, Wolf D, Steurer M, Gastl G, Gunsilius E, Grubeck- Loebenstein B. Increase of regulatory T cells in the peripheral blood of cancer patients. Clin Cancer Res. 2003 ;9 (2) : 606-12.

Curiel JT, nat med 2004

Bates GJ, Fox SB, Han C, Leek RD, Garcia JF, Harris AL, Banham AH. Quantification of regulatory T cells enables the identification of high-risk breast cancer patients and those at risk of late relapse. J Clin Oncol. 2006 Dec 1;24 (34) :5373-80.

Perez SA, Karamouzis MV, Skarlos DV, Ardavanis A, Sotiriadou NN, Iliopoulou EG, Salagianni ML, Orphanos G,

Baxevanis CN, Rigatos G, Papamichail M. CD4+CD25+ regulatory T-cell frequency in HER-2/neu (HER) -positive and HER-negative advanced-stage breast cancer patients. Clin Cancer Res. 2007 May 1;13 (9) : 2714-21 Fulton A, Miller F, Weise A, Wei WZ. Prospects of controlling breast cancer metastasis by immune intervention.

Walser TC, Ma X, Kundu N, Dorsey R, Goloubeva O, Fulton AM. Immune-mediated modulation of breast cancer growth and metastasis by the chemokine Mig (CXCL9) in a murine model. J Immunother. 2007 Jul-Aug; 30 (5) : 490-8.

Modlich O, Prisack HB, Munnes M, Audretsch W, Bojar H. Predictors of primary breast cancers responsiveness to preoperative epirubicin/cyclophosphamide-based chemotherapy: translation of microarray data into clinically useful predictive signatures. J Transl Med. 2005 3:32.

Rody A, Karn T, Gatje R, Ahr A, Solbach C, Kourtis K, Munnes M, Loibl S, Kissler S, Ruckhaberle E, Holtrich U, von Minckwitz G, Kaufmann M. Gene expression profiling of breast cancer patients treated with docetaxel, doxorubicin, and cyclophosphamide within the GEPARTRIO trial: HER-2, but not topoisomerase II alpha and microtubule-associated protein tau, is highly predictive of tumor response. Breast. 2007 16:86-93. Rody A, Karn T, Gatje R, Kourtis K, Minckwitz G, Loibl S, Munnes M, Ruckhaberle E, Holtrich U, Kaufmann M, Ahr A. Gene expression profiles of breast cancer obtained from core cut biopsies before neoadjuvant docetaxel, adriamycin, and cyclophoshamide chemotherapy correlate with routine prognostic markers and could be used to identify predictive signatures. Zentralbl Gynakol. 2006 128:76-81. Rody A, Karn T, Solbach C, Gaetje R, Munnes M, Kissler S, Ruckhaberle E, Minckwitz GV, Loibl S, Holtrich U, Kaufmann M. The erbB2+ cluster of the intrinsic gene set predicts tumor response of breast cancer patients receiving neoadjuvant chemotherapy with docetaxel, doxorubicin and cyclophosphamide within the GEPARTRIO trial. Breast. 2007 6:235-240. Denkert C, Loibl S, Noske A, Roller M, Muller BM, Komor M, Budczies J, Darb-Esfahani S, Kronenwett R, Hanusch C, von

Tδrne C, Weichert W, Engels K, Solbach C, Schrader I, Dietel M, von Minckwitz G (2009) . Tumor-Associated Lymphocytes As an Independent Predictor of Response to Neoadjuvant Chemotherapy in Breast Cancer. J Clin Oncol. 2009 Nov 16. [Epub ahead of print]

Claims

What is claimed is:

1. A method for predicting a response to and/or benefit from chemotherapy in a patient suffering from cancer comprising the steps of a) classifying a tumor into at least two classes said at least two classes being selected from the group consisting of a a Her 2/neu negative, ESR negative (basal / triple negative) class of tumors, and a Her 2/neu negative, ESR positive (luminal class) class of tumors,

b) determining in a tumor sample the expression of at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class, and

c) depending on said gene expression, predicting said response and/or benefit;

wherein said at least one marker gene comprises a plurality of genes for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR positive (luminal class) tumors said plurality of genes comprising the genes CD3D, CXCL9, and UBE2C; or

wherein said at least one marker gene comprises a plurality of genes for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR negative (basal or triple negative class) tumors, said plurality of genes comprising the genes STMNl, HER2/NEU, and NFKBIA.

2. Method according to claim 1 wherein the plurality of genes is used for predicting a response to and/or benefit from chemotherapy in Her 2/neu negative, ESR positive (luminal class) tumors said plurality of genes further comprising the gene ESRl.

3. A method for predicting a response to and/or benefit from chemotherapy in a patient suffering from cancer comprising the steps of a) classifying a tumor as belonging to at least one class,

b) determining in a tumor sample the expression of at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class, and

c) depending on said gene expression, predicting said response and/or benefit;

wherein said at least one marker gene comprises a gene selected from the group consisting of TMSL8, ABCCl, EGFR, MVP, ACOX2 , HER2/NEU, MYHIl, TOBl, AKRlCl, ERBB4, NFKBIA, TOP2A, AKR1C3, ESRl, OLFMl, T0P2B, ALCAM, FRAPl, PGR, TP53, BCL2, GADD45A, PRKABl, TUBAlA, C16orf45, HIFlA, PTPRC, TUBB, CA12, IGKC, RACGAPl, UBE2C, CD14, IKBKB, S100A7, VEGFA, CD247, KRT5, SEPT8, YBXl, CD3D, MAPK3, SLC2A1, CDKNlA, MAPT, SLC7A8, CHPTl, MLPH, SPONl, CXCL13, MMPl, STATl, CXCL9, MMP7, STC2, DCN, MUCl, STMNl and combinations thereof.

4. Method of claim 3, wherein said tumor is classified into HER2/NEU positive or negative, Luminal and/or Basal / triple negative classes.

5. Method according to claim 3 wherein said at least one marker gene for predicting a response to and/or benefit from chemotherapy in Her 2/neu positive tumors is selected from the group consisting of CHPTl, BCL2, MLPH, SPONl and combinations thereof.

6. Method of any of the preceding claims wherein said classification is performed by determining in a tumor sample the expression of at least one gene indicative for each class and depending on said gene expression, classifying the tumor.

7. A method of claim 6, wherein the expression of said at least one gene indicative for a given class is determined by a immunohistochemistry method

8. The method of any of the preceding claims wherein said gene expression is determined on a RNA level by a PCR based method and/or a microarray based method.

9. A method of any of claims 3 to 8, wherein for Her2/neu positive tumors said at least one marker gene is selected from the group consisting of ERBB4, CHPTl, BCL2 MLPH, and the combinations of CHPT1/ERBB4, and CHPT1/SPON1.

10. A method of any of claims 3 to 9, wherein for luminal tumors said at least one marker gene is selected from the group consisting of CXCL9, MUCl, IGKC, CD3Z, and the combinations of CD3D/MUC1, FRAPl /MUCl, ACOX2/CD3D, ACOX2/CD3Z, and AKR1C3/EGFR.

11. A method of any of claims 3 to 9, wherein for of basal / triple negative tumors said at least one marker gene is selected from the group consisting of TMSL8, ERBB2 (HER2/neu), MUCl and the combinations of STMNl, HER2/neu/STMNl, HER2/neu/TMSL8 , HER2/neu/NFKBIA.

12. A method of any of the above claims, wherein the expression level of no more than five marker genes are determined in a given class.

13. A method of any of the above claims, wherein the expression level of said 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 .

14. Method of any of the above claims, wherein the expression levels of a plurality of marker genes are mathematically combined to give a score indicative of a response to and/or benefit from chemotherapy.

15. A kit for performing the method of any of the preceding claims comprising at least one probe specific for a gene or gene product for each at least one marker gene indicative of a response to chemotherapy for a tumor in each respective class.

16. Use of the Kit of numbered paragraph 5 for performing the method according to any of claims 1 to 14.