AU2012261820A1 - Molecular diagnostic test for cancer - Google Patents

Description

MOLECULAR DIAGNOSTIC TEST FOR CANCER

CROSS-REFERENCE TO RELATED APPLICATIONS

[0001] The present invention claims the priority benefit of U.S. Provisional Patent

Application 61/492,488 filed June 2, 2011, which is incorporated herein by reference.

FIELD OF THE INVENTION

[0002] The present invention relates to a molecular diagnostic test useful for diagnosing cancers from different anatomical sites that includes the use of a common subtype related to angiogenesis. The invention includes the derivation of a gene classification model from gene expression levels. One application is the stratification of response to, and selection of patients for cancer therapeutic drug classes and thus guide patient treatment selection. Another application is the stratification of cancer patients into those that respond and those that do not respond to anti-angiogenic therapeutics. The present invention provides a test that can guide therapy selection as well as selecting patient groups for enrichment strategies during clinical trial evaluation of novel therapeutics. The invention can be used as a prognostic indicator for certain cancers including ovarian cancer, breast cancer, and glioblastoma. The angiogenesis subtype can be identified from fresh/frozen (FF) or formalin fixed paraffin embedded FFPE patient samples.

BACKGROUND

[0003] The pharmaceutical industry continuously pursues new drug treatment options that are more effective, more specific or have fewer adverse side effects than currently administered drugs. Drug therapy alternatives are constantly being developed because genetic variability within the human population results in substantial differences in the effectiveness of many established drugs. Therefore, although a wide variety of drug therapy options are currently available, more therapies are always needed in the event that a patient fails to respond.

[0004] Traditionally, the treatment paradigm used by physicians has been to prescribe a first-line drug therapy that results in the highest success rate possible for treating a disease. Alternative drug therapies are then prescribed if the first is ineffective. This paradigm is clearly not the best treatment method for certain diseases. For example, in diseases such as cancer, the first treatment is often the most important and offers the best opportunity for successful therapy, so there exists a heightened need to choose an initial drug that will be the most effective against that particular patient's disease. [0005] Ovarian cancer is the leading cause of death among all gynecological cancers in western countries. This high death rate is due to the diagnosis at an advanced stage in most patients. Epithelial ovarian cancer (EOC) constitutes 90% of ovarian malignancies and is classified into distinct histologic categories including serous, mucinous, endometrioid, clear cell, transitional, mixed, and undifferentiated subtypes. There is increasing evidence that these differed histologies arise from different aetiologies. There have been recent advances in the methodology used to classify epithelial ovarian cancer (McCluggage, W.G. "Morphological subtypes of ovarian carcinoma: a review with emphasis on new developments and pathogenesis," PATHOLOGY 2011 Aug;43(5):420-32). One of the consequences of this is that many tumors that would previously been classified as endometrioid are now being classified as serous.

[0006] The current standard treatment for ovarian cancer is debulking surgery and standard platinum taxane based cytotoxic chemotherapy. However, not all patients respond to this, and of those that do, approximately 70% will experience a recurrence. Specific targeted therapies for ovarian cancer based on histological or molecular classification have not yet reached the marketplace. Similarly for other types of cancer, there is still no accurate way of selecting appropriate cytotoxic chemotherapeutic agents.

[0007] The advent of microarrays and molecular genomics has the potential for a significant impact on the diagnostic capability and prognostic classification of disease, which may aid in the prediction of the response of an individual patient to a defined therapeutic regimen. Microarrays provide for the analysis of large amounts of genetic information, thereby providing a genetic fingerprint of an individual. There is much enthusiasm that this technology will ultimately provide the necessary tools for custom-made drug treatment regimens.

[0008] Currently, healthcare professionals have few mechanisms to help them identify cancer patients who will benefit from chemotherapeutic agents. Identification of the optimal first-line drug has been difficult because methods are not available for accurately predicting which drug treatment would be the most effective for a particular cancer's physiology. This deficiency results in relatively poor single agent response rates and increased cancer morbidity and death. Furthermore, patients often needlessly undergo ineffective, toxic drug therapy.

[0009] Angiogenesis is a key component of neo-vascularisation of tumors and essential to tumorigenesis and metastatsis. As such, it is a key area for therapeutic intervention and has been correlated to poor prognosis and reduced survival. This has promoted the development of a number of agents that target angiogenesis related processes and pathways, including the market leader and first FDA-approved anti-angiogenic, bevacizumab (Avastin), produced by Genentech/Roche. [0010] Treatment regimens that include bevacizumab have demonstrated broad clinical activity ⁰. However, no overall survival (OS) benefit has been shown after the addition of bevacizumab to cytotoxic chemotherapy in most cancers ⁸' ^12~13. This suggests that a substantial proportion of tumours are either initially resistant or quickly develop resistance to VEGF blockade (the mechanism of action of bevacizumab). In fact, 21% of ovarian, 10% of renal and 33% of rectal cancer patients show partial regression when receiving bevacizumab monotherapy, suggesting that bevacizumab may be active in small subgroups of patients, but that such incremental benefits do not reach significance in unselected patients. As such, the use of a biomarker of response to bevacizumab would improve assessment of treatment outcomes and thus enable the identification of patient subgroups that would receive the most clinical benefit from bevacizumab treatment. This would be particularly relevant in the case of metastatic breast cancer, where the absence of a clinically beneficial biomarker has undermined the use of bevacizumab. Thus far, no such biomarker has been clinically validated to predict bevacizumab efficacy. Hypertension and VEGF polymorphisms are so far the only biomarkers to show potential, but important questions remain about their use in a clinical setting.

[0011] Another approach to anti-angiogenic therapy is simulataneous targeting of multiple angiogenic pathways rather than selective targeting of the VEGF pathway. Theoretically, multitargeted anti-angiogenic agents should more completely inhibit angiogenesis than agents such as bevacizumab and thus may produce greater therapeutic benefit. It has been postulated that in some tumors, angiogenesis may require VEGF only in the early stages of disease but is driven by additional angiogenic pathways as the disease progresses. Therefore, by targeting multiple pathways, it may be possible to counteract compensatory escape mechanisms that could lead to resistance to VEGF inhibition.

[0012] As for other types of cancer there is still no accurate way of selecting which patients will or will not respond to standard of care with an anti-angiogenic therapeutic or single agent anti-angiogenic therapy.

[0013] What is therefore needed is a molecular diagnostic test that would facilitate the stratification of patients based upon their predicted response to anti-angiogenic therapeutics, either in combination with standard of care or as a single-agent therapeutic. This would allow for the rapid identification of those patients who should receive alternative therapies. Such a molecular diagnostic test should be predictive of therapeutic responsiveness across different cancer types with sufficient accuracy. SUMMARY OF THE INVENTION

[0014] Disclosed are methods of using a collection of biomarkers expressed in cancer such that when some or all of the transcripts are over or under-expressed, they identify a subtype of cancer that has an up-regulation in molecular signaling relating to angiogenesis. The invention also provides methods for indicating responsiveness or non-responsiveness to anti- angiogenic agents. In different aspects, this collection of biomarkers may form the basis of a single parameter or multiparametric predictive test that could be delivered using methods known in the art such as microarray, Q-PCR, immunohistochemistry, ELISA or other technologies that can quantify mRNA or protein expression.

[0015] In addition, the cancer sub-type described herein is common to many types of cancer and is not limited to a single cancer disease type. Therefore, the expression signatures disclosed herein may be used to predict responsiveness or non-responsiveness of cancer therapeutics across different cancer types in different tissues. In one embodiment of the invention, these biomarkers are useful for evaluating a cancer tumor's responsiveness to anti- angiogenic therapeutics. In one exemplary embodiment, the cancer is ovarian cancer. In another exemplary embodiment, the cancer is glioblastoma. In another exemplary embodiment, the cancer is breast cancer.

[0016] The invention described herein is not limited to any one drug; it can be used to identify responders and non responders to any of a range of drugs currently in use, under development and novel that directly or indirectly affect or target angiogeneic processes. In one embodiment, the present invention may be used to evaluate adjuvant or neoadjuvant bevacizumab or dasatanib, either as single agents, or in combination with standard of care therapy. In another embodiment, the present invention may be used to evaluate Avastin, VEGF- TRAP, treatment in ovarian cancer.

[0017] The present invention relates to prediction of response to drugs using at least or up to different 10 classifications of response, such as overall survival, progression free survival, radiological response, as defined by RECIST, complete response, partial response, stable disease and serological markers such as, but not limited to, PSA, CEA, CA125, CA15-3 and CA19-9. In cetain embodiments this invention can be used to evaluate survival in ovarian, breat, and glioblastoma.

[0018] In another aspect, the present invention relates to the identification of an angiogenesis sub-type in cancers. The sub-type can be detected by determining the expression levels of certain biomarkers. An expression signature defines a set of biomarkers whose expression is predictive of cancer types that are responsive or non-responsive to anti-angiogenic agents. In certain exemplary embodiments, the expression signature comprises two or more biomarkers selected from the biomarkers listed in Table 1A and IB. In another exemplary embodiment, the expression signature comprises two or more biomarkers selected from the sequences of SEQ ID NOs: 632-801 (Group I) or SEQ ID NOs: 802-974 (Group II). In another exemplary embodiment, the expression signature comprises two or more biomarkers selected from the biomarkers listed in Table 2 A and 2B. In another exemplary embodiment, the expression signature comprises the biomarkers listed in Table 2A and 2B and their corresponding weights as determined using a PLS classifier.

[0019] In another aspect, the present invention relates to kits for conventional diagnostic uses listed above such as qPCR, microarray, and immunoassays such as immunohistochemistry, ELISA, Western blot and the like. Such kits include appropriate reagents and directions to assay the expression of the genes or gene products and quantify mRNA or protein expression.

[0020] Also disclosed are methods for identifying human tumors with or without the angiogenesis phenotype. In certain exemplary embodiments, such methods may be used to identify patients that are sensitive to and respond to drugs that inhibit, either directly or indirectly, processes relating to angiogenesis. In certain other exemplary embodiments, such methods may be used to identify patients that are resistant to or do not respond to drugs that inhibit, either directly or indirectly, processes relating to angiogenesis.

[0021] In another aspect, the invention can be used as a prognostic indicator in certain cancer types. In one exemplary embodiment, the cancer is ovarian cancer. In another exemplary embodiment, the cancer is breast cancer. In yet another exemplary embodiment, the cance is glioblastoma.

[0022] This invention also relates to guiding effective treatment of patients. Further, methods relating to selection of patient treatment regimens and selecting patients for clinical trials of current, or developmental stage drugs that directly or indirectly affect angiogenesis are provided.

[0023] In addition, methods that accommodate the use of archived formalin fixed paraffin-embedded (FFPE) biopsy material, as well as fresh/frozen (FF) tissue, for assay of all transcripts, and are therefore compatible with the most widely available type of biopsy material, are described herein. A biomarker expression level may be determined using RNA obtained from FFPE tissue, fresh frozen tissue or fresh tissue that has been stored in solutions such as RNAlater®.

BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 provides a heatmap representing the hierarchical agglomerative clustering analysis of the most variable genes across 199 serous samples of the Almac Diagnostics' epithelial ovarian cancer sample set. The functional analysis of the probeset clusters is summarized on the right hand side of the image. The legend across the top of the image indicates the classifier group each sample was assigned to for classifier generation (i.e. Class labels).

[0025] Figures 2A and 2B represents the functional analysis results of the angiogenesis probe set cluster in the 199 serous only samples in an epithelial ovarian cancer training set using a functional enrichment analysis. FIG. 2A shows a histogram representing the significance of the top 10 enriched Gene Ontology Biological processes. Red bars indicate significance of a process at a p-value of 0.05 after false discovery rate correction. FIG. 2B represents the subset of the Gene Ontology Biological processes tree where processes include one or more genes encoded by the probe sets in cluster 3. Red coloured processes indicate significance of that process at a p-value of 0.05 after false discovery rate correction. Black coloured processes include one or more genes encoded by the probe sets in cluster 3, but are not significant.

[0026] Figure 3 represents the functional enrichment results of the genes within an exemplary 25 -gene expression signature that identifies the molecular subtype related to angiogenesis. Red bars indicate significance of a process at a p-value of 0.05 after False Discovery Rate correction.

[0027] Figure 4 provides Kaplan-Meier curves for recurrence-free survival (time-to-event in weeks) following initial surgical resection from patients with high-grade glioblastoma without prior treatment (Phillips HS, Kharbanda S, Chen R, Forrest WF et al. "Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis," CANCER CELL 2006 Mar;9(3): 157-73. PMID: 16530701; Costa BM, Smith JS, Chen Y, Chen J et al. "Reversing HOXA9 oncogene activation by PI3K inhibition: epigenetic mechanism and prognostic significance in human glioblastoma," CANCER RES 2010 Jan 15;70(2):453-62. PMID: 20068170).

[0028] Figure 5 provides a diagram of a ROC curve of the classification performance of an exemplary 25 -gene classifier model within 16 prostate cell-lines following treatment with Dasatanib. The AUC is approximately 0.84 following application of the classifier model. The 95% confidence limits were determined using 1000 bootstrap iterations (Wang XD, Reeves K, Luo FR, Xu LA et al. "Identification of candidate predictive and surrogate molecular markers for dasatinib in prostate cancer: rationale for patient selection and efficacy monitoring," GENOME BIOL 2007;8(11):R255. PMID: 18047674).

[0029] Figure 6 provides a heatmap representing the hierarchical agglomerative clustering analysis of the most variable genes across 265 serous samples of an epithelial ovarian cancer sample set reclassified according to updated pathological classification criteria. The functional analysis of the probeset clusters is summarized on the right hand side of the image. The legend across the top of the image indicates the classifier group each sample was assigned to for classifier generation (i.e. Class labels).

[0030] Figures 7A and 7B represents the functional analysis results of the angiogenesis probeset of the 265 serous only samples in an epithelial ovarian cancer training set using a functional enrichment tool (FET) algorithm. FIG. 7A shows a histogram representing the significance of the top 10 enriched Gene Ontology biological processes. Red bars indicate significance of a process at a p-value of 0.05 after False Discovery Rate correction. FIG. 7B represents the subset of the Gene Ontology Biological processes tree where processes include one or more genes encoded by the probe sets in cluster 2 (angiogenesis). Red coloured processes indicate significance of that process at a p-value of 0.05 after false discovery rate correction. Black coloured processes include one or more genes encoded by the probe sets in the angiogenesis cluster, but are not significant.

[0031] Figure. 8 represents the functional enrichment results of the genes within an exemplary 45 -gene classifier model that identifies the molecular sub-type related to angiogenesis. Red bars indicate significance of a process at a p-value of 0.05 after False Discovery Rate correction.

[0032] Figure 9 provides a diagram of a ROC curve of the classification performance of the 45 -gene classifier model within 16 prostate cell-lines following treatment with Dasatanib. The AUC is -0.95 following application of the classifier model. The 95% confidence limits were determined using 1000 bootstrap iterations.

[0033] Figure 10 provides a heatmap representing the hierarchical agglomerative clustering analysis of the most variable genes across 265 serous samples of an epithelial ovarian cancer sample set reclassified according to updated pathological classification criteria. The functional analysis of the probeset clusters is summarized on the right hand side of the image. The legend across the top of the image indicates the classifier group each sample should be assigned to for generation of a non-angiognesis or non-responsive group classifier.

[0034] Figure 11 provides Kaplan-Meier curves for progression-free survival (in weeks) for the non-angiogenesis sample group (FIG 3 a, sample cluster 1) versus the angiogenesis sample group (Figure 10, sample clusters 2 and 3) in the reclassified ovarian sample set.

[0035] Figures 12A and 12B provide heatmaps representing the hierarchical agglomerative clustering analysis of the most variable genes across 51 ER negative samples (Figure 12 A) of a breast cancer sample set and the most variable genes across 56 ER positive samples of a breast cancer sample set (Figure 12B). The functional analysis of the probe set clusters is summarized on the right hand side of the image for those cluster showing vasculature development/angiogenesis or immune response/IFN signalling.

SEQUENCE LISTING

[0036] The nucleic and amino acid sequences listed in the accompanying sequence listing are shown using standard letter abbreviations for nucleotide bases, as defined in 37 C.F.R. § 1.822. Only one strand of each nucleic acid sequence is shown, but the complementary strand is understood as included by any reference to the displayed strand.

[0037] The Sequence Listing is submitted as an ASCII text file in the form of the file named ADL_0801WP_ST25.txt, which was created on June 4, 2012 and is 352,839 bytes, which is incorporated by reference herein.

DETAILED DESCRIPTION OF THE INVENTION

[0038] Unless defined otherwise, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Definitions of common terms in molecular biology may be found in Benjamin Lewin, Genes IX, published by Jones and Bartlet, 2008 (ISBN 0763752223); Kendrew et al. (eds.), The Encyclopedia of Molecular Biology, published by Blackwell Science Ltd., 1994 (ISBN 0632021829); Robert A. Meyers (ed.), Molecular Biology and Biotechnology: a Comprehensive Desk Reference, published by VCH Publishers, Inc., 1995 (ISBN 9780471185710); Singleton et al., Dictionary of Microbiology and Molecular Biology 2nd ed., J. Wiley & Sons (New York, N.Y. 1994), and March, Advanced Organic Chemistry Reactions, Mechanisms and Structure 4th ed., John Wiley & Sons (New York, N.Y. 1992).

[0039] The singular terms "a," "an," and "the" include plural referents unless context clearly indicates otherwise. Similarly, the word "or" is intended to include "and" unless the context clearly indicates otherwise. The term "comprises" means "includes." In case of conflict, the present specification, including explanations of terms, will control.

[0040] As used herein terms "marker panel," "expression classifier," "classifier,"

"expression signature," or "signature" may be used interchangeably.

[0041] All publications, published patent documents, and patent applications cited in this application are indicative of the level of skill in the art(s) to which the application pertains. All publications, published patent documents, and patent applications cited herein are hereby incorporated by reference to the same extent as though each individual publication, published patent document, or patent application was specifically and individually indicated as being incorporated by reference. Overview

[0042] A major goal of current research efforts in cancer is to increase the efficacy of perioperative systemic therapy in patients by incorporating molecular parameters into clinical therapeutic decisions. Pharmacogenetics/genomics is the study of genetic/genomic factors involved in an individuals' response to a foreign compound or drug. Agents or modulators which have a stimulatory or inhibitory effect on expression of a biomarker of the invention can be administered to individuals to treat (prophylactically or therapeutically) cancer in the patient. It is ideal to also consider the pharmacogenomics of the individual in conjunction with such treatment. Differences in metabolism of therapeutics may possibly lead to severe toxicity or therapeutic failure by altering the relationship between dose and blood concentration of the pharmacologically active drug. Thus, understanding the pharmacogenomics of an individual permits the selection of effective agents (e.g., drugs) for prophylactic or therapeutic treatments. Such pharmacogenomics can further be used to determine appropriate dosages and therapeutic regimens. Accordingly, the level of expression of a biomarker of the invention in an individual can be determined to thereby select appropriate agent(s) for therapeutic or prophylactic treatment of the individual.

[0043] The present invention relates to a molecular diagnostic tests useful for diagnosing cancers from different anatomical sites that includes the use of a common subtype related to angiogenesis. The invention includes expression signatures that identify a subject as responsive or non-responsive to anti-angiogenic therapeutics. The expression signature is derived by obtaining the expression profiles of samples from a sample set of known pathology and/or clinical outcome. The samples may originate from the same sample tissue type or different tissue types.. As used herein an "expression profile" comprises a set of values representing the expression level for each biomarker analyzed from a given sample.

[0044] The expression profiles from the sample set are then analyzed using a mathematical model. Different mathematical models may be applied and include, but are not limited to, models from the fields of pattern recognition (Duda et al. Pattern Classification, 2^nd ed., John Wiley, New York 2001), machine learning (Scholkopf et al. Learning with Kernels, MIT Press, Cambridge 2002, Bishop, Neural Networks for Pattern Recognition, Clarendon Press, Oxford 1995), statistics (Hastie et al. The Elements of Statistical Learning, Springer, New York 2001), bioinformatics (Dudoit et al, 2002, J. Am. Statist. Assoc. 97:77-87, Tibshirani et al, 2002, Proc. Natl. Acad. Sci. USA 99:6567-6572) or chemometrics (Vandeginste, et al, Handbook of Chemometrics and Qualimetrics, Part B, Elsevier, Amsterdam 1998). The mathematical model identifies one or more biomarkers expressed in the sample set that are most predictive of a given disease phenotype. These one ore more biomarkers define an expression signature. Accordingly, an expression signature includes the biomarkers identified as most predictive of a given disease phenotype. In certain exemplary embodiments, the mathematical model defines a variable, such as a weight, for each identified biomarker. In certain exemplary embodiments, the mathematical model defines a decision function. The decision function may further define a threshold score which separates the sample set into two disease phenotypes such as, but not limited to, samples that are responsive and non-responsive to anti-angiogenic therapeutics. In one exemplary embodiment, the decision function and expression signature are defined using a linear classifier.

[0045] To classify new samples using a defined expression signature, the biomarkers defined by the expression signature are isolated and an expression profile of the biomarker(s) determined. The new sample biomarker expression profile is analyzed with the same mathematical model used to define the expression signature. In certain exemplary embodiments, the mathematical model defines an expression score for the new sample. The expression score may be determined by combining the expression values of the biomarkers with corresponding scalar weights using non-linear, algebraic, trigonometric or correlative means to derive a single scalar value. The expression score is compared to the threshold score and the sample classified as responsive or non-responsive to anti-angiogenic therapeutics. In one exemplary embodiment, a sample expression value greater than the reference expression value indicates a patient will be responsive to an anti-angiogenic therapeutic. In another exemplary embodiment, a sample expression score below the threshold score indicates the patient will not be responsive to an anti-angiogenic therapeutic. In another exemplary embodiment, a sample expression score below the threshold expression score indicates the patient has a cancer type, or is at risk of developing a cancer type, that is not responsive to an anti-angiogenic therapeutic. In another exemplary embodiment, a sample expression score above the reference expression score indicates the patient has a cancer type, or is at risk of developing a cancer type, that is responsive to an anti-angiogenic therapeutic. Where an expression signature is derived from a tissue sample set comprising one type of cancer tissue, the expression signature is not limited to identifying the same cancer sub-type only in tissues of the same cancer type, but may be used with other cancer types that share the same cancer sub-type. For example, where an expression signature is derived from ovarian cancer samples, the expression signature may be used to identify a similar angiogenesis sub-type in different cancers such as glioblastoma or breast cancer.

[0046] One application of the expression signatures disclosed herein is the stratification of response to, and selection of patients for therapeutic drug classes that encompass anti- angiogenic therapies. By examining the expression of a collection of the identified biomarkers in a tumor, it is possible to determine which therapeutic agent or combination of agents will be most likely to reduce the growth rate of a cancer. It is also possible to determine which therapeutic agent or combination of agents will be the least likely to reduce the growth rate of a cancer. By examining the expression of a collection of biomarkers, it is therefore possible to eliminate ineffective or inappropriate therapeutic agents. Importantly, in certain embodiments, these determinations can be made on a patient-by-patient basis or on an agent-by-agent basis. Thus, one can determine whether or not a particular therapeutic regimen is likely to benefit a particular patient or type of patient, and/or whether a particular regimen should be continued. The present invention provides a test that can guide therapy selection as well as selecting patient groups for enrichment strategies during clinical trial evaluation of novel therapeutics. For example, when evaluating a putative anti-angiogeneic agent or treatment regime, the expression signatures and methods disclosed herein may be used to select individuals for clinical trials that have cancer types that are responsive to anti-angiogenic agents. The angiogenesis subtype can be identified from a fresh/frozen (FF) or formalin fixed paraffin embedded (FFPE) patient sample. In one exemplary embodiment, the cancer type is ovarian cancer. In another exemplary embodiment, the cancer type is a glioblastoma. In a further exemplary embodiment, the cancer type is breast cancer.

[0047] A cancer is "responsive" to a therapeutic agent if its rate of growth is inhibited as a result of contact with the therapeutic agent, compared to its growth in the absence of contact with the therapeutic agent. Growth of a cancer can be measured in a variety of ways. For instance, the size of a tumor or measuring the expression of tumor markers appropriate for that tumor type.

[0048] A cancer is "non-responsive" to a therapeutic agent if its rate of growth is not inhibited, or inhibited to a very low degree, as a result of contact with the therapeutic agent when compared to its growth in the absence of contact with the therapeutic agent. As stated above, growth of a cancer can be measured in a variety of ways, for instance, the size of a tumor or measuring the expression of tumor markers appropriate for that tumor type. The quality of being non-responsive to a therapeutic agent is a highly variable one, with different cancers exhibiting different levels of "non-responsiveness" to a given therapeutic agent, under different conditions. Still further, measures of non-responsiveness can be assessed using additional criteria beyond growth size of a tumor such as, but not limited to, patient quality of life, and degree of metastases.

Identifying Expression Signatures

[0049] The expression signatures of the present invention are identified by analyzing the expression profiles of certain biomarkers in a patient sample set. Biomarkers suitable for use in the present invention include DNA, RNA, and proteins. The biomarkers are isolated from a patient sample and their expression levels determined to derive a set of expression profiles for each sample analyzed in the patient sample set.

a. Expression Profiles

[0050] In certain embodiments, the expression profile obtained is a genomic or nucleic acid expression profile, where the amount or level of one or more nucleic acids in the sample is determined. In these embodiments, the sample that is assayed to generate the expression profile employed in the diagnostic or prognostic methods is one that is a nucleic acid sample. The nucleic acid sample includes a population of nucleic acids that includes the expression information of the phenotype determinative biomarkers of the cell or tissue being analyzed. In some embodiments, the nucleic acid may include RNA or DNA nucleic acids, e.g., mRNA, cRNA, cDNA etc., so long as the sample retains the expression information of the host cell or tissue from which it is obtained. The sample may be prepared in a number of different ways, as is known in the art, e.g., by mRNA isolation from a cell, where the isolated mRNA is used as isolated, amplified, or employed to prepare cDNA, cRNA, etc., as is known in the field of differential gene expression. Accordingly, determining the level of mRNA in a sample includes preparing cDNA or cRNA from the mRNA and subsequently measuring the cDNA or cRNA. The sample is typically prepared from a cell or tissue harvested from a subject in need of treatment, e.g., via biopsy of tissue, using standard protocols, where cell types or tissues from which such nucleic acids may be generated include any tissue in which the expression pattern of the to be determined phenotype exists, including, but not limited to, disease cells or tissue, body fluids, etc.

[0051] The expression profile may be generated from the initial nucleic acid sample using any convenient protocol. While a variety of different manners of generating expression profiles are known, such as those employed in the field of differential gene expression/biomarker analysis, one representative and convenient type of protocol for generating expression profiles is array-based gene expression profile generation protocols. Such applications are hybridization assays in which a nucleic acid that displays "probe" nucleic acids for each of the genes to be assayed/profiled in the profile to be generated is employed. In these assays, a sample of target nucleic acids is first prepared from the initial nucleic acid sample being assayed, where preparation may include labeling of the target nucleic acids with a label, e.g., a member of a signal producing system. Following target nucleic acid sample preparation, the sample is contacted with the array under hybridization conditions, whereby complexes are formed between target nucleic acids that are complementary to probe sequences attached to the array surface. The presence of hybridized complexes is then detected, either qualitatively or quantitatively. Specific hybridization technology which may be practiced to generate the expression profiles employed in the subject methods includes the technology described in U.S. Pat. Nos. 5,143,854; 5,288,644; 5,324,633; 5,432,049; 5,470,710; 5,492,806; 5,503,980; 5,510,270; 5,525,464; 5,547,839; 5,580,732; 5,661,028; 5,800,992; the disclosures of which are herein incorporated by reference; as well as WO 95/21265; WO 96/31622; WO 97/10365; WO 97/27317; EP 373 203; and EP 785 280. In these methods, an array of "probe" nucleic acids that includes a probe for each of the biomarkers whose expression is being assayed is contacted with target nucleic acids as described above. Contact is carried out under hybridization conditions, e.g., stringent hybridization conditions as described above, and unbound nucleic acid is then removed. The resultant pattern of hybridized nucleic acids provides information regarding expression for each of the biomarkers that have been probed, where the expression information is in terms of whether or not the gene is expressed and, typically, at what level, where the expression data, i.e., expression profile, may be both qualitative and quantitative.

b. Diseases and Sample Tissue Sources

[0052] In certain exemplary embodiments, the patient sample set comprises cancer tissue samples, such as archived samples. The patient sample set is preferably derived from cancer tissue samples having been characterized by prognosis, likelihood of recurrence, long term survival, clinical outcome, treatment response, diagnosis, cancer classification, or personalized genomics profile. As used herein cancer includes, but is not limited to, leukemia, brain cancer, prostate cancer, liver cancer, ovarian cancer, stomach cancer, colorectal cancer, throat cancer, breast cancer, skin cancer, melanoma, lung cancer, sarcoma, cervical cancer, testicular cancer, bladder cancer, endocrine cancer, endometrial cancer, esophageal cancer, glioma, lymphoma, neuroblastoma, osteosarcoma, pancreatic cancer, pituitary cancer, renal cancer, and the like. In one embodiment, the methods described herein refer to cancers that are treated with anti- angiogenic agents, anti-angiogenic targeted therapies, inhibitors of angiogenesis signaling, but not limited to these classes. These cancers also include subclasses and subtypes of these cancers at various stages of pathogenesis. In certain exemplary embodiments the patient sample set comprises ovarian cancer samples. In another exemplary embodiment, the patient sample set comprises, breast cancer samples. In yet another exemplary embodiment, the patient sample set comprises glioblastoma samples.

[0053] "Biological sample", "sample", and "test sample" are used interchangeably herein to refer to any material, biological fluid, tissue, or cell obtained or otherwise derived from an individual. This includes blood (including whole blood, leukocytes, peripheral blood mononuclear cells, buffy coat, plasma, and serum), sputum, tears, mucus, nasal washes, nasal aspirate, breath, urine, semen, saliva, meningeal fluid, amniotic fluid, glandular fluid, lymph fluid, nipple aspirate, bronchial aspirate, synovial fluid, joint aspirate, ascites, cells, a cellular extract, and cerebrospinal fluid. This also includes experimentally separated fractions of all of the preceding. For example, a blood sample can be fractionated into serum or into fractions containing particular types of blood cells, such as red blood cells or white blood cells (leukocytes). If desired, a sample can be a combination of samples from an individual, such as a combination of a tissue and fluid sample. The term "biological sample" also includes materials containing homogenized solid material, such as from a stool sample, a tissue sample, or a tissue biopsy, for example. The term "biological sample" also includes materials derived from a tissue culture or a cell culture. Any suitable methods for obtaining a biological sample can be employed; exemplary methods include, e.g., phlebotomy, swab (e.g., buccal swab), and a fine needle aspirate biopsy procedure. Samples can also be collected, e.g., by micro dissection (e.g., laser capture micro dissection (LCM) or laser micro dissection (LMD)), bladder wash, smear (e.g., a PAP smear), or ductal lavage. A "biological sample" obtained or derived from an individual includes any such sample that has been processed in any suitable manner after being obtained from the individual, for example, fresh frozen or formalin fixed and/or paraffin embedded.

[0054] As used herein, the term "patient" includes human and non-human animals. The preferred patient for treatment is a human. "Patient" and "subject" are used interchangeably herein.

c. Biomarkers

[0055] As used herein, the term "biomarker" can refer to a gene, an mR A, cDNA, an antisense transcript, a miR A, a polypeptide, a protein, a protein fragment, or any other nucleic acid sequence or polypeptide sequence that indicates either gene expression levels or protein production levels. When a biomarker indicates or is a sign of an abnormal process, disease or other condition in an individual, that biomarker is generally described as being either over- expressed or under-expressed as compared to an expression level or value of the biomarker that indicates or is a sign of a normal process, an absence of a disease or other condition in an individual. "Up-regulation", "up-regulated", "over-expression", "over-expressed", and any variations thereof are used interchangeably to refer to a value or level of a biomarker in a biological sample that is greater than a value or level (or range of values or levels) of the biomarker that is typically detected in similar biological samples from healthy or normal individuals. The terms may also refer to a value or level of a biomarker in a biological sample that is greater than a value or level (or range of values or levels) of the biomarker that may be detected at a different stage of a particular disease. [0056] "Down-regulation", "down-regulated", "under-expression", "under-expressed", and any variations thereof are used interchangeably to refer to a value or level of a biomarker in a biological sample that is less than a value or level (or range of values or levels) of the biomarker that is typically detected in similar biological samples from healthy or normal individuals. The terms may also refer to a value or level of a biomarker in a biological sample that is less than a value or level (or range of values or levels) of the biomarker that may be detected at a different stage of a particular disease.

[0057] Further, a biomarker that is either over-expressed or under-expressed can also be referred to as being "differentially expressed" or as having a "differential level" or "differential value" as compared to a "normal" expression level or value of the biomarker that indicates or is a sign of a normal process or an absence of a disease or other condition in an individual. Thus, "differential expression" of a biomarker can also be referred to as a variation from a "normal" expression level of the biomarker.

[0058] The terms "differential biomarker expression" and "differential expression" are used interchangeably to refer to a biomarker whose expression is activated to a higher or lower level in a subject suffering from a specific disease, relative to its expression in a normal subject, or relative to its expression in a patient that responds differently to a particular therapy or has a different prognosis. The terms also include biomarkers whose expression is activated to a higher or lower level at different stages of the same disease. It is also understood that a differentially expressed biomarker may be either activated or inhibited at the nucleic acid level or protein level, or may be subject to alternative splicing to result in a different polypeptide product. Such differences may be evidenced by a variety of changes including mR A levels, miRNA levels, antisense transcript levels, or protein surface expression, secretion or other partitioning of a polypeptide. Differential biomarker expression may include a comparison of expression between two or more genes or their gene products; or a comparison of the ratios of the expression between two or more genes or their gene products; or even a comparison of two differently processed products of the same gene, which differ between normal subjects and subjects suffering from a disease; or between various stages of the same disease. Differential expression includes both quantitative, as well as qualitative, differences in the temporal or cellular expression pattern in a biomarker among, for example, normal and diseased cells, or among cells which have undergone different disease events or disease stages.

[0059] In certain exemplary embodiments, the biomarker is an RNA transcript. As used herein "RNA transcript" refers to both coding and non-coding RNA, including messenger RNAs (mRNA), alternatively spliced mRNAs, ribosomal RNA (rRNA), transfer RNA (tRNA), small nuclear RNAs (snRNA), and antisense RNA. Measuring mRNA in a biological sample may be used as a surrogate for detection of the level of the corresponding protein and gene in the biological sample. Thus, any of the biomarkers or biomarker panels described herein can also be detected by detecting the appropriate R A. Methods of biomarker expression profiling include, but are not limited to quantitative PCR, NGS, northern blots, southern blots, microarrays, SAGE, immunoassays (ELISA, EIA, agglutination, nephelometry, turbidimetry, Western blot, immunoprecipitation, immunocytochemistry, flow cytometry, Luminex assay), and mass spectrometry. The overall expression data for a given sample may be normalized using methods known to those skilled in the art in order to correct for differing amounts of starting material, varying efficiencies of the extraction and amplification reactions.

[0060] In certain exemplary embodiments, biomarkers useful for distinguishing between cancer types that are responsive and non-responsive to anti-angiogenic therapeutics can be determined by identifying biomarkers exhibiting the highest degree of variability between samples in the patient data set as determined using the expression detection methods and patient sample sets discussed above. Standard statistical methods known in the art for identifying highly variable data points in expression data may be used to identify the highly variable biomarkers. For example, a combined background and variance filter to the patient data set. The background filter is based on the selection of probe sets with expression E and expression variance va¾ above the thresholds defined by background standard deviation oBg (from the Expression Console software) and quantile of the standard normal distribution z_a at a specified significance a probesets were kept if:

E > log₂((z_ao Bg)); log₂((va¾) > 2 [log₂(o_Bg) - E - log₂(log(2))]

[0061] where a defines a significance threshold. In certain exemplary embodiment, the significance threshold is 6.3 · 10^~5. In another exemplary embodiment, the significance threshold may be between 1.0 · 10^"7 to 1.0 · 10^"3.

[0062] In certain exemplary embodiments, the highly variable biomarkers may be further analyzed to group samples in the patient data set into sub-types or clusters based on similar gene expression profiles. For examples, biomarkers may be clustered based on how highly correlated the up-regulation or down-regulation of their expression is to one another. Various clustering analysis techniques known in the art may be used. In one exemplary embodiment, hierarchical agglomerative clustering is used to identify the cancer sub-types. To determine the biological relevance of each sub-type, the biomarkers within each cluster may be further mapped to their corresponding genes and annotated by cross-reference to one or more gene ontology databases containing information on biological activity and biological pathways associated with the gene. In one exemplary embodiment, biomarker in clusters enriched for angiogenesis, vasculature development and immune response general functional terms are grouped into a putative angiogenesis sample group and used for expression signature generation. In another exemplary embodiment, biomarkers in clusters that are up regulated and enriched for angiogeneis, vasculature development and immune response general functional terms are grouped into a putative angiongenesis sample group and used for expression signature generation. In another exemplary embodiment, biomarkers in clusters that are down regulated and enriched for angiogenesis, vasculature development and immune response general functional terms are grouped into a putative angiongenesis sample group and used for expression signature generation. Further details for conducting functional analysis of biomarker clusters is provided in the Examples section below.

[0063] In one exemplary embodiment, the biomarkers useful in deriving an expression signature for distinguishing cancer sub-types that are, or are not, responsive to anti-angiogenic therapeutics include those biomarkers listed in Table 1A, Table IB, or both. In another exemplary embodiment, the biomarkers useful in deriving an expression signature for distinguishing cancer sub-types that are, or are not, responsive to anti-angiogenic therapeutics include those biomarkers listed in Group I (comprising SEQ ID NOs: 632-801) or Group II (comprising SEQ ID NOs: 802-974), or both. These biomarkers are identified as having predictive value to determine a patient response to a therapeutic agent. Their expression correlates with the response, or lack thereof, to an agent, and more specifically, an anti- angiogenic therapeutic agent. By examining the expression of a collection of the identified biomarkers in a tumor, it is possible to determine which therapeutic agent or combination of agents will be most likely to reduce the growth rate of a cancer. By examining a collection of identified biomarkers in a tumor, it is also possible to determine which therapeutic agent or combination of agents will be the least likely to reduce the growth rate of a cancer. By examining the expression of a collection of biomarkers, it is therefore possible to eliminate ineffective or inappropriate therapeutic agents. Importantly, in certain embodiments, these determinations can be made on a patient-by-patient basis or on an agent-by-agent basis. Thus, one can determine whether or not a particular therapeutic regimen is likely to benefit a particular patient or type of patient, and/or whether a particular regimen should be continued.

Table 1 A: An io enesis and immune res onse cluster enes of Fi 1

5 Sense MARCKS

6 Sense AL0X5

7 Sense C0L8A1

8 Sense CTSS

9 Sense HAVCR2

10 Sense RAB31

11 Sense KCNAB2

12 Sense THBS1

13 Sense SH3BP4

14 Sense CTGF

15 Sense CTGF

16 Sense VCAN

17 Sense IGKC

18 Sense IGKC

19 Sense IGKC

20 Sense SGK1

21 Sense NFATC1

22 Sense HMHA1

FCGR1C /// FCGR1A ///

23 Sense

FCGR1B

24 Sense EDA2R

25 Sense COL8A1

26 Sense COL12A1

27 Sense HLA-B

28 Sense HLA-F

29 Sense HLA-C

30 Sense SPP1

31 Sense MY01F

32 Sense SPARC

33 Sense SPARC

34 Sense UBD /// GABBRl

35 Sense CCND1

36 Sense COL1A1

37 Sense EGR1 38 Sense EGR1

39 Sense TNFSF10

40 Sense SULF2

41 Sense CERCAM

42 Sense ATF3

43 Sense MIR21

44 Sense BASP1

45 Sense IFIT2

46 Sense SULF1

47 Sense IGLC2 /// IGLC3

48 Sense IGLC2 /// IGLC3

49 Sense (Fully OExonic) IGLC2 /// IGLC3

50 Sense IGLC2 /// IGLC3

51 Sense IGLC2 /// IGLC3

52 Sense IGLCl

53 Sense IGLCl

54 Sense IGLC2 /// IGLC3

55 Sense ANGPTL2

56 Sense COL5A2

57 Sense IGJ

58 Sense THY1

59 Sense NDN

60 Sense RGS2

61 Sense MEIS3P1 /// MEIS3P2

62 Sense GBP2

63 Sense CSF1R

64 Sense C1R

65 Sense FAT1

66 Sense COL1A1

67 Sense RHOB

68 Sense MMP11

69 Sense GADD45B

70 Sense MMP14

71 Sense MMP14 72 Sense IGHG4

73 Sense DDX60L

74 Sense SPP1

75 Sense ROR2

76 Sense CTSK

77 Sense FCGR2B

78 Sense PTAFR

79 Sense ICAM1

80 Sense HCLS1

81 No Transcript Match —

82 Sense SLFN11

83 No Transcript Match —

84 Sense JAM3

85 Sense TMEM49

86 Sense TMEM49

87 Sense LTBP2

88 Sense IRS1

89 Sense COL5A2

90 Sense C17orf91

91 Sense GPNMB

92 Sense FAM198B

93 Sense MICAL2

94 Sense TMEM2

95 Sense CHST15

96 Sense SECTM1

97 Sense DCN

98 Sense VCAM1

99 Sense TNFAIP3

100 Sense C1QA

101 Sense C1QA

102 Sense FBX032

103 Sense COL12A1

104 Sense CPE

105 Sense CIITA 106 Sense GAS 7

107 Sense C0L3A1

108 Sense FN1

109 Sense IFI30

110 Sense ITGB2

111 Sense ELN

112 Sense CMTM3

113 Sense ANTXR1

114 Sense ARHGDIB

115 Sense LAPTM5

116 Sense S0X4

117 Sense IFI44L

118 Sense IL4I1

119 Sense ANTXR2

120 Sense IGLC2 /// IGLC3

121 Sense EPSTI1

122 Sense BIRC3

123 Sense IGLC2 /// IGLC3

124 Sense BST2

125 Sense TNFSF10

126 Sense COL10A1

127 Sense IGLC2 /// IGLC3

128 Sense FBP1

129 Sense RHOBTB3

130 Sense CDK6

131 Sense CD74

132 Sense ISM1

133 Sense C1QC

134 Sense BIN2

135 Sense CSRNP1

136 Sense TYROBP

137 Sense C1QTNF3

138 Sense DCN

139 Sense IGFBP4 140 Sense AOAH

141 Sense SIRPA

142 Sense FOSB

143 Sense CCDC80

144 Sense IGLCl

145 Sense HCST

146 Sense IFI35

147 Sense BIRC3

148 Sense COL3A1

149 Sense IFITM2

150 Sense ZFP36

151 Sense MMP11

152 Sense COL1A2

153 Sense HLA-DPA1

154 Sense TWIST 1

155 Sense ZNF154

156 Sense EGR1

157 Sense IGLC2 /// IGLC3

158 Sense TNFSF10

159 Sense IGKC

160 Sense IGHGl /// IGHG4

161 Sense GBP5

162 Sense COL1A2

163 Sense APOC1

164 No Transcript Match —

165 Sense COL3A1

166 AntiSense PXDN

167 AntiSense EGR1

168 Sense GBP3

169 Sense ISG15

170 Sense —

171 Sense KIAA0146

172 Sense CMAH

173 Sense APBB2 174 Sense TPM1

175 No Transcript Match —

176 Sense DMD

177 No Genome Match —

178 Sense IL10RA

179 Sense —

180 No Transcript Match —

181 Sense DUSP1

182 Sense GBP1

183 Sense PARVG

184 Sense MAFF

185 Sense PDGFC

186 Sense MSN

187 Sense RSAD2

188 Sense TPM1

189 Sense EMB

190 Sense C6orfl55

191 Sense FOS

192 Sense DEXI

193 Sense RNF19A

194 Sense FBX032

195 Sense DPYSL3

196 Sense PRICKLE 1

197 AntiSense EGR1

198 AntiSense NRP2

199 Sense B2M

200 AntiSense MIR21

201 Sense MMP2

202 Sense CDR1

203 Sense HLA-B

204 Sense CTGF

205 Sense DCN

206 Sense SOD2

207 Sense FN1 208 Sense COL8A2

209 Sense SGK1

210 Sense TIMP3

211 Sense ACTA2

212 Sense SRGN

213 Sense L0XL1

214 Sense CCR1

215 Sense GBP1

216 Sense CDH11

217 Sense FCGR3A

218 Sense LUM

219 Sense NNMT

220 Sense COL1A2

221 Sense RGS1

222 Sense GJA1

223 Sense SPARCL1

224 Sense DAB2

225 AntiSense CTHRC1

226 Sense RGS16

227 Sense FBLN1

228 Sense SPP1

229 Sense CTSB

230 Sense SPP1

231 Sense SDC1

232 Sense PLAU

233 Sense PDGFRA

234 Sense SERPINFl

235 Sense BGN

236 Sense COL6A3

237 AntiSense C3

238 AntiSense C3

239 AntiSense SPP1

240 AntiSense HLA-DQAl

241 AntiSense GAS1 242 Sense VCAN

243 AntiSense —

IGHG4 /// IGHG2 /// IGHGl ///

244 Sense

IGHGP

245 Sense IGHG2

246 Sense C3orf26

247 AntiSense ATF3

248 AntiSense ATF3

249 AntiSense SULF1

250 Sense FN1

251 AntiSense CALD1

252 AntiSense CALD1

253 Sense TMEM49

254 Sense TMEM49

255 Sense CHD5

256 AntiSense EGR1

257 AntiSense SNAI2

258 AntiSense ITPRIPL2

259 AntiSense GABBRl /// UBD

260 AntiSense GABBRl /// UBD

261 AntiSense TWIST 1

262 AntiSense TWIST 1

263 AntiSense BATF2

264 AntiSense NFKBIZ

265 Sense C3orf26

266 AntiSense LOXL1

267 Sense —

268 AntiSense TIMP2

269 AntiSense FN1

270 AntiSense COL1A1

271 AntiSense DCN

272 Sense TREH

273 AntiSense UBE2L6

274 AntiSense APOL1 275 AntiSense BIRC3

276 AntiSense BIRC3

277 Sense LILRB4

278 Sense FGD2

279 Sense TMEM49

280 Sense NCF4

281 Sense COL10A1

282 Sense GAL3ST4

283 Sense HCK

284 Sense TAGLN

285 Sense TWIST 1

286 Sense HCLS1

287 Sense LPAR6

288 Sense ITGB2

289 Sense LST1

290 Sense HLA-B

291 Sense C17orf91

292 Sense ZC3H12A

293 Sense KLF10

294 Sense BASP1

295 Sense BASP1

Table IB: Angiogenesis and immune response cluster genes of Fig 6

SEQ NO: Orientation Gene Symbol

2 Sense (Fully Exonic) PDGFC

3 Sense (Fully Exonic) TGFB3

4 Sense (Fully Exonic) RAC2

5 Sense (Fully Exonic) MARCKS

6 Sense (Fully Exonic) ALOX5

7 Sense (Fully Exonic) COL8A1

11 Sense (Fully Exonic) KCNAB2

12 Sense (Fully Exonic) THBS1

14 Sense (Fully Exonic) CTGF

15 Sense (Fully Exonic) CTGF

16 Sense (Fully Exonic) VCAN

18 Sense (Fully Exonic) IGKC

19 Sense (Fully Exonic) IGKC

21 Sense (includes Intronic) NFATC1

22 Sense (Fully Exonic) HMHA1

23 Sense (Fully Exonic) FCGR1B 24 Sense (Fully Exonic) EDA2R

25 Sense (Fully Exonic) COL8A1

26 Sense (Fully Exonic) COL12A1

27 Sense (Fully Exonic) HLA-B

28 Sense HLA-F

37 Sense (Fully Exonic) EGR1

40 Sense (Fully Exonic) SULF2

41 Sense (Fully Exonic) CERCAM

42 Sense (Fully Exonic) ATF3

43 Sense (Fully Exonic) MIR21

45 Sense (Fully Exonic) IFIT2

47 Sense (Fully Exonic) IGLC3

48 Sense (Fully Exonic) IGLC3

49 Sense (Fully Exonic) IGLC3

50 Sense (Fully Exonic) IGLC3

51 Sense (Fully Exonic) IGLC3

54 Sense (Fully Exonic) IGLC3

55 Sense (Fully Exonic) ANGPTL2

56 Sense (Fully Exonic) COL5A2

58 Sense (Fully Exonic) THY1

59 Sense (Fully Exonic) NDN

60 Sense (Fully Exonic) RGS2

61 Sense (Fully Exonic) MEIS3P2

62 Sense (Fully Exonic) GBP2

65 Sense (Fully Exonic) FAT1

66 Sense (Fully Exonic) COL1A1

68 Sense (Fully Exonic) MMP11

69 Sense (Fully Exonic) GADD45B

71 Sense (Fully Exonic) MMP14

72 Sense (Fully Exonic) IGHG4

80 Sense (Fully Exonic) HCLS1

83 No Transcript match

84 Sense (Fully Exonic) JAM3

86 Sense (Fully Exonic) TMEM49

87 Sense (Fully Exonic) LTBP2

88 Sense (Fully Exonic) IRS1

90 Sense (Fully Exonic) C17orf91

91 Sense (Fully Exonic) GPNMB

92 Sense (Fully Exonic) FAM198B

95 Sense (Fully Exonic) CHST15

97 Sense (Fully Exonic) DCN

98 Sense (Fully Exonic) VCAM1

105 Sense (Fully Exonic) CIITA

106 Sense (Fully Exonic) GAS 7

107 Sense (Fully Exonic) COL3A1

110 Sense (Fully Exonic) ITGB2

111 Sense (Fully Exonic) ELN

112 Sense (Fully Exonic) CMTM3

113 Sense (Fully Exonic) ANTXR1

118 Sense (Fully Exonic) IL4I1 119 Sense (Fully Exonic) ANTXR2

120 Sense (Fully Exonic) IGLC2 /// IGLC3

123 Sense (Fully Exonic) IGLC3

124 Sense (Fully Exonic) BST2

126 Sense (Fully Exonic) COL10A1

127 Sense (Fully Exonic) IGLC3

128 Sense (Fully Exonic) FBP1

129 Sense (Fully Exonic) RHOBTB3

131 Sense (Fully Exonic) CD74

132 Sense (Fully Exonic) ISM1

135 Sense (Fully Exonic) CSRNP1

138 Sense (Fully Exonic) DCN

139 Sense (Fully Exonic) IGFBP4

143 Sense (Fully Exonic) CCDC80

148 Sense (Fully Exonic) COL3A1

150 Sense (Fully Exonic) ZFP36

151 Sense (Fully Exonic) MMP11

152 Sense (Fully Exonic) COL1A2

153 Sense (Fully Exonic) HLA-DPA1

154 Sense (Fully Exonic) TWIST 1

155 Sense (Fully Exonic) ZNF154

157 Sense (Fully Exonic) IGLC3

159 Sense (Fully Exonic) IGKC

160 Sense (Fully Exonic) IGHG1

162 Sense (Fully Exonic) COL1A2

163 Sense (Fully Exonic) APOC1

167 AntiSense EGR1

171 Sense (Fully Exonic) KIAA0146

174 Sense (Fully Exonic) TPM1

176 Sense (includes Intronic) DMD

180 No Transcript match

181 Sense (Fully Exonic) DUSP1

182 Sense (Fully Exonic) GBP1

185 Sense (includes Intronic) PDGFC

186 Sense (includes Intronic) MSN

188 Sense (includes Intronic) TPM1

189 Sense (Fully Exonic) EMB

191 Sense (Fully Exonic) FOS

195 Sense (includes Intronic) DPYSL3

197 AntiSense EGR1

198 AntiSense NRP2

201 Sense (Fully Exonic) MMP2

204 Sense (Fully Exonic) CTGF

211 Sense (Fully Exonic) ACTA2

213 Sense (Fully Exonic) LOXL1

216 Sense (Fully Exonic) CDH11

218 Sense (Fully Exonic) LUM

219 Sense (Fully Exonic) NNMT

222 Sense (Fully Exonic) GJA1

225 AntiSense CTHRC1 229 Sense (Fully Exonic) CTSB

232 Sense (Fully Exonic) PLAU

233 Sense (Fully Exonic) PDGFRA

242 Sense (Fully Exonic) VCAN

243 AntiSense —

244 Sense (Fully Exonic) IGHG4 /// IGHG2 /// IGHGl

///I GHGP

245 Sense (Fully Exonic) IGHG2

246 Sense (includes Intronic) C3orf26

247 AntiSense ATF3

248 AntiSense ATF3

250 Sense (Fully Exonic) FN1

251 AntiSense CALD1

252 AntiSense CALD1

256 AntiSense EGR1

261 AntiSense TWIST 1

262 AntiSense TWIST 1

263 AntiSense BATF2

264 AntiSense NFKBIZ

265 Sense (includes Intronic) C3orf26

266 AntiSense LOXL1

267 Sense (includes Intronic) —

269 AntiSense FN1

270 AntiSense COL1A1

272 Sense (Fully Exonic) TREH

274 AntiSense APOL1

281 Sense (Fully Exonic) COL10A1

282 Sense (Fully Exonic) GAL3ST4

284 Sense (Fully Exonic) TAGLN

285 Sense (Fully Exonic) TWIST 1

286 Sense (Fully Exonic) HCLS1

288 Sense (Fully Exonic) ITGB2

290 Sense (Fully Exonic) HLA-B

291 Sense (Fully Exonic) C17orf91

296 Sense (Fully Exonic) FBLIM1

297 Sense (Fully Exonic) COL15A1

298 Sense (Fully Exonic) AQP7P3

299 AntiSense IGFBP5

300 Sense (Fully Exonic) FANK1

301 AntiSense INS

302 Sense (Fully Exonic) COL27A1

303 Sense (Fully Exonic) COL5A1

304 Sense (Fully Exonic) PRICKLE2

305 Sense (Fully Exonic) N/A

306 Sense (Fully Exonic) GXYLT2

307 Sense (includes Intronic) KLF12

308 No Transcript match

309 Sense (Fully Exonic) FBX032

310 No Transcript match

311 Sense (Fully Exonic) ASAH2B 312 AntiSense PPFIBP1

313 AntiSense XIST

314 Sense (Fully Exonic) IGFBP6

315 Sense (Fully Exonic) ROBOl

316 Sense (Fully Exonic) TPM1

317 AntiSense N/A

318 AntiSense PLEKHG1

319 Sense (Fully Exonic) NR2F1

320 Sense (Fully Exonic) NPDC1

321 AntiSense INS

322 Sense (Fully Exonic) TRAF5

323 Sense (Fully Exonic) CALD1

324 Sense (includes Intronic) CHRM3

325 Sense (Fully Exonic) AMOTL1

326 Sense (includes Intronic) COL12A1

327 Sense (Fully Exonic) PLXNA4

328 Sense (includes Intronic) TMEM43

329 Sense (includes Intronic) PvORA

330 AntiSense INS

331 Sense (Fully Exonic) TSPAN18

332 No Transcript match

333 Sense (Fully Exonic) TNC

334 Sense (Fully Exonic) TYRO 3

335 AntiSense EFNA5

336 Sense (Fully Exonic) MYL9

337 Sense (Fully Exonic) MIR198

338 Sense (includes Intronic) N/A

339 Sense (includes Intronic) PLA2R1

340 Sense (Fully Exonic) COL14A1

341 Sense (Fully Exonic) NRP1

342 Sense (Fully Exonic) FSCN1

343 Sense (includes Intronic) PDGFD

344 No Transcript match

345 Sense (includes Intronic) DOCK4

346 Sense (Fully Exonic) TRIM 13

347 Sense (Fully Exonic) IGFBP5

348 Sense (Fully Exonic) C19orf63

349 AntiSense KLF6

350 AntiSense TRIO

351 Sense (Fully Exonic) COL4A1

352 Sense (Fully Exonic) EPDR1

353 Sense (Fully Exonic) FNDC1

354 Sense (Fully Exonic) IL1R1

355 Sense (Fully Exonic) CES4

356 Sense (Fully Exonic) GPR176

357 Sense (includes Intronic) GXYLT2

358 AntiSense WHSC1L1

359 Sense (Fully Exonic) N/A

360 Sense (Fully Exonic) RGN

361 Sense (includes Intronic) CA3 362 Sense (Fully Exonic) TIMP3

363 Sense (Fully Exonic) EFNA5

364 Sense (Fully Exonic) RASGRF2

365 Sense (includes Intronic) RELL1

366 AntiSense ACSS3

367 Sense (Fully Exonic) STMN3

368 Sense (Fully Exonic) N/A

369 AntiSense C7orf29

370 Sense (Fully Exonic) HOXC6

371 Sense (Fully Exonic) KLF8

372 Sense (includes Intronic) SERINC5

373 Sense (Fully Exonic) AKT3

374 Sense (Fully Exonic) TGFB2

375 AntiSense WNT5A

376 No Transcript match

377 No Transcript match

378 AntiSense IGFBP7

379 No Transcript match

380 Sense (includes Intronic) SULT1C4

381 Sense (Fully Exonic) AASS

382 Sense (Fully Exonic) HEPH

383 Sense (Fully Exonic) ADH5

384 Sense (Fully Exonic) TIMP2

385 Sense (Fully Exonic) EMP1

386 Sense (Fully Exonic) CXCL14

387 Sense (Fully Exonic) ZNF548

388 Sense (Fully Exonic) SGCB

389 Sense (includes Intronic) ASH2L

390 Sense (includes Intronic) SERINC5

391 No Genome match

392 Sense (Fully Exonic) TMEM159

393 Sense (includes Intronic) RBMS3

394 Sense (Fully Exonic) TMEM49

395 Sense (includes Intronic) PvORA

396 No Transcript match

397 AntiSense ZNF608

398 No Genome match

399 Sense (Fully Exonic) ADAMTS2

400 Sense (Fully Exonic) APCDD1

401 AntiSense GXYLT2

402 Sense (Fully Exonic) XIST

403 Sense (Fully Exonic) MBNL2

404 Sense (Fully Exonic) SHF

405 Sense (includes Intronic) APBB2

406 No Transcript match

407 Sense (Fully Exonic) COL14A1

408 Sense (Fully Exonic) IGFBP5

409 Sense (Fully Exonic) CREB5

410 AntiSense INS

411 Sense (Fully Exonic) BAHCC1 412 Sense (Fully Exonic) RFXAP

413 Sense (Fully Exonic) INS

414 Sense (Fully Exonic) DDR2

415 Sense (Fully Exonic) CA12

416 Sense (Fully Exonic) RHOB

417 Sense (Fully Exonic) N/A

418 Sense (Fully Exonic) SNORD 116-4

419 Sense (Fully Exonic) MEG3

420 Sense (Fully Exonic) WNT4

421 Sense (Fully Exonic) FBLN2

422 AntiSense DAAM1

423 No Transcript match

424 Sense (Fully Exonic) CHN1

425 Sense (includes Intronic) APBB2

426 Sense (Fully Exonic) PTRF

427 AntiSense IGF1

428 Sense (Fully Exonic) UST

429 Sense (Fully Exonic) SMARCA1

430 Sense (includes Intronic) N/A

431 Sense (Fully Exonic) IGLC3

432 AntiSense INS

433 Sense (Fully Exonic) KANK4

434 AntiSense IGF1

435 Sense (Fully Exonic) CYP27A1

436 AntiSense EIF2B5

437 No Transcript match

438 Sense (Fully Exonic) SNRNP25

439 Sense (Fully Exonic) SETD7

440 Sense (Fully Exonic) MSX1

441 Sense (Fully Exonic) HOPX

442 Sense (Fully Exonic) NID2

443 Sense (Fully Exonic) IGF1

444 Sense (Fully Exonic) PSD3

445 Sense (Fully Exonic) FGFR1

446 Sense (Fully Exonic) ETV1

447 Sense (Fully Exonic) ZNF655

448 No Genome match

449 AntiSense INS

450 Sense (Fully Exonic) SFRP2

451 Sense (Fully Exonic) SPAG16

452 AntiSense NR2F2

453 Sense (includes Intronic) SYNP02

454 Sense (Fully Exonic) FAM101B

455 AntiSense IGF2

456 Sense (Fully Exonic) CA3

457 Sense (Fully Exonic) XIST

458 No Transcript match

459 Sense (Fully Exonic) WNT7A

460 Sense (includes Intronic) N/A

461 Sense (Fully Exonic) FGFR1 462 AntiSense FXYD6

463 Sense (Fully Exonic) FGFR1

464 Sense (includes Intronic) IGFBP7

465 Sense (Fully Exonic) TIMP2

466 Sense (Fully Exonic) DUSP1

467 Sense (includes Intronic) SERINC5

468 No Transcript match

469 Sense (Fully Exonic) ABLIM1

470 Sense (Fully Exonic) ARL4A

471 AntiSense SH3TC2

472 AntiSense NR2F2

473 Sense (Fully Exonic) ENG

474 Sense (Fully Exonic) MGP

475 Sense (Fully Exonic) MEG3

476 AntiSense FAM115A

477 Sense (Fully Exonic) EGR1

478 Sense (Fully Exonic) SNORD 116-3

479 Sense (Fully Exonic) AEBP1

480 Sense (includes Intronic) SDK1

481 Sense (Fully Exonic) ENC1

482 Sense (Fully Exonic) SNORD 116-7

483 Sense (Fully Exonic) N/A

484 Sense (Fully Exonic) APOD

485 AntiSense N/A

486 AntiSense GAS1

487 Sense (Fully Exonic) VPS36

488 No Transcript match

489 Sense (Fully Exonic) SPHK2

490 Sense (Fully Exonic) SNORD 116-8

491 Sense (Fully Exonic) MYO10

492 Sense (Fully Exonic) HOXC6

493 Sense (Fully Exonic) RNF149

494 Sense (Fully Exonic) BTG2

495 Sense (includes Intronic) MAP3K1

496 Sense (Fully Exonic) SNORD 116-23

497 Sense (includes Intronic) ACSL4

498 Sense (Fully Exonic) CYP27C1

499 Sense (includes Intronic) COL12A1

500 Sense (Fully Exonic) IGFBP5

501 Sense (Fully Exonic) DUSP4

502 Sense (Fully Exonic) PFKFB3

503 Sense (Fully Exonic) SDC2

504 AntiSense FXYD6

505 Sense (Fully Exonic) COL5A1

506 Sense (Fully Exonic) MARCKS

507 Sense (Fully Exonic) IRS2

508 Sense (Fully Exonic) N/A

509 AntiSense FSCN1

510 Sense (Fully Exonic) FYN

511 Sense (Fully Exonic) IGFBP5 512 Sense (Fully Exonic) NUDT4P1

513 Sense (Fully Exonic) NFKBIZ

514 Sense (Fully Exonic) N/A

515 Sense (Fully Exonic) C7orf41

516 Sense (Fully Exonic) MEG3

517 Sense (Fully Exonic) N/A

518 Sense (Fully Exonic) PLEKHG1

519 Sense (Fully Exonic) ZNF827

520 Sense (Fully Exonic) ZNF175

521 Sense (Fully Exonic) XIST

522 Sense (includes Intronic) GSN

523 Sense (includes Intronic) RORA

524 Sense (Fully Exonic) CA13

525 AntiSense TMX4

526 Sense (Fully Exonic) KIT

527 Sense (includes Intronic) WDR78

528 Sense (Fully Exonic) ECEL1

529 Sense (Fully Exonic) XIST

530 Sense (Fully Exonic) PROCR

531 Sense (Fully Exonic) C9orfl67

532 Sense (Fully Exonic) MUC6

533 Sense (includes Intronic) P4HA2

534 Sense (Fully Exonic) FAM69C

535 Sense (Fully Exonic) NOX4

536 Sense (includes Intronic) N/A

537 No Transcript match

538 Sense (Fully Exonic) SMOX

539 Sense (Fully Exonic) KIAA0922

540 No Transcript match

541 Sense (Fully Exonic) XIST

542 Sense (Fully Exonic) NPAS2

543 Sense (Fully Exonic) NAV1

544 Sense (includes Intronic) N/A

545 Sense (Fully Exonic) HLA-A

546 Sense (Fully Exonic) FAM46C

547 Sense (Fully Exonic) N/A

548 Sense (Fully Exonic) SLAMF7

549 Sense (Fully Exonic) FCER1G

550 Sense (Fully Exonic) CIS

551 Sense (Fully Exonic) NUPR1

552 AntiSense C1QC

553 AntiSense SAT1

554 Sense (Fully Exonic) SOD2

555 Sense (Fully Exonic) IRF1

556 Sense (Fully Exonic) SFN

557 AntiSense LTB

558 Sense (Fully Exonic) ARID5A

559 Sense (Fully Exonic) BST2

560 Sense (Fully Exonic) HLA-F

561 Sense (Fully Exonic) XAF1 562 Sense (Fully Exonic) TCOF1

563 Sense (Fully Exonic) RPL23AP1

564 Sense (Fully Exonic) IL1RN

565 Sense (Fully Exonic) IFIT5

566 Sense (Fully Exonic) B2M

567 AntiSense GBP1

568 Sense (Fully Exonic) HLA-F

569 Sense (Fully Exonic) DGKA

570 Sense (Fully Exonic) XBP1

571 Sense (Fully Exonic) PLCG2

572 Sense (Fully Exonic) FAM46C

573 No Genome match

574 Sense (Fully Exonic) TREM2

575 Sense (Fully Exonic) LGALS9

576 Sense (Fully Exonic) HLA-DPB1

577 AntiSense ODF3B

578 Sense (Fully Exonic) MX1

579 Sense (Fully Exonic) STAT1

580 Sense (Fully Exonic) CTSB

581 Sense (Fully Exonic) FAM26F

582 Sense (includes Intronic) PARP14

583 AntiSense SAT1

584 Sense (Fully Exonic) CTSS

585 No Transcript match

586 Sense (Fully Exonic) CTSB

587 Sense (Fully Exonic) ADAM8

588 Sense (includes Intronic) B2M

589 Sense (Fully Exonic) FLVCR2

590 Sense (Fully Exonic) TYROBP

591 AntiSense SAMD9L

592 Sense (Fully Exonic) SAMD9L

593 Sense (Fully Exonic) SIGLECl

594 Sense (Fully Exonic) MMP7

595 Sense (Fully Exonic) APOL1

596 Sense (Fully Exonic) CYLD

597 Sense (Fully Exonic) HLA-B

598 Sense (Fully Exonic) SAT1

599 Sense (Fully Exonic) C1QB

600 Sense (Fully Exonic) HLA-DMB

601 Sense (Fully Exonic) NLRC5

602 Sense (Fully Exonic) FAM20A

603 AntiSense N/A

604 Sense (Fully Exonic) STAT1

605 Sense (includes Intronic) STAT1

606 Sense (Fully Exonic) STAT1

607 AntiSense N/A

608 Sense (Fully Exonic) DERL3

609 Sense (Fully Exonic) HLA-F

610 Sense (Fully Exonic) MAFB

611 Sense (Fully Exonic) CD4 612 Sense (Fully Exonic) HLA-A

613 Sense (Fully Exonic) UBE2L6

614 Sense (Fully Exonic) C1QC

615 Sense (Fully Exonic) CD 163

616 Sense (Fully Exonic) LRMP

617 Sense (Fully Exonic) Cl lorfl7

618 Sense (Fully Exonic) XAF1

619 Sense (Fully Exonic) GLRX

620 Sense (Fully Exonic) IFIH1

621 Sense (Fully Exonic) CD44

622 Sense (Fully Exonic) LITAF

623 Sense (Fully Exonic) CCDC69

624 Sense (Fully Exonic) GBP5

625 Sense (Fully Exonic) PML

626 Sense (Fully Exonic) SAMD9

627 Sense (Fully Exonic) CBR3

628 Sense (Fully Exonic) RASGRP2

629 Sense (Fully Exonic) FCGR2A

630 Sense (Fully Exonic) BST2

631 Sense (Fully Exonic) HLA-A

[0064] In certain exemplary embodiments, all or a portion of the biomarkers recited in

Table 1A and Table IB, may be used in an expression signature. In certain other exemplary embodiments, all or a portion of the biomarkers recited in Group I or Group II may be used in an expression signature. For example, expression signatures comprising the biomarkers in Table 1A, Table IB, Group I, and Group II, can be generated using the methods provided herein and can comprise between one, and all of the markers set forth in Tables 1A, IB, Group I, Group II and each and every combination in between (e.g., four selected markers, 16 selected markers, 74 selected markers, etc.). In some embodiments, the expression signature comprises at least 5, 10, 20, 40, 60, 100, 150, 200, or 300 or more markers. In other embodiments, the predictive biomarker panel comprises no more than 5, 10, 20, 40, 60, 100, 150, 200, 300, 400, 500, 600 or 700 markers. In one exemplary embodiment, the expression signature includes a plurality of markers listed in Table 1A. In another exemplary embodiment, the expression signature includes a plurality of biomarkers listing in Table IB. In yet another exemplary embodiment, the expression signature includes a plurality of biomarkers listed in Table 1A and Table IB. In another exemplary embodiment, the expression signature includes a plurality of biomarkers listed in Group I. In another exemplary embodiment, the expression signature includes a plurality of biomarkers listed in table Group II. In yet another exemplary embodiment, the expression signature includes a plurality of biomarkers listed in Groups I and II. In some embodiments the expression signature includes at least about 1%, about 5%, about 10%, about 20%, about 30%, about 40%, about 50%, about 60%, about 70%, about 80%, about 90%, about 95%, about 96%, about 97%, about 98%, or about 99% of the markers listed in Table 1A, Table IB, Group I, Group II, or a combination thereof. Selected expression signatures can be assembled from the biomarkers provided using methods described herein and analogous methods known in the art. In one embodiment, the expression signature contains all 250 genes or gene products in Table 1A. In another exemplary embodiment, the expression signature contains all 486 genes or gene products in Table IB. In another exemplary embodiment the expression signature comprises SEQ ID NOs: 632-801. In aother exemplary embodiment, the expression signature comprises SEQ ID NOs: 802-974.

4. Mathematical Models

[0065] The following methods may be used to derive expression signatures for distinguishing between subjects that are responsive or non-responsive to anti-angiogenic therapeutics, or as prognostic indicators of certain cancer types, including expression signatures derived from the biomarkers disclosed above. In certain other exemplary embodiments, the expression signature is derived using a decision tree (Hastie et al. The Elements of Statistical Learning, Springer, New York 2001), a random forest (Breiman, 2001 Random Forests, Machine Learning 45:5), a neural network (Bishop, Neural Networks for Pattern Recognition, Clarendon Press, Oxford 1995), discriminant analysis (Duda et al. Pattern Classification, 2nd ed., John Wiley, New York 2001), including, but not limited to linear, diagonal linear, quadratic and logistic discriminant analysis, a Prediction Analysis for Microarrays (PAM, (Tibshirani et al, 2002, Proc. Natl. Acad. Sci. USA 99:6567-6572)) or a Soft Independent Modeling of Class Analogy analysis. (SIMCA, (Wold, 1976, Pattern Recogn. 8: 127-139)).

[0066] Biomarker expression values may be defined in combination with corresponding scalar weights on the real scale with varying magnitude, which are further combined through linear or non-linear, algebraic, trigonometric or correlative means into a single scalar value via an algebraic, statistical learning, Bayesian, regression, or similar algorithms which together with a mathematically derived decision function on the scalar value provide a predictive model by which expression profiles from samples may be resolved into discrete classes of responder or non-responder, resistant or non-resistant, to a specified drug, drug class, or treatment regimen. Such predictive models, including biomarker membership, are developed by learning weights and the decision threshold, optimized for sensitivity, specificity, negative and positive predictive values, hazard ratio or any combination thereof, under cross-validation, bootstrapping or similar sampling techniques, from a set of representative expression profiles from historical patient samples with known drug response and/or resistance.

[0067] In one embodiment, the biomarkers are used to form a weighted sum of their signals, where individual weights can be positive or negative. The resulting sum ("expression score") is compared with a pre-determined reference point or value. The comparison with the reference point or value may be used to diagnose, or predict a clinical condition or outcome.

[0068] As described above, one of ordinary skill in the art will appreciate that the biomarkers included in the classifier provided in Tables 1A, IB, Group I and Group II will carry unequal weights in a classifier for responsiveness or resistance to a therapeutic agent. Therefore, while as few as one sequence may be used to diagnose or predict an outcome such as responsiveness to therapeutic agent, the specificity and sensitivity or diagnosis or prediction accuracy may increase using more sequences.

[0069] As used herein, the term "weight" refers to the relative importance of an item in a statistical calculation. The weight of each biomarker in a gene expression classifier may be determined on a data set of patient samples using analytical methods known in the art.

[0070] In certain exemplary embodiments, the expression signature is defined by a decision function. A decision function is a set of weighted expression values derived using a linear classifier. All linear classifiers define the decision function using the following equation:

f(x) = w'•x + b =∑Wi ' x_i +b (1)

[0071] All measurement values, such as the microarray gene expression intensities x for a certain sample are collected in a vector x. Each intensity is then multiplied with a corresponding weight Wj to obtain the value of the decision function f(x) after adding an offset term b. In deriving the decision function, the linear classifier will further define a threshold value that splits the gene expression data space into two disjoint halves. Exemplary linear classifiers include but are not limited to partial least squares (PLS), (Nguyen et al, Bioinformatics 18 (2002) 39-50), support vector machines (SVM) (Scholkopf et al, Learning with Kernels, MIT Press, Cambridge 2002), and shrinkage discriminant analysis (SDA) (Ahdesmaki et al., Annals of applied statistics 4, 503-519 (2010)). In one exemplary embodiment, the linear classifier is a PLS linear classifier.

[0072] The decision function is empirically derived on a large set of training samples, for example from patients showing responsiveness or resistance to a therapeutic agent. The threshold separates a patient group based on different characteristics such as, but not limited to, responsiveness/non-responsiveness to treatment. The interpretation of this quantity, i.e. the cutoff threshold responsiveness or resistance to a therapeutic agent, is derived in the development phase ("training") from a set of patients with known outcome. The corresponding weights and the responsiveness/resistance cut-off threshold for the decision score are fixed a priori from training data by methods known to those skilled in the art. In one exemplary embodiment, Partial Least Squares Discriminant Analysis (PLS-DA) is used for determining the weights. (L. Stahle, S. Wold, J. Chemom. 1 (1987) 185-196; D. V. Nguyen, D.M. Rocke, Bioinformatics 18 (2002) 39-50).

[0073] Effectively, this means that the data space, i.e. the set of all possible combinations of biomarker expression values, is split into two mutually exclusive halves corresponding to different clinical classifications or predictions, for example, one corresponding to responsiveness to a therapeutic agent and the other to non-responsiveness. In the context of the overall classifier, relative over-expression of a certain biomarker can either increase the decision score (positive weight) or reduce it (negative weight) and thus contribute to an overall decision of, for example, responsiveness or resistance to a therapeutic agent.

[0074] In certain exemplary embodiments of the invention, the data is transformed non- linearly before applying a weighted sum as described above. This non-linear transformation might include increasing the dimensionality of the data. The non-linear transformation and weighted summation might also be performed implicitly, for example, through the use of a kernel function. (Scholkopf et al. Learning with Kernels, MIT Press, Cambridge 2002).

[0075] In certain exemplary embodiments, the patient training set data is derived by isolated RNA from a corresponding cancer tissue sample set and determining expression values by hybridizing the isolated RNA to a microarray. In certain exemplary embodiments, the microarray used in deriving the expression signature is a transcriptome array. As used herein a "transcriptome array" refers to a microarray containing probe sets that are designed to hybridize to sequences that have been verified as expressed in the diseased tissue of interest. Given alternative splicing and variable poly-A tail processing between tissues and biological contexts, it is possible that probes designed against the same gene sequence derived from another tissue source or biological context will not effectively bind to transcripts expressed in the diseased tissue of interest, leading to a loss of potentially relevant biological information. Accordingly, it is beneficial to verify what sequences are expressed in the disease tissue of interest before deriving a microarray probe set. Verification of expressed sequences in a particular disease context may be done, for example, by isolating and sequencing total RNA from a diseased tissue sample set and cross-referencing the isolated sequences with known nucleic acid sequence databases to verify that the probe set on the transcriptome array is designed against the sequences actually expressed in the diseased tissue of interest. Methods for making transcriptome arrays are described in United States Patent Application Publication No. 2006/0134663, which is incorporated herein by reference. In certain exemplary embodiments, the probe set of the transcriptome array is designed to bind within 300 nucleotides of the 3' end of a transcript. Methods for designing transcriptome arrays with probe sets that bind within 300 nucleotides of the 3 ' end of target transcripts are disclosed in United States Patent Application Publication No. 2009/0082218, which is incorporated by reference herein. In certain exemplary embodiments, the microarray used in deriving the gene expression profiles of the present invention is the Almac Ovarian Cancer DSA™ microarray (Almac Group, Craigavon, United Kingdom).

[0076] An optimal linear classifier can be selected by evaluating a linear classifier's performance using such diagnostics as "area under the curve" (AUC). AUC refers to the area under the curve of a receiver operating characteristic (ROC) curve, both of which are well known in the art. AUC measures are useful for comparing the accuracy of a classifier across the complete data range. Linear classifiers with a higher AUC have a greater capacity to classify unknowns correctly between two groups of interest (e.g., ovarian cancer samples and normal or control samples). ROC curves are useful for plotting the performance of a particular feature (e.g., any of the biomarkers described herein and/or any item of additional biomedical information) in distinguishing between two populations (e.g., individuals responding and not responding to a therapeutic agent). Typically, the feature data across the entire population (e.g., the cases and controls) are sorted in ascending order based on the value of a single feature. Then, for each value for that feature, the true positive and false positive rates for the data are calculated. The true positive rate is determined by counting the number of cases above the value for that feature and then dividing by the total number of cases. The false positive rate is determined by counting the number of controls above the value for that feature and then dividing by the total number of controls. Although this definition refers to scenarios in which a feature is elevated in cases compared to controls, this definition also applies to scenarios in which a feature is lower in cases compared to the controls (in such a scenario, samples below the value for that feature would be counted). ROC curves can be generated for a single feature as well as for other single outputs, for example, a combination of two or more features can be mathematically combined (e.g., added, subtracted, multiplied, etc.) to provide a single sum value, and this single sum value can be plotted in a ROC curve. Additionally, any combination of multiple features, in which the combination derives a single output value, can be plotted in a ROC curve. These combinations of features may comprise a test. The ROC curve is the plot of the true positive rate (sensitivity) of a test against the false positive rate (1 -specificity) of the test.

[0077] In one exemplary embodiment the expression signature is directed to the 25 biomarkers detailed in Table 2A with corresponding ranks and weights detailed in the table or alternative rankings and weightings, depending, for example, on the disease setting. In another exemplary embodiment, the expression signature is directed to the 45 biomarkers detailed in Table 2B with corresponding ranks and weights detailed in the table or alternative rankings and weightings, depending, for example, on the disease setting. Tables 2A and 2B rank the biomarkers in order of decreasing weight in the classifier, defined as the rank of the average weight in the compound decision score function measured under cross-validation.

Table 2A Gene symbols and corresponding ranked weights for a 25 gene signature

Table 2B Gene symbols and corresponding ranked weights for a 45 gene signature

6 BICC1 0.0304

7 CDH11 0.0283

8 MRVI1 0.0256

9 PMP22 0.0240

10 COL1 1A1 -0.0237

11 IGFL2 0.0222

12 LUM -0.0220

13 NTM -0.0218

14 BGN 0.021 1

15 COL3A1 -0.0210

16 COL10A1 0.0197

17 RAB31 0.0180

18 ANGPTL2 0.0166

19 PLAU 0.0166

20 COL8A1 0.0164

21 MIR1245 0.0153

22 POLD2 0.0146

23 N D2 0.0145

24 FZD1 0.0143

25 COPZ2 0.0139

26 ITGA5 0.0136

27 VGLL3 0.0125

28 INHBA -0.0118

29 MMP14 0.0110

30 VCAN 0.0100

31 THBS2 -0.0087

32 RUNX2 0.0083

33 TIMP3 0.0081

34 SFRP2 -0.0079

35 COL1A2 0.0078

36 COL5A2 -0.0072

37 SERPINFl 0.0068

38 IF26B -0.0052

39 TNFAIP6 0.0050

40 MMP2 0.0040

41 FN1 0.0031

42 ALP 2 0.0024

43 CTS 0.0015

44 LOXL1 -0.0014

45 FAP 0.0000 [0078] In one exemplary embodiment, the expression signature comprises one or more biomarkers selected from the group consisting of: CCDC80, INHBA, THBS2, SFRP2, MMP2, PLAU, FAP, FN1, COL8A1, RAB31, FAM38B, VCAN, GJB2, ITGA5, CRISPLD2, C17, f91, BGN, TIMP3, ALPK2, LUM, NKD2, LOX, MIR1245, LOXL1, and CXCL12.

[0079] In another exemplary embodiment, the expression signature comprises at least

CCDC80, INHBA, THBS2 and SFRP2 and at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, or 21. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarkers CCDC80, INHBA, THBS2 and SFRP2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 or 21. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker CCDC80 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23 or 24. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker INHBA and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23 or 24. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker THBS2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23 or 24. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker SFRP2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23 or 24.

[0080] In another exemplary embodiment, the expression signature comprises one or more biomarkers selected from the group consisting of: TMEM200A, GJB2, MMP13, GFPT2, POSTN, BICC1, CDH11, MRVI1, PMP22, COL11A1, IGFL2, LUM, NTM, BGN, COL3A1, COL10A1, RAB31, ANGPTL2, PLAU, COL8A1, MIR1245, POLD2, NKD2, FZD1, COPZ2, ITGA5, VGLL3, INHBA, MMP14, VCAN, THBS2, RUNX2, TIMP3, SFRP2, COL1A2, COL5A2, SERPINFl, KIF26B, TNFAIP6, MMP2, FN1, ALPK2, CTSK, LOXL1 and FAP.

[0081] In another exemplary embodiment, the expression signature comprises at least

TMEM200A, GJB2, MMP13 and GFPT2 and at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, or 41. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarkers TMEM200A, GJB2, MMP13 and GFPT2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2B, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40 or 41. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker TMEM200A and one of at least N additional biomarkers selected from the list of biomarkers in Table 2B, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker GJB2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2a, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27,

28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker MMP13 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2B, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker GFPT2 and one of at least N additional biomarkers selected from the list of biomarkers in Table 2B, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18,

29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44.

[0082] In one exemplary embodiment, the expression signature comprises one or more biomarkers selected from the group consisting of: ALPK2, BGN, COL8A1, FAP, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN.

[0083] In another exemplary embodiment, the expression signature comprises at least

ALPK2, BGN, COL8A1, FAP and at least N additional biomarkers selected from the list of biomarkers FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarkers ALPK2, BGN, COL8A1 and FAP and one of at least N additional biomarkers selected from the list of biomarkers FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker ALPK2 and one of at least N additional biomarkers selected from the list of biomarkers BGN, COL8A1, FAP, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker BGN and one of at least N additional biomarkers selected from the list of biomarkers ALPK2, COL8A1, FAP, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker COL8A1 and one of at least N additional biomarkers selected from the list of biomarkers ALPK2, BGN, , FAP, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. In a further aspect, therapeutic responsiveness is predicted in an individual by conducting an assay on a biological sample from the individual and detecting biomarker values that each correspond to the biomarker FAP and one of at least N additional biomarkers selected from the list of biomarkers ALPK2, BGN, COL8A1, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3 and VCAN, wherein N equals 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14 or 15. [0084] In one exemplary embodiment, the expression signature comprises a set of biomarkers that are down regulated. In one exemplary embodiment the expression signature comprises at least GJB2, INHBA, THBS2, SFRP2, PLAU and at least N additional biomarkers from Table 1 A or Table IB, wherein N is 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, or 70. In another exemplary embodiment, the expression signature comprises at least GJB2 and at least N additional biomarkers from Table 1A or Table IB, wherein N is 2, 3, 4, 5, 6, 7, 8, 9,

10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35,

36, 37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60,

61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, or 74. In another exemplary embodiment, the expression signature comprises INHBA and at least N additional biomarkers from Table 1A or Table IB, wherein N is . 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, or 74. In another exemplary embodiment, the expression signature comprises THBS2 and at least N additional biomarkers from Table 1A or Table IB, wherein N is 2, 3, 4, 5, 6, 7, 8, 9, 10,

11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36,

37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61,

62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, or 74. In another exemplary embodiment, the expression signature comprises at least SFRP2 wherein N is 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, or 74. In another exemplary embodiment, the expression signature comprises PLAU and at least N additional biomarkers from Table 1A or Table IB, wherein N is 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 29, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34,35, 36, 37, 38, 39, 40, 41, 42, 43 or 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, or 70. In another exemplary embodiment the expression signature includes GJB2, INHBA, THBS2, SFRP2, PLAU and at least about 1%, about 5%, about 10%, about 20%, about 30%>, about 40%>, about 50%>, about 60%, about 70%, about 80%, about 90%, about 95%, or about 99% of the biomarkers listed in Table 1 A, Table IB, or a combination thereof.

Classifying New Test Samples using an Expression Signature

[0085] To classify new test samples using an expression signature, such as those described above, the relative expression levels of biomarkers in a cancer tissue are measured to form a test sample expression profile. In certain exemplary embodiments, the test sample expression profile is summarized in the form of a compound decision score ("expression score") and compared to a threshold score that is mathematically derived from a training set of patient data. The score threshold separates a patient group based on different characteristics such as, but not limited to, responsiveness/non-responsiveness to treatment. The patient training set data is preferably derived from cancer tissue samples having been characterized by prognosis, likelihood of recurrence, long term survival, clinical outcome, treatment response, diagnosis, cancer classification, or personalized genomics profile. Expression profiles, and corresponding decision scores from patient samples may be correlated with the characteristics of patient samples in the training set that are on the same side of the mathematically derived score decision threshold. The threshold of the linear classifier scalar output is optimized to maximize the sum of sensitivity and specificity under cross-validation as observed within the training dataset.

[0086] The overall expression data for a given sample is normalized using methods known to those skilled in the art in order to correct for differing amounts of starting material, varying efficiencies of the extraction and amplification reactions, etc.

[0087] In one embodiment, the biomarker expression profile of a patient tissue sample is evaluated by a linear classifier. As used herein, a linear classifier refers to a weighted sum of the individual biomarker intensities into a compound decision score ("decision function"). The decision score is then compared to a pre-defined cut-off score threshold, corresponding to a certain set-point in terms of sensitivity and specificity which indicates if a sample is above the score threshold (decision function positive) or below (decision function negative).

[0088] Using a linear classifier on the normalized data to make a diagnostic or prognostic call (e.g. responsiveness or resistance to therapeutic agent) effectively means to split the data space, i.e. all possible combinations of expression values for all genes in the classifier, into two disjoint halves by means of a separating hyperplane. This split is empirically derived on a large set of training examples, for example from patients showing responsiveness or resistance to a therapeutic agent. Without loss of generality, one can assume a certain fixed set of values for all but one biomarker, which would automatically define a threshold value for this remaining biomarker where the decision would change from, for example, responsiveness or resistance to a therapeutic agent. Expression values above this dynamic threshold would then either indicate resistance (for a biomarker with a negative weight) or responsiveness (for a biomarker with a positive weight) to a therapeutic agent. The precise value of this threshold depends on the actual measured expression profile of all other biomarkers within the classifier, but the general indication of certain biomarkers remains fixed, i.e. high values or "relative over-expression" always contributes to either a responsiveness (genes with a positive weight) or resistance (genes with a negative weights). Therefore, in the context of the overall gene expression classifier, relative expression can indicate if either up- or down-regulation of a certain biomarker is indicative of responsiveness or resistance to a therapeutic agent.

[0089] There are a number of suitable methods for measuring expression profiles of test samples depending on the type of biomarker to be assayed. Measuring mRNA in a biological sample may be used as a surrogate for detection of the level of the corresponding protein in the biological sample. Thus, any of the biomarkers or biomarker panels described herein can also be detected by detecting the appropriate RNA. Methods of gene expression profiling include, but are not limited to, microarray, RT-PCT, qPCR, NGS, northern blots, SAGE, mass spectrometry.

[0090] mRNA expression levels are measured by reverse transcription quantitative polymerase chain reaction (RT-PCR followed with qPCR). RT-PCR is used to create a cDNA from the mRNA. The cDNA may be used in a qPCR assay to produce fluorescence as the DNA amplification process progresses. By comparison to a standard curve, qPCR can produce an absolute measurement such as number of copies of mRNA per cell. Northern blots, microarrays, Invader assays, and RT-PCR combined with capillary electrophoresis have all been used to measure expression levels of mRNA in a sample. See Gene Expression Profiling: Methods and Protocols, Richard A. Shimkets, editor, Humana Press, 2004.

[0091] miRNA molecules are small RNAs that are non-coding but may regulate gene expression. Any of the methods suited to the measurement of mRNA expression levels can also be used for the corresponding miRNA. Recently many laboratories have investigated the use of miRNAs as biomarkers for disease. Many diseases involve widespread transcriptional regulation, and it is not surprising that miRNAs might find a role as biomarkers. The connection between miRNA concentrations and disease is often even less clear than the connections between protein levels and disease, yet the value of miRNA biomarkers might be substantial. Of course, as with any RNA expressed differentially during disease, the problems facing the development of an in vitro diagnostic product will include the requirement that the miRNAs survive in the diseased cell and are easily extracted for analysis, or that the miRNAs are released into blood or other matrices where they must survive long enough to be measured. Protein biomarkers have similar requirements, although many potential protein biomarkers are secreted intentionally at the site of pathology and function, during disease, in a paracrine fashion. Many potential protein biomarkers are designed to function outside the cells within which those proteins are synthesized.

[0092] Gene expression may also be evaluated using mass spectrometry methods. A variety of configurations of mass spectrometers can be used to detect biomarker values. Several types of mass spectrometers are available or can be produced with various configurations. In general, a mass spectrometer has the following major components: a sample inlet, an ion source, a mass analyzer, a detector, a vacuum system, and instrument-control system, and a data system. Difference in the sample inlet, ion source, and mass analyzer generally define the type of instrument and its capabilities. For example, an inlet can be a capillary-column liquid chromatography source or can be a direct probe or stage such as used in matrix-assisted laser desorption. Common ion sources are, for example, electrospray, including nanospray and microspray or matrix-assisted laser desorption. Common mass analyzers include a quadrupole mass filter, ion trap mass analyzer and time-of-flight mass analyzer. Additional mass spectrometry methods are well known in the art (see Burlingame et al, Anal. Chem. 70:647 R- 716R (1998); Kinter and Sherman, New York (2000)).

[0093] Protein biomarkers and biomarker values can be detected and measured by any of the following: electrospray ionization mass spectrometry (ESI-MS), ESI-MS/MS, ESI- MS/(MS)n, matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF-MS), surface-enhanced laser desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS), desorption/ionization on silicon (DIOS), secondary ion mass spectrometry (SIMS), quadrupole time-of-flight (Q-TOF), tandem time-of-flight (TOF/TOF) technology, called ultraflex III TOF/TOF, atmospheric pressure chemical ionization mass spectrometry (APCI-MS), APCI-MS/MS, APCI-(MS).sup.N, atmospheric pressure photoionization mass spectrometry (APPI-MS), APPI-MS/MS, and APPI-(MS).sup.N, quadrupole mass spectrometry, Fourier transform mass spectrometry (FTMS), quantitative mass spectrometry, and ion trap mass spectrometry.

[0094] Sample preparation strategies are used to label and enrich samples before mass spectroscopic characterization of protein biomarkers and determination biomarker values. Labeling methods include but are not limited to isobaric tag for relative and absolute quantitation (iTRAQ) and stable isotope labeling with amino acids in cell culture (SILAC). Capture reagents used to selectively enrich samples for candidate biomarker proteins prior to mass spectroscopic analysis include but are not limited to aptamers, antibodies, nucleic acid probes, chimeras, small molecules, an F(ab')₂ fragment, a single chain antibody fragment, an Fv fragment, a single chain Fv fragment, a nucleic acid, a lectin, a ligand-binding receptor, affybodies, nanobodies, ankyrins, domain antibodies, alternative antibody scaffolds (e.g. diabodies etc) imprinted polymers, avimers, peptidomimetics, peptoids, peptide nucleic acids, threose nucleic acid, a hormone receptor, a cytokine receptor, and synthetic receptors, and modifications and fragments of these.

[0095] The foregoing assays enable the detection of biomarker values that are useful in methods for predicting responsiveness of a cancer therapeutic agent, where the methods comprise detecting, in a biological sample from an individual, at least N biomarker values that each correspond to a biomarker selected from the group consisting of the biomarkers provided in Tables 1A, IB, 2A, 2B or Groups I and II, wherein a classification, as described in detail below, using the biomarker values indicates whether the individual will be responsive to a therapeutic agent. While certain of the described predictive biomarkers are useful alone for predicting responsiveness to a therapeutic agent, methods are also described herein for the grouping of multiple subsets of the biomarkers that are each useful as a panel of two or more biomarkers. Thus, various embodiments of the instant application provide combinations comprising N biomarkers, wherein N is at least three biomarkers. It will be appreciated that N can be selected to be any number from any of the above-described ranges, as well as similar, but higher order, ranges. In accordance with any of the methods described herein, biomarker values can be detected and classified individually or they can be detected and classified collectively, as for example in a multiplex assay format,

b) Microarray methods

[0096] In one embodiment, the present invention makes use of "oligonucleotide arrays"

(also called herein "microarrays"). Microarrays can be employed for analyzing the expression of biomarkers in a cell, and especially for measuring the expression of biomarkers of cancer tissues.

[0097] In one embodiment, biomarker arrays are produced by hybridizing detectably labeled polynucleotides representing the mRNA transcripts present in a cell (e.g., fluorescently- labeled cDNA synthesized from total cell mRNA or labeled cRNA) to a microarray. A microarray is a surface with an ordered array of binding (e.g., hybridization) sites for products of many of the genes in the genome of a cell or organism, preferably most or almost all of the genes. Microarrays can be made in a number of ways known in the art. However produced, microarrays share certain characteristics. The arrays are reproducible, allowing multiple copies of a given array to be produced and easily compared with each other. Preferably the microarrays are small, usually smaller than 5 cm², and they are made from materials that are stable under binding (e.g., nucleic acid hybridization) conditions. A given binding site or unique set of binding sites in the microarray will specifically bind the product of a single gene in the cell. In a specific embodiment, positionally addressable arrays containing affixed nucleic acids of known sequence at each location are used.

[0098] It will be appreciated that when cDNA complementary to the RNA of a cell is made and hybridized to a microarray under suitable hybridization conditions, the level of hybridization to the site in the array corresponding to any particular gene will reflect the prevalence in the cell of mRNA transcribed from that gene/biomarker. For example, when detectably labeled (e.g., with a fluorophore) cDNA or cR A complementary to the total cellular mR A is hybridized to a microarray, the site on the array corresponding to a gene (i.e., capable of specifically binding the product of the gene) that is not transcribed in the cell will have little or no signal (e.g., fluorescent signal), and a gene for which the encoded mRNA is prevalent will have a relatively strong signal. Nucleic acid hybridization and wash conditions are chosen so that the probe "specifically binds" or "specifically hybridizes' to a specific array site, i.e., the probe hybridizes, duplexes or binds to a sequence array site with a complementary nucleic acid sequence but does not hybridize to a site with a non-complementary nucleic acid sequence. As used herein, one polynucleotide sequence is considered complementary to another when, if the shorter of the polynucleotides is less than or equal to 25 bases, there are no mismatches using standard base-pairing rules or, if the shorter of the polynucleotides is longer than 25 bases, there is no more than a 5% mismatch. Preferably, the polynucleotides are perfectly complementary (no mismatches). It can be demonstrated that specific hybridization conditions result in specific hybridization by carrying out a hybridization assay including negative controls using routine experimentation.

[0099] Optimal hybridization conditions will depend on the length (e.g., oligomer vs. polynucleotide greater than 200 bases) and type (e.g., RNA, DNA, PNA) of labeled probe and immobilized polynucleotide or oligonucleotide. General parameters for specific (i.e., stringent) hybridization conditions for nucleic acids are described in Sambrook et al., supra, and in Ausubel et al, "Current Protocols in Molecular Biology", Greene Publishing and Wiley- interscience, NY (1987), which is incorporated in its entirety for all purposes. When the cDNA microarrays are used, typical hybridization conditions are hybridization in 5xSSC plus 0.2% SDS at 65C for 4 hours followed by washes at 25°C in low stringency wash buffer (lxSSC plus 0.2% SDS) followed by 10 minutes at 25°C in high stringency wash buffer (0.1SSC plus 0.2% SDS) (see Shena et al, Proc. Natl. Acad. Sci. USA, Vol. 93, p. 10614 (1996)). Useful hybridization conditions are also provided in, e.g., Tijessen, Hybridization With Nucleic Acid Probes", Elsevier Science Publishers B.V. (1993) and Kricka, "Nonisotopic DNA Probe Techniques", Academic Press, San Diego, Calif. (1992).

c) Immunoassay methods

[0100] Immunoassay methods are based on the reaction of an antibody to its corresponding target or analyte and can detect the analyte in a sample depending on the specific assay format. To improve specificity and sensitivity of an assay method based on immunoreactivity, monoclonal antibodies are often used because of their specific epitope recognition. Polyclonal antibodies have also been successfully used in various immunoassays because of their increased affinity for the target as compared to monoclonal antibodies Immunoassays have been designed for use with a wide range of biological sample matrices Immunoassay formats have been designed to provide qualitative, semi-quantitative, and quantitative results.

[0101] Quantitative results may be generated through the use of a standard curve created with known concentrations of the specific analyte to be detected. The response or signal from an unknown sample is plotted onto the standard curve, and a quantity or value corresponding to the target in the unknown sample is established.

[0102] Numerous immunoassay formats have been designed. ELISA or EIA can be quantitative for the detection of an analyte/biomarker. This method relies on attachment of a label to either the analyte or the antibody and the label component includes, either directly or indirectly, an enzyme. ELISA tests may be formatted for direct, indirect, competitive, or sandwich detection of the analyte. Other methods rely on labels such as, for example, radioisotopes (I¹²⁵) or fluorescence. Additional techniques include, for example, agglutination, nephelometry, turbidimetry, Western blot, immunoprecipitation, immunocytochemistry, immunohistochemistry, flow cytometry, Luminex assay, and others (see ImmunoAssay: A Practical Guide, edited by Brian Law, published by Taylor & Francis, Ltd., 2005 edition).

[0103] Exemplary assay formats include enzyme-linked immunosorbent assay (ELISA), radioimmunoassay, fluorescent, chemiluminescence, and fluorescence resonance energy transfer (FRET) or time resolved-FRET (TR-FRET) immunoassays. Examples of procedures for detecting biomarkers include biomarker immunoprecipitation followed by quantitative methods that allow size and peptide level discrimination, such as gel electrophoresis, capillary electrophoresis, planar electrochromatography, and the like.

[0104] Methods of detecting and/or quantifying a detectable label or signal generating material depend on the nature of the label. The products of reactions catalyzed by appropriate enzymes (where the detectable label is an enzyme; see above) can be, without limitation, fluorescent, luminescent, or radioactive or they may absorb visible or ultraviolet light. Examples of detectors suitable for detecting such detectable labels include, without limitation, x- ray film, radioactivity counters, scintillation counters, spectrophotometers, colorimeters, fluorometers, luminometers, and densitometers.

[0105] Any of the methods for detection can be performed in any format that allows for any suitable preparation, processing, and analysis of the reactions. This can be, for example, in multi-well assay plates (e.g., 96 wells or 384 wells) or using any suitable array or microarray. Stock solutions for various agents can be made manually or robotically, and all subsequent pipetting, diluting, mixing, distribution, washing, incubating, sample readout, data collection and analysis can be done robotically using commercially available analysis software, robotics, and detection instrumentation capable of detecting a detectable label. Kits

[0106] Reagents, tools, and/or instructions for performing the methods described herein can be provided in a kit. For example, the kit can contain reagents, tools, and instructions for determining an appropriate therapy for a cancer patient. Such a kit can include reagents for collecting a tissue sample from a patient, such as by biopsy, and reagents for processing the tissue. The kit can also include one or more reagents for performing a gene or gene product expression analysis, such as reagents for performing RT-PCR, qPCR, northern blot, proteomic analysis, or immunohistochemistry to determine expression levels of gene or gene product markers in a sample of a patient. For example, primers for performing RT-PCR, probes for performing northern blot analyses, and/or antibodies for performing proteomic analysis such as Western blot, immunohistochemistry and ELISA analyses can be included in such kits. Appropriate buffers for the assays can also be included. Detection reagents required for any of these assays can also be included. The appropriate reagents and methods are described in further detail below.

[0107] The kits featured herein can also include an instruction sheet describing how to perform the assays for measuring gene or gene product expression. The instruction sheet can also include instructions for how to determine a reference cohort, including how to determine expression levels of gene or gene product markers in the reference cohort and how to assemble the expression data to establish a reference for comparison to a test patient. The instruction sheet can also include instructions for assaying gene or gene product expression in a test patient and for comparing the expression level with the expression in the reference cohort to subsequently determine the appropriate chemotherapy for the test patient. Methods for determining the appropriate chemotherapy are described above and can be described in detail in the instruction sheet.

[0108] Informational material included in the kits can be descriptive, instructional, marketing or other material that relates to the methods described herein and/or the use of the reagents for the methods described herein. For example, the informational material of the kit can contain contact information, e.g., a physical address, email address, website, or telephone number, where a user of the kit can obtain substantive information about performing a gene expression analysis and interpreting the results, particularly as they apply to a human's likelihood of having a positive response to a specific therapeutic agent.

[0109] The kits featured herein can also contain software necessary to infer a patient's likelihood of having a positive response to a specific therapeutic agent from the gene product marker expression.

Therapeutic agents [0110] As described above, the methods described herein permit the classification of a patient as responsive or non-responsive to a therapeutic agent that targets angiogenic processes and signaling within tumors. Some current such therapeutics used to treat cancer include, but are not limited to, the following agent; VEGF pathway-targeted therapeutic agent, including multi-targeted pathway inhibitors (VEGF/PDGF/FGF/EGFT/FLT-3/c-KIT), Angiopoietin-TIE2 pathway inhibitors, endogenous angiogenic inhibitors, immunomodulatory Agents. VEGF specific inhibitors include, but are not limited to, Bevacizumab (Avastin), Afibercept (VEGF Trap), IMC- 112 IB (Ramucirumab). Multi-targeted pathway inhibitors include, but are not limited to, Imatinib (Gleevec), Sorafenib (Nexavar), Gefitinib (Iressa), Sunitinib (Sutent), Erlotinib, Tivozinib, Cediranib (Recentin), Pazopanib (Votrient), BIBF 1120 (Vargatef), Dovitinib, Semaxanib (Sugen), Axitinib (AGO 13736), Vandetanib (Zactima), Nilotinib (Tasigna), Dasatinib (Sprycel), Vatalanib, Motesanib, ABT-869, TKI-258. Angiopoietin-TIE2 pathway inhibitors include, but are not limited to, AMG-386, PF-4856884 CVX-060, CEP- 11981, CE-245677, MEDI-3617, CVX-241, Trastuzumab (Herceptin). Endogenous angiogenic inhibitors include, but are not limited to, Thombospondin, Endostatin, Tumstatin, Canstatin, Arrestin, Angiostatin, Vasostatin, Interferon alpha. Immunomodulatory Agents include, but are not limited to, Thalidomide and Lenalidomide.

[0111] This invention is further illustrated by the following examples, which are not to be construed in any way as imposing limitations upon the scope thereof. On the contrary, it is to be clearly understood that resort may be had to various other embodiments, modifications, and equivalents thereof which, after reading the description herein, may suggest themselves to those skilled in the art without departing from the spirit of the present invention and/or scope of the appended claims.

EXAMPLES

Example 1: Tissue processing, hierarchical clustering, subtype identification and classifier development

Tumor Material.

[0112] Exemplary expression signatures were identified from gene expression analysis of a cohort of macrodissected epithelial serous ovarian tumor FFPE tissue samples sourced from the NHS Lothian and University of Edinburgh.

[0113] The protocol for histological classification of epithelial ovarian cancer to define serous, endometrioid, clear cell and mucinous histologies has recently been updated. One of the consequences of this is that many tumors that would previously been classified as endometrioid are now being classified as serous. (McCluggage, W.G. "Morphological subtypes of ovarian carcinoma: a review with emphasis on new developments and pathogenesis," PATHOLOGY 2011 Aug;43(5):420-32). Serous samples that were used in this study were among a larger set of epithelial ovarian cancer samples of all histologies that were harvested between 1984 and 2006. Pathology to assign histology status was performed by the pathologists at each of the centers at the time of harvesting. During March and April 2012, 357 of these epithelial ovarian samples were reviewed for histology classification by two independent consultant ovarian cancer pathologists according to the revised protocol. This resulted in a reclassification of several of these samples as indicated Table 3.

Table 3: results of pathology review of 357 epithelial ovarian cancer samples.

(Original histological status is presented in rows, and updated histological status is presented in columns)

The original three serous subtypes identified below, and consequently a 25 gene signature described in the example below (Fig 1) were identified from 199 samples that were classified as serous according to the original pathologist reports. Bioinformatic analysis was similarly performed on the 277 samples classified as stage III and IV high grade serous ovarian cancer using the updated pathology classification method. This analysis identified the updated serous subgroups detailed in Fig 2 and consequently used to define a 45 gene signature.

Gene Expression Profiling from FFPE

[0114] Total RNA was extracted from macrodissected FFPE tissue using the High Pure RNA Paraffin Kit (Roche Diagnostics GmbH, Mannheim, Germany). RNA was converted into complementary deoxyribonucleic acid (cDNA), which was subsequently amplified and converted into single-stranded form using the SPIA® technology of the WT-Ovation™ FFPE RNA Amplification System V2 (NuGEN Technologies Inc., San Carlos, CA, USA). The amplified single-stranded cDNA was then fragemented and biotin labeled using the FL- Ovation™ cDNA Biotin Module V2 (NuGEN Technologies Inc.). The fragmented and labeled cDNA was then hybridized to the Almac Ovarian Cancer DSA™. Almac's Ovarian Cancer DSA™ research tool has been optimised for analysis of FFPE tissue samples, enabling the use of valuable archived tissue banks. The Almac Ovarian Cancer DSA™ research tool is an innovative microarray platform that represents the transcriptome in both normal and cancerous ovarian tissues. Consequently, the Ovarian Cancer DSA™ provides a comprehensive representation of the transcriptome within the ovarian disease and tissue setting, not available using generic microarray platforms. Arrays were scanned using the Affymentrix Genechip® Scanner 7G (Affymetrix Inc., Santa Clara, CA).

Data preparation

[0115] Quality Control (QC) of profiled samples was carried out using MAS5 preprocessing algorithm. Different technical aspects were addressed: average noise and background homogeneity, percentage of present call (array quality), signal quality, RNA quality and hybridization quality. Distributions and Median Absolute Deviation of corresponding parameters were analyzed and used to identify possible outliers.

[0116] Almac's Ovarian Cancer DSA™ contains probes that primarily target the area within 300 nucleotides from the 3' end. Therefore standard Affymetrix RNA quality measures were adapted - for housekeeping genes intensities of 3 ' end probesets with ratios of 3 ' end probeset intensity to the average background intensity were used in addition to usual 375' ratios. Hybridization controls were checked to ensure that their intensities and present calls conform to the requirements specified by Affymetrix.

[0117] Pre-processing of the raw data generated from expression profiling of the epithelial serous ovarian cancer training set was performed in Expression Console vl .l with Robust Multi- array Analysis (RMA).

Hierarchical Clustering and Functional Analysis

a. Hierarchical clustering analysis

[0118] Hierarchical clustering techniques were applied to microarray data from the epithelial serous ovarian tumors analysed using the Ovarian Cancer DSA™ (disease specific array) platform. Raw expression data was preprocessed using the standard Robust Multichip Algorithm (RMA) procedure. Non-biological systematic variance in the data set was identified and removed. Those probe sets whose expression levels varied significantly from tumor to tumor were identified. These probe sets formed the intrinsic list.

[0119] Two dimensional cluster analysis (tumor, probeset) was performed to establish tumor relationships based on the intrinsic list. Hierarchical agglomerative clustering was applied (Pearson correlation - Original analysis - or Euclidean distance - updated analysis - and Ward's linkage). Optimal partition number was selected using the GAP index (Tibshirani et al., 2002, J. R. Stat. Soc, 63:411-423). All probesets available in the cluster subgroups were mapped to genes names.

b. Functional analysis of gene clusters

[0120] To establish the functional significance of the probeset clusters, probesets were mapped to genes (Entrez gene ID) and an enrichment analysis was performed. Enrichment significance was calculated based on the hypergeometric function (False Discovery Rate applied (Benjamini and Hochberg, 1995, J. R. Stat. Soc. 57:289:300)). Over-representation of biological processes and pathways were analysed for each gene group generated by the hierarchical clustering for the epithelial serous ovarian cancer samples using Almac Diagnostics' proprietary Functional Enrichment Tool (FET). Antisense probesets were excluded from the analysis. Hypergeometric p-values were assessed for each enriched functional entity class. Functional entity classes with the highest p-values were selected as representative of the group and a general functional category representing these functional entities was assigned to the gene clusters based on significance of representation (i.e. p-value).

[0121] To generate an angiogenesis classifier using the original 199 epithelial serous ovarian tumors, genes in clusters enriched for angiogenesis, vasculature development and immune response general functional terms were grouped into a putative angiogenesis gene group and used for the signature generation. The sample clusters presenting high expression for the genes involved in angiogenesis, vasculature development and immune response general functional terms were selected for classification and labeled as 'angiogenesis'. Those not showing high expression for the genes involved in these functional terms were labeled as 'non-angiogenesis'.

[0122] To generate an angiogenesis classifier using the reclassified 265 epithelial serous ovarian tumors, genes in clusters enriched for angiogenesis and vasculature development general functional terms were grouped into a putative angiogenesis gene group and used for the signature generation. The sample clusters presenting high expression for the genes involved in angiogenesis and vasculature development general functional terms were selected for classification and labeled as 'angiogenesis'. Those not showing high expression for the genes involved in these functional terms were labeled as 'non-angiogenesis'. Classifier development at a sene level

[0123] To facilitate validation of the classifier across multiple array platforms, the angiogenesis classifier was generated at the gene level. The following steps outline the procedures that were taken for gene level signature development (each step performed over internal cross validation using 10 repeats of 5 -fold cross-validation):

Gene level signature development

^■ Pre-processing:

o PvMA background correction.

o Reference set of genes are those genes (sense probesets only) unique to the ovarian DSA platform.

o Gene level summarization was performed in two steps; first probes to probeset summarization was performed by calculating the median expression of the probes in a probeset; secondly, median expression of the (sense only) probesets mapping to each gene in the reference distribution is calculated, yielding a "gene level" expression matrix.

o Quantile normalization was performed on the full gene expression data matrix and a reference quantile derived from the training data was used to normalize the test samples within each round of cross validation.

^■ Feature selection: Filtering 75% of genes by variance, intensity and correlation to cDNA concentration, followed by either recursive feature elimination (RFE) or filter feature selection (FFS) based on CAT scores.

^■ Classification algorithms: Partial Least Squares (PLS), SDA (Shrinkage Discriminate Analysis) and DSDA (Diagonal SDA).

Model selection

[0124] The criteria used for model selection were AUC over internal cross-validation and feature elimination. Functional enrichment of the signatures over cross validation using FET based on the gene ontologies, interim validation sets which included two sets of technical replicates for which standard deviation in signature scores for repeated samples was evaluated over cross-validation and feature elimination and an assessment of the independence to clinical and technical factors over cross validation (factors listed in Table 4).

[0125] It should be noted, since the subgroup (i.e. class label) derivation was performed using microarray expression from the same sample cohort that was used for signature development, there was an expected positive bias in any performance estimates based on AUC. This highlights the importance of widening the criteria used for model selection, by including additional metrics such as functional enrichment and assessing the independence to clinical and technical factors.

Table 4 List of clinical and technical factors investigated

Calculating classifier scores for validation data sets

[0126] All datasets were pre-processed using RMA. For each validation set, the probesets that map to the classifier genes were determined, excluding anti-sense probesets (if applicable). Annotation for Affymetrix Plus 2.0 and U133A arrays are available from the Affymetrix website. The median intensity over all probesets mapping to each gene in the classifier was calculated, resulting in a gene intensity matrix. The classifier was then applied to this data matrix to produce a classifier score/prediction for each sample.

Univariate and Multivariate analysis

[0127] Univariate and multivariate analysis may be carried out in relation to the glioblastoma dataset to assess respectively the association between the angiogenesis subtype classifier and survival, and to determine if the association, if any, was independent to known clinical predictors. The p-values for univariate analysis were calculated using logistic regression in MATLAB. For the multivariate analysis we used a likelihood ratio test from logistic regression was used, where the p-values represent the drop in the log-likelihood when comparing the model with the clinical covariates and the predictions to a reduced model with clinical covariates only. The likelihood ratio test measures the importance of the gene predictor in modeling survival, and highlights its independence as a predictor relative to the clinical predictors. In both univariate and multivariate analysis, a p-value < 0.05 was used as the criterion for significance. Furthermore, samples with unknown clinical factors were excluded in this assessment.

Results

Identification of subgroups and seneration of signature from original and updated histolosv classification

Hierarchical Clustering Analysis [0128] Feature selection resulted in the selection of 1200 probesets from the original epithelial serous ovarian cancer data set (199 samples) and 1400 PS from the reclassified epithelial serous ovarian cancer data set (265 samples). The GAP analysis revealed three sample clusters and three probeset cluster groups within both sample sets (FIG 1, FIG 6).

Classification of tumors into 'angiogenesis' or 'non-angiogenesis' sample groups

[0129] The classification of samples as 'anigogenesis' or 'non-angiogenesis' was based upon the results of the functional analysis of the epithelial serous ovarian cancer dataset (FIG 1 FIG 6). The objective of this study was to characterize at a transcriptomic level a set of genes that would be capable of determining responsiveness or resistance of a pathogenic cell to anti- angiogenic agents and potentially identify patients who could benefit from anti-angiogenic therapy. With this in mind, those samples within the epithelial serous ovarian cancer datasets that best represented this biology were to be selected and compared to the remaining samples for classifier generation (see next section). It was decided that the samples from the sample angiogenesis cluster within the original epithelial serous ovarian cancer sample set (199 samples) were the most relevant samples for this selection as these samples demonstrated an up- regulation of genes involved in signaling related to angiogenic and immune response processes and pathways as defined by functional analysis (FIG 2A and 2B) It was decided that the samples from sample cluster three within the reclassified epithelial serous ovarian cancer sample set (265 samples) were the most relevant samples for this selection as these samples demonstrated an up-regulation of genes involved in signaling related to angiogenic processes and pathways as defined by functional analysis (FIG 2A and 2B)

[0130] An identical hierarchial clustering approach was applied to 105 breast cancer samples. The dominant biology in Breast cancer is ER status and therefore in order to identify the structure in the biology of the samples this cohort was divided into 2 populations for cluster analysis. We identified angiogenesis and vasculature development subtypes (Figure 12A and 12B) demonstrating the expositing of an angiogenesis subtype from breast cancer samples.

Development and validation of the ansiosenesis subtype classifier models

[0131] For ease of reference, the following steps are detailed in reference to expression signatures derived from Table 1A or Table IB. However, a similar proceduce can be applied to other putative clusters of angiogenesis sub-type related biomarkers such as those disclosed in SEQ ID NOs: 632-801 and SEQ ID NOs: 802-974). Following the identification of a class of tumors, that form the putative 'angiogenesis' subgroup, computational classification of these tumors versus all others in the tumor cohort with reference to the functional 'angiogenesis' (angiogenesis, vasculature development, immune response) gene list (Table 1A or Table IB) was performed to identify a refined gene classification model, which classifies the 'angiogenesis' subtype.

[0132] The classification pipeline was used to derive a model using the set of epithelial serous ovarian cancer samples. The classification pipeline has been developed in accordance with commonly accepted good practice (MAQC Consortium, Nat Biotechnol 2010). The process will, in parallel: 1) derive gene classification models from empirical data; and 2) assess the classification performance of the models, both under cross-validation. The performance and success of the classifier generation depends on a number of parameters that can be varied, for instance the choice of classification method or probe set filtering. Taking this into account, two feature sets were evaluated (i) the full feature list with 75% variance/intensity filtering (with forced inclusion of the angiogenesis gene list, Table 1 A) and (ii) the angiogenesis gene list only; and three classification algorithms were evaluated, namely PLS; SDA and DSDA. RFE was used throughout model development, which is an iterative procedure removing a fraction of the lowest-ranked features at each iteration; stopping when only a minimum number of features remain. The AUC was used to assess the classification performance, as this measure is independent of cut-off between groups and prevalence rates in the data. It is also one of the recognized measurements of choice for classification performance. As such, the best number of features for each model was chosen based on the average AUC under cross-validation.

[0133] From the analysis described above, the PLS FFS model was deemed to be the most suitable classifier model. Weights were calculated for each gene using PLS regression, resulting in the final gene classifier models (25 -gene classifier model for the original approach, and a 45- gene classifier for samples reclassified reflecting recent changes to standard histology protocols) that may be used for validation on external data sets from different array platforms. The gene signature development process was focused upon identification of the ontological processes and pathways relevant to angiogenesis to ensure biological relevance of any signature developed. As such, functional analysis was performed upon both signatures to qualify their relevance to angiogenesis and related processes. The significance processes in FIG 3 and FIG 8 are related to angiogenesis and vasculature development.

Example 2: In silico validation of the angiogenesis subtype and angiogenesis classifier models

[0134] The performance of both the 25-gene (original approach) and 45-gene

(reclassification approach) angiogenesis classifier models were validated by the Area Under the ROC (Receiver Operator Characteristic) Curve (AUC) within the original Almac epithelial serous ovarian cancer dataset and two independent datasets. The AUC is a statistic calculated on the observed disease scale and is a measure of the efficacy of prediction of a phenotype using a classifier model (Wray et. al., PLoS Genetics Vol 6, 1-9). An AUC of 0.5 is typical of a random classifier, and an AUC of 1.0 would represent perfect separation of classes. Therefore, in order to determine if the angiogenesis subtype classifier model is capable of predicting response to, and selecting patients for anti-angiogenic ovarian cancer therapeutic drug classes either as single agent or in combination with standard of care therapies, the hypothesis is that the AUCs following application within these datasets should be above 0.5 with the lowest confidence interval also above 0.5.

Application of classifier model to independent microarray clinical datasets

[0135] To assess the prognostic power of the 25-gene and 45-gene classifier models, they were applied to a dataset of 77 glioblastoma samples taken at the time of initial surgical resection from patients (>21 years of age) without prior therapy (Phillips et. al., 2006).

This analysis revealed that the 25-gene classifier model was independently associated with prognosis in glioblastoma. Importantly, in a multivariate Cox analysis, the angiogenesis signature was found to be prognostic of survival, independent from World Health Organisation (WHO) tumor grade and/or the presence of necrosis (p = 0.37). Both of these clinical factors are correlated with survival in glioblastoma. The angiogenesis-high group was associated with significantly worse survival compared to the angiogenesis-low group (Hazard Ratio = 1.7814, p = 0.0269). This indicates that the 25-gene classifier is an independent prognostic biomarker of survival in glioblastoma.

Application of the classifier models to an independent prostate cancer cell-line dataset

[0136] To assess the predictive power of the 25-gene and 45-gene classifier models, they were applied to a dataset of 16 prostate cell-lines following treatment with Dasatanib. The cell- lines were defined as being either a 'responder' or 'non-responder' based upon cell-proliferation assays. This analysis revealed that the 25-gene classifier model is associated with response to Dasatanib, with an AUC of 0.8364 (CI = 0.5255 - 1.0000), indicating that the 25-gene classifier is predictive of response to Dasatanib. The analysis revealed that the 45-gene classifier model is associated with respone to the same compound, with an AUC of 0.9455 (CI = 0.7949 - 1.0000) indicating that the 45-gene classifier is also predictive of response to Dasatanib.

Example 4: Identification of and in silico validation of an anti-angiogenic "non-responsive" subgroup of ovarian cancer

[0137] The expression of angiogenesis genes in probeset cluster 2 is down regulated in all samples in Cluster 2 of hierarchial clustering of 265 samples newly classified as serous (Fig 6 and Fig 10). These samples in sample cluster 2 have a better prognosis than the rest of the serous samples in samples from cluster 1 and 3 combined together as demonstrated in Fig 11. This indicated that this group is defined by down regulation of expression of the angiogenesis genes identified in Table 2B. Patients with downregulation of genes involved in angiogeneisis and therefore this subgroup is termed a "non-responsive" group." This phenotype has aslo been identified in ER+ and ER- breast cancer as can be see the middle sample group in Fig 12A and the second sample group in Fig 12B.

References

1. Friedman HS, Prados MD, Wen PY, et al. Bevacizumab alone and in combination with irinotecan in recurrent glioblastoma. J Clin Oncol;21 /4733-40 (2009).

2. Hurwitz H, Fehrenbacher L, Novotny W, et al. Bevacizumab plus irinotecan, fluorouracil, and leucovorin for metastatic colorectal cancer. N Engl J Med;350:2335-42 (2004).

3. Rini BI, Halabi S, Rosenberg JE, et al. Bevacizumab plus interferon alfa compared with interferon alfa monotherapy in patients with metastatic renal cell carcinoma: CALGB 90206. J Clin Oncol; 26:5422-8 (2008).

4. Sandler A, Gray R, Perry MC, et al. Paclitaxel-carboplatin alone or with bevacizumab for non-small-cell lung cancer. N Engl J Med; 355: 2542-50 (2006).

5. Wolmark N, Yothers G, O'Connell MJ, et al. A phase III trial comparing mFOLFOX6 to mFOLFOX6 plus bevacizumab in stage II or III carcinoma of the colon: results of NSABP protocol C-08. J Clin Oncol; 27.LBA4 (2009).

6. Yang JC, Haworth L, Sherry RM, et al., A randomized trial of bevacizumab, an anti-vascular endothelial growth factor antibody, for metastatic renal cancer, N Engl J Med 349 427-434 (2003).

7. Willett CG, Boucher Y, di Tomaso E, et al, Direct evidence that the VEGF-specific antibody bevacizumab has antivascular effects in human rectal cancer, Nat. Med. 10, 145-147 (2004).

8. Miller K, Wang M, Gralow J, et al, Paclitaxel plus bevacizumab versus paclitaxel alone for metastatic breast cancer, N EnglJ Med 357 2666-2676 (2007).

9. Miller KD, Chap LI , Holmes FA, et al, Randomized phase III trial of capecitabine compared with bevacizumab plus capecitabine in patients with previously treated metastatic breast cancer, J Clin Oncol 23 792-799 (2005).

10. O'Shaughnessy J, Miles D, Gray Rj, et al, A meta-analysis of overall survival data from three randomized trials of bevacizumab (BV) and first-line chemotherapy as treatment for patients with metastatic breast cancer (MBC), J Clin Oncol 28 (suppl) (abstr 1005) (2010).

11. Reck M, von Pawel J, Zatloukal P, et al, Phase III trial of cisplatin plus gemcitabine with either placebo or bevacizumab as first-line therapy for nonsquamous non-small-cell lung cancer: AVAil, J C/m Oncol 21, 1227-1234 (2009).

12. Escudier B, Bellmunt J, Negrier S et al, Phase III trial of bevacizumab plus interferon alfa- 2a in patients with metastatic renal cell carcinoma (AVOREN): final analysis of overall survival, J Clin Oncol 2%, 2144-2150 (2010)

13. Burger RA, Sill MW, Monk BJ, et. al. Phase II trial of bevacizumab in persistent or recurrent epithelial ovarian cancer or primary peritoneal cancer: a Gynecologic Oncology Group Study. J Clin Oncol; 20;25(33):5165-71 (2007).

Claims

CLAIMS We claim:

1. A method of predicting a subject's responsiveness to an anti-angiogenic therapeutic agent comprising:

measuring expression levels of one or more biomarkers in a test sample of diseased tissue obtained from the subject to determine a sample expression score, wherein the biomarkers are defined by an expression signature;

comparing the sample expression score to a threshold expression score; and

classifying the subject as responsive or non-responsive, based on whether the sample expression score is above or below the threshold expression score.

2. The method of claim 1 , wherein the diseased tissue is cancer tissue.

3. The method of claim 2, wherein the cancer is ovarian cancer, glioblastoma or breast cancer.

4. The method of any one of claims 1 to 3, wherein the sample expression value is determined by measuring an expression level for each biomarker and multiplying it by a corresponding weight, wherein the weight for each biomarker is determined by the expression signature.

5. The method of claim 4, wherein the expression signature is derived by a method comprising:

isolating total R A from a sample set of diseased tissue;

hybridizing the isolated total RNA to a microarray to obtain a sample expression data set;

selecting those probes on the microarray with a variability above a defined significance threshold to form a preliminary biomarker set;

generating clusters of biomarkers within the preliminary biomarker set having a similar expression profile using a clustering algorithm;

identifying the biological processes or biological pathways for each cluster of biomarkers;

selecting the cluster corresponding to the biological process or biological pathway of interest; and defining an expression signature by analysing the expression levels of the biomarkers in the selected cluster in a the sample set of diseased tissue using a supervised or unsupervised training algorithm.

6. The method of claim 5, wherein the microarray is a transcriptome array comprising probe sets that binds to RNA transcripts verified as expressed in a test sample set by isolating and sequencing RNA transcripts from the test sample set and cross-validating the isolated RNA transcript sequences to known RNA transcript sequences, wherein the test sample set comprises the same tissue as the patient test sample

7. The method of claim 6, wherein the probe set comprises probes that bind within 300 nucleotides of the 3 ' end of each RNA transcript.

8. The method of any one of claims 5 to 7, wherein RNA transcripts comprising coding and non-coding transcripts that include messenger RNAs (mRNA), alternatively spliced mRNAs, ribosomal RNAs (rRNA), transfer RNAs (tRNA), small nuclear RNAs (snRNA), microRNAs (miRNAs) and antisense RNAs.

9. The method of any one of claims 5 to 8, wherein the expression signature is defined using a PLS classifier, a SVM classifier, a SDA classifier, or a DSDA classifier.

10. The method of claim 9, wherein the expression signature is defined using a PLS classifier.

11. The method of any one of claims 1 to 10, wherein the expression signature comprises two or more genes from Table 1 A or Table IB.

12. The method of any one of claims 1 to 10, wherein the expression signature comprises two or more genes from Table 2 A or 2B.

13. The method of any one of claims 1 to 10, wherein the expression signature comprises two or more genes comprising the sequences of SEQ ID NOs: 632-801 (Group I) or SEQ ID NOs: 802-974 (Group II).

14. The method of any one of claims 1 to 10, wherein the expression signature comprises ALPK2, BGN, COL8A1, FAP, FNl, GJB2, INHBA, ITGA5, LOXLl, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3, VCAN.

15. The method of any one of claims 1 to 10, wherein the expression signatures comprises the genes listed in Table 2A.

16. The method of any one of claims 1 to 10, wherein the expression signature comprises the genes listed in Table 2B.

17. The method of any one of claims 1 to 10, wherein the expression signature comprises the genes comprising the sequences of Group I or Group II.

18. The method of any one of claims 1 to 10„ wherein the expression signature comprises GJB2, INHBA, THBS2, SFRP2, and PLAU.

19. The method of any one of claims 1 to 18, wherein the anti-angiogenic therapeutic agent is a VEGF-pathway-targeted therapeutic agent, an angiopoietin-TIE2 pathway inhibitor, endogenous angiogenic inhibitors, and immunomodulatory agents.

20. The method of claim 19, wherein the VEGF pathway-targeted therapeutic agents include Bevacizumab (Avastin), Afibercept (VEGF Trap), IMC-1121B (Ramucirumab), Imatinib (Gleevec), Sorafenib (Nexavar), Gefitinib (Iressa), Sunitinib (Sutent), Erlotinib, Tivozinib, Cediranib (Recentin), Pazopanib (Votrient), BIBF 1120 (Vargatef), Dovitinib, Semaxanib (Sugen), Axitinib (AGO 13736), Vandetanib (Zactima), Nilotinib (Tasigna), Dasatinib (Sprycel), Vatalanib, Motesanib, ABT-869, TKI-258 or a combination thereof.

21. The method of claim 19, wherein the angiopoietin-TIE2 pathway inhibitor includes AMG-386, PF-4856884 CVX-060, CEP-11981, CE-245677, MEDI-3617, CVX-241, Trastuzumab (Herceptin) or a combination thereof.

22. The method of claim 19, wherein the endogenous angiogenic inhibitors include Thombospondin, Endostatin, Tumstatin, Canstatin, Arrestin, Angiostatin, Vasostatin, Interferon alpha or a combination thereof.

23. The method of claim 19, wherein the immunomodulatory agents include thalidomide and lenalidomide or a combination thereof.

24. A method of diagnosing a subject as having a cancer or being susceptible to developing the cancer response to anti-angiogenic therapeutics comprising:

measuring expression levels of biomarkers in a test sample of a diseased tissue obtained from the subject to determine obtain a sample expression score, wherein the biomarkers are defined by an expression signature;

comparing the sample expression value to a threshold score; and

classifying the subject as responsive or non-responsive based on whether the expression score is above or below the threshold score.

25. The method of claim 24, wherein the cancer is ovarian cancer or a glioblastoma.

26. The method of any one of claims 24 to 25, wherein the reference expression value is determined by calculating an expression value for each biomarker and multiplying it by a corresponding weight, wherein the weight for each biomarker is determined by the expression signature.

27. The method of claim 24, wherein the expression signature is derived by a method comprising:

isolating total R A from a sample set of the diseased tissue;

hybridizing the isolated total RNA to a microarray to obtain a set of expression values; selecting those probes on the microarray with the variability above a defined significance threshold to form a preliminary biomarker set;

generating clusters of biomarkers within the preliminary biomarker set having a similar gene expression profile using a clustering algorithm;

identifying the biological process or biological pathway for each cluster of biomarkers; selecting the cluster corresponding to the biological processes or biological pathways of interest; and

deriving the expression signature by analysing the expression levels of the biomarkers in the selected cluster in a the sample set of cancer tissue using a supervised or unsupervised training algorithm, wherein the expression value defines a set of biomarkers, corresponding weights for each biomarker and the reference expression value.

28. The method of claim 27, wherein the microarray is a transcriptome array comprising a probe set that binds to RNA transcripts verified as expressed in a sample set of cancer tissue by isolating and sequencing RNA transcripts from the cancer tissue and cross-validating isolated RNA transcript sequences to known RNA transcript sequences.

29. The method of claim 28, wherein the probe set comprises probes that bind within 300 nucleotides of the 3 ' end of each RNA transcript.

30. The method of any one of claims 27 to 29, wherein RNA transcripts comprising coding and non-coding transcripts that include messenger RNAs (mRNA), alternatively spliced mRNAs, ribosomal RNAs (rRNA), transfer RNAs (tRNA), small nuclear RNAs (snRNA), microRNAs (miRNAs) and antisense RNA.

31. The method of any one of claims 25 to 30, wherein the expression signature is defined using a PLS classifier, a SVM classifier, a SDA classifier, or a DSDA classifier.

32. The method of claim 31, wherein the expression signature is defined using a PLS classifier.

33. The method of any one of claims 24 to 32, wherein the expression signature comprises two or more genes from Table 1 A or Table IB.

34. The method of any one of claims 24 to 32, wherein the expression signature comprises two or more genes from Table 2 A or Table 2B.

35. The method of claims 24 to 32, wherein the expression signature comprises two or more genes comprising the sequences of Group I or Group II.

36. The method of any one of claims 24 to 32, wherein the expression signature comprises ALPK2, BGN, COL8A1, FAP, FNl, GJB2, INHBA, ITGA5, LOXLl, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3, VCAN.

37. The method of any one of claims 24 to 32, wherein the expression signature comprises the genes listed in Table 2A.

38. The method of any one of claims 24 to 32, wherein the expression signature comprises the genes listed in Table 2B.

39. The method of any one of claims 24 to 32, wherein the expression signature comprises the genes comprising the sequences of Group I or Group II.

40. The method of any one of claims 24 to 32, wherein the expression signature comprises GJB2, INHBA, THBS2, SFRP2, and PLAU

41. The method of any one of claims 24 to 40, wherein the anti-angiogenic agent is a VEGF- pathway-targeted therapeutic agent, an angiopoietin-TIE2 pathway inhibitor, endogenous angiogenic inhibitors, and immunomodulatory agents.

42. The method of claim 41, wherein the VEGF pathway-targeted therapeutic agents include Bevacizumab (Avastin), Afibercept (VEGF Trap), IMC-1121B (Ramucirumab), Imatinib (Gleevec), Sorafenib (Nexavar), Gefitinib (Iressa), Sunitinib (Sutent), Erlotinib, Tivozinib, Cediranib (Recentin), Pazopanib (Votrient), BIBF 1120 (Vargatef), Dovitinib, Semaxanib (Sugen), Axitinib (AGO 13736), Vandetanib (Zactima), Nilotinib (Tasigna), Dasatinib (Sprycel), Vatalanib, Motesanib, ABT-869, and TKI-258.

43. The method of claim 41, wherein the angiopoietin-TIE2 pathway inhibitor includes AMG-386, PF-4856884 CVX-060, CEP-11981, CE-245677, MEDI-3617, CVX-241, Trastuzumab (Herceptin).

44. The method of claim 41, wherein the endogenous angiogenic inhibitors include Thombospondin, Endostatin, Tumstatin, Canstatin, Arrestin, Angiostatin, Vasostatin, Interferon alpha.

45. The method of claim 41, wherein the immunomodulatory agents include thalidomide and lenalidomide.

46. A method of determing a prognosis of a subject with cancer comprising: measuring expression levels of one or more biomarkers in a test sample of diseased tissue obtained from the subject to determine a sample expression score, wherein the biomarkers are defined by an expression signature;

comparing the sample expression score to a threshold expression score; and

classifying the subject as responsive or non-responsive, based on whether the sample expression score is above or below the threshold expression score.

47. The method of claim 46, wherein the diseased tissue is cancer tissue.

48. The method of claim 47, wherein the cancer is ovarian cancer, glioblastoma, or breast cancer.

49. The method of any one claims 46 to 48, wherein the sample expression value is determined by measuring an expression level for each biomarker and multiplying it by a corresponding weight, wherein each weight for each biomarker is determined by the expression signature.

50. The method of claim 49, wherein the expression signature is derived by a method comprising:

isolating total R A from a sample set of diseased tissue;

hybridizing the isolated total RNA to a microarray to obtain a sample expression data set;

selecting those probes on the microarray with a variability above a defined significance threshold to form a preliminary biomarker set;

generating clusters of biomarkers within the preliminary biomarker set having a similar expression profile using a clustering algorithm;

identifying the biological processes or biological pathways for each cluster of biomarkers;

selecting the cluster corresponding to the biological process or biological pathway of interest; and

defining an expression signature by analysing the expression levels of the biomarkers in the selected cluster in a the sample set of diseased tissue using a supervised or unsupervised training algorithm.

51. The method of claim 50, wherein the microarray is a transcriptome array comprising probe sets that binds to RNA transcripts verified as expressed in a test sample set by isolating and sequencing RNA transcripts from the test sample set and cross-validating the isolated RNA transcript sequences to known RNA transcript sequences, wherein the test sample set comprises the same tissue as the patient test sample

52. The method of claim 51, wherein the probe set comprises probes that bind within 300 nucleotides of the 3 ' end of each RNA transcript.

53. The method of any one of claims 50 to 57, wherein RNA transcripts comprising coding and non-coding transcripts that include messenger RNAs (mRNA), alternatively spliced mRNAs, ribosomal RNAs (rRNA), transfer RNAs (tRNA), small nuclear RNAs (snRNA), microRNAs (miRNAs) and antisense RNAs.

54. The method of any one of claims 50 to 53, wherein the expression signature is defined using a PLS classifier, a SVM classifier, a SDA classifier, or a DSDA classifier.

55. The method of claim 46 to 54, wherein the expression signature is defined using a PLS classifier.

56. The method of any one of claims 46 to 55, wherein the expression signature comprises two or more genes from Table 1 A or Table IB.

57. The method of any one of claims 46 to 55, wherein the expression signature comprises two or more genes from Table 2 A or 2B.

58. The method of any one of claims 46 to 55, wherein the expression signature comprises two or more genes comprising the sequences of SEQ ID NOs: 632-801 (Group I) or SEQ ID NOs: 802-974 (Group II).

59. The method of any one of claims 46 to 55, wherein the expression signature comprises ALPK2, BGN, COL8A1, FAP, FN1, GJB2, INHBA, ITGA5, LOXL1, LUM, MIR1245, MMP2, NKD2, PLAU, RAB31, SFRP2, THBS2, TIMP3, VCAN.

60. The method of any one of claims 46 to 55, wherein the expression signatures comprises the genes listed in Table 2A.

61. The method of any one of claims 46 to 55, wherein the expression signature comprises the genes listed in Table 2B.

62. The method of any one of claims 46 to 55, wherein the expression signature comprises the genes comprising the sequences of Group I or Group II.

63. The method of any one of claims 46 to 55, wherein the expression signature comprises GJB2, INHBA, THBS2, SFRP2, and PLAU.

64. The method of any one of claims 46 to 63, wherein the anti-angiogenic therapeutic agent is a VEGF-pathway-targeted therapeutic agent, an angiopoietin-TIE2 pathway inhibitor, endogenous angiogenic inhibitors, and immunomodulatory agents.

65. The method of claim 64, wherein the VEGF pathway-targeted therapeutic agents include Bevacizumab (Avastin), Afibercept (VEGF Trap), IMC-1121B (Ramucirumab), Imatinib (Gleevec), Sorafenib (Nexavar), Gefitinib (Iressa), Sunitinib (Sutent), Erlotinib, Tivozinib, Cediranib (Recentin), Pazopanib (Votrient), BIBF 1120 (Vargatef), Dovitinib, Semaxanib (Sugen), Axitinib (AGO 13736), Vandetanib (Zactima), Nilotinib (Tasigna), Dasatinib (Sprycel), Vatalanib, Motesanib, ABT-869, TKI-258 or a combination thereof.

66. The method of claim 64, wherein the angiopoietin-TIE2 pathway inhibitor includes AMG-386, PF-4856884 CVX-060, CEP-11981, CE-245677, MEDI-3617, CVX-241, Trastuzumab (Herceptin) or a combination thereof.

67. The method of claim 64, wherein the endogenous angiogenic inhibitors include Thombospondin, Endostatin, Tumstatin, Canstatin, Arrestin, Angiostatin, Vasostatin, Interferon alpha or a combination thereof.

68. The method of claim 64, wherein the immunomodulatory agents include thalidomide and lenalidomide or a combination thereof.