WO2010127322A1 - Gene expression profile algorithm and test for likelihood of recurrence of colorectal cancer and response to chemotherapy - Google Patents

Abstract

Algorithm-based molecular assays that involve measurement of expression levels of prognostic and/or predictive genes, or co-expressed genes thereof, from a biological sample obtained from a cancer patient, and analysis of the measured expression levels to provide information concerning the likelihood of recurrence of colorectal cancer and/or the likelihood of a beneficial response to chemotherapy for the patient are provided herein. Methods of analysis of gene expression values of prognostic and/or predictive genes, as well as methods of identifying gene expression-tumor region ratios, tumor-associated stromal surface area, and gene cliques, i.e. genes that co-express with a validated biomarker and thus may be substituted for that biomarker in an assay, are also provided.

Description

GENE EXPRESSION PROFILE ALGORITHM AND TEST FOR

LIKELIHOOD OF RECURRENCE OF COLORECTAL CANCER AND

RESPONSE TO CHEMOTHERAPY

CROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims priority benefit of United States Provisional Application

Serial No. 61/174,890 filed on May 1, 2009 and United States Provisional Application Serial No.61/239,420 filed September 2, 2009, each of which is incorporated herein by reference in its entirety.

TECHNICAL FIELD

[0002] The present disclosure relates to molecular diagnostic assays that provide information concerning prognosis and prediction of response to chemotherapy in colorectal cancer patients. The present disclosure also provides methods of identifying genes that co- express with one or more biomarker genes.

INTRODUCTION

[0003] Colorectal cancer is the third most common malignant neoplasm worldwide, and the second leading cause of cancer-related mortality in the United States and the European Union. It is estimated that there will be approximately 150,000 new cases diagnosed each year in the United States, with about 65% of these being diagnosed as stage II/III colorectal cancer, as discussed below.

[0004] Clinical diagnosis of colorectal cancer generally involves evaluating the progression status of the cancer using standard classification criteria. Two classification systems have been widely used in colorectal cancer, the modified Duke's (or Astler-Coller) staging systems and more recently TNM staging as developed by the American Joint Committee on Cancer. Estimates of recurrence risk and treatment decisions in colorectal cancer are currently based primarily on tumor stage.

[0005] A series of trials carried out during the 1980's demonstrated that postoperative adjuvant therapy with fluorouracil ("5 -FU") and levamisole or leucovorin ("LV") led to a significant survival benefit for colon cancer patients. However, the benefits of adjuvant therapy are not enjoyed equally by all patients. For example, adjuvant 5-FU/LV chemotherapy has been shown to benefit a relatively small (-3%) but statistically significant subset of patients with stage II colon cancer, while the addition of oxaliplatin significantly improved overall DFS with no survival benefit seen in with stage II disease. (See, R. Gray et al., Lancet 370:2020-29 (2007), T. Andre, et al., N Engl J Med (2004), J. Kuebler, et al, J Clin Oncol (2007).) Moreover, significant neurotoxicity and GI toxicity is common and toxic deaths (0.5% in published studies) are well documented in other randomized trials. [0006] These results underline the importance of identifying prognostic and predictive tests which better define for individual patients their likelihood of recurrence and/or magnitude of benefit that they can expect from adjuvant chemotherapy. Under current guidelines, many patients who would be cured by surgery are unnecessarily given adjuvant therapy, while other patients who would benefit from such therapy do not receive it.

SUMMARY

[0007] Algorithm-based molecular assays that involve measurement of expression levels of prognostic and/or predictive genes, or co-expressed genes thereof, from a biological sample obtained from a cancer patient, and analysis of the measured expression levels to provide information concerning the likelihood of recurrence of colorectal cancer (Recurrence Score or RS) and/or the likelihood of a beneficial response to chemotherapy (Treatment Score or TS) for the patient are provided herein. Methods of analysis of gene expression values of prognostic and/or predictive genes, as well as methods of identifying gene cliques, i.e. genes that co-express with a validated biomarker and exhibit correlation of expression with the validated biomarker, and thus may be substituted for that biomarker in an assay, are also provided. One skilled in the art would recognize that such substitutions may impact the algorithm, for example the risk profile and weighting of the gene groups may need to be adjusted.

[0008] In exemplary embodiments, expression levels of a gene from gene subsets comprising a stromal group and a cell cycle group may be used to calculate a Recurrence Score (RS). The stromal group includes at least one of the following: BGN, FAP, INHBA, or a gene that that co-expresses with BGN, FAP, or INHBA. The cell cycle group includes at least one of the following: MYB L2, Ki-67, cMYC, MAD2L1, or a gene that co-expresses with MYB L2, Ki-67, cMYC, or MAD2L1. In other exemplary embodiments, the stromal gene is BGN and the cell cycle gene is Ki-67.

[0009] In exemplary embodiments, gene expression levels of one or more genes from additional gene subsets may be measured and used to calculate the RS, including a cell signaling group, and angiogenesis group, and/or an apoptosis group. The cell signaling group includes GADD45B and genes that co-express with GADD45B. The apoptosis group includes BIK and genes that co-express with BIK. The angiogenesis group includes EFNB2 and genes that co-express with EFNB2. The calculation may be performed on a computer programmed to execute the RS algorithm.

[0010] In exemplary embodiments, the method can further include measuring expression levels of predictive genes in a tumor sample obtained from the patient; and calculating a Treatment Score (TS) for the patient using measured gene expression levels, wherein the TS is calculated by assigning the measured expression levels to gene subsets of a TS algorithm, wherein the gene subsets comprise at least one gene each from an MSI group, an apoptosis group, and a stromal group. Calculation of the TS may be performed on a computer programmed to execute the TS algorithm. In exemplary embodiments, a benefit score for the patient based on the RS and the TS may be calculated. In exemplary embodiments, the MSI group can include AXIN2 and genes that co-express with AXIN2. In exemplary embodiments, the apoptosis group can include BIK and genes that co-express with BIK. In exemplary embodiments, the stromal group can include EFNB2 and genes that co- express with EFNB2. In exemplary embodiments, the gene subsets can further include a transcription factor group, where, e.g., the transcription factor group comprises RUNXl and genes that co-express with RUNXl. In exemplary embodiments, the gene subsets can further include a cell cycle group, where, e.g., the cell cycle group includes MAD2L1 and HSPEl, and genes that co-express with MAD2L1 and HSPEl. In exemplary embodiments, the at least one gene from the gene subsets may be replaced by a substitute gene from the group consisting of RANB P2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2, and genes that co-express with RANB P2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2.

[0011] In exemplary embodiments, the expression level for each gene subset may be weighted according to a contribution of the gene subset to risk of recurrence and/or response to chemotherapy.

[0012] The present disclosure provides methods to analyze gene expression taking into account variability of expression of certain gene subsets within particular regions of the tumor. In exemplary embodiments, this method may be incorporated into a RS algorithm. For example, the gene expression levels for the stromal group may be calculated as a ratio of stromal gene expression values per stroma unit area of a colorectal tumor. Similarly, gene expression levels for the cell cycle group may be calculated as a ratio of cell cycle expression values per epithelial unit area of the colorectal tumor. [0013] The present disclosure provides methods to estimate likelihood of colon cancer recurrence based on analysis of measurements of the surface area of the tumor- associated stroma in a colon tumor sample obtained from a patient. In exemplary embodiments, this method may be incorporated into a RS algorithm. [0014] The present disclosure provides methods to use a threshold value for expression values used in an algorithm-based gene expression analysis, which methods involve measuring an expression level of a gene in a tissue section obtained from a patient; and comparing the measured expression level to a threshold value for said gene; wherein if the threshold value is less than the expression level of said gene, the expression value is used in an expression algorithm, and wherein if the expression level of said gene is greater than or equal to the threshold value, the expression level is used in an expression algorithm. [0015] In exemplary embodiments, the the threshold value is based on a C_t value. The threshold value can be, for example, one or more from those listed in Table 3. [0016] The present disclosure provides gene expression analysis methods to identify a gene that is co-expressed with a target gene which methods involve normalizing microarray gene expression data for cancer tumor samples based on array probes; calculating a correlation coefficient based on gene expression levels for every unique pair of array probes; determining significant probe pairs, wherein significant probe pairs are a target gene probe and an array probe with a correlation co-efficient greater than a significant threshold value; mapping the target gene to its corresponding target gene probe, selecting a candidate probe set, wherein each candidate probe is part of a significant probe pair; and identifying a gene associated with each candidate probe; wherein said gene associated with each candidate probe is a co-expressed gene.

[0017] The present disclosure also provides methods of assessing gene expression, the method comprising measuring a normalized expression level of a gene in a cancer tumor sample obtained from a patient calculating a ratio of normalized expression of the gene to a tissue unit area in the colorectal sample, wherein the tissue unit area is a tumor-associated stroma unit area or a tumor epithelial unit area; and calculating a recurrence score (RS) or a treatment score (TS) for the patient using the ratio. In related embodiments, the gene is a stromal group gene. In related embodiments, the tissue unit area is a tumor-associated stroma unit area. In further related embodiments, the gene is a cell cycle group gene. In related embodiments, the tissue unit area is a tumor epithelial unit area unit area. [0018] The present disclosure provides methods of determining a prognosis for a cancer patient, comprising measuring a stromal area of a tumor sample obtained from the cancer patient to obtain a Stromal Risk Score, wherein increased stromal area of the tumor sample is positively correlated with an increased risk of recurrence of cancer for said cancer patient, and generating a report based on the Stromal Risk Score. In related embodiments, the tumor sample is a colorectal cancer tumor.

BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a set of graphs providing hazard ratio estimates and 95% confidence intervals for gene expression from univariate Cox PH regression models of recurrence-free interval (RFI) in NSABP C-01/02 patients and CCF patients for the 65 genes that were significantly related to RFI in both studies.

[0020] Figure 2 is a series of graphs providing hazard ratio estimates and 95% confidence intervals for gene expression from univariate Cox PH regression models of RFI in

C-01/02/04/06 and CCF patients for 48 gene significantly related to RFI in both surgery only and surgery plus FU-based chemotherapy.

[0021] Figure 3a is a graph illustrating Kaplan-Meier estimates of recurrence-free interval Stage II patients treated with surgery only, by tertile of recurrence score.

[0022] Figure 3b is a graph illustrating Kaplan-Meier estimates of recurrence-free interval Stage III patients treated with surgery only, by tertile of recurrence score.

[0023] Figure 4a provides a graph and a table illustrating a risk profile and recurrence scores (RS) for recurrence in Stage II colon cancer patients.

[0024] Figure 4b provides a graph and a table illustrating a risk profile and recurrence scores (RS) for recurrence in Stage III colon cancer, surgery only patients.

[0025] Figure 5 is a graph providing a chemotherapy benefit plot for Stage II patients.

[0026] Figure 6 provides a collection of graphs illustrating thresholding analysis for

BGN, FAP and INHBA.

[0027] Figure 7 provides a collection of graphs illustrating thresholding analysis for cMYC, Ki-67 and MYB L2.

[0028] Figure 8 provides a collection of graphs illustrating thresholding analysis for

GADD45B.

[0029] Figure 9 provides a collection of graphs illustrating thresholding analysis for

EFNB2, RUNXl and BIK.

[0030] Figure 10 provides a collection of graphs illustrating thresholding analysis for

MAD2L1, HSPEl and AXIN2.

[0031] Figure 11 is a schematic illustrating seeding of gene cliques. [0032] Figure 12 is a Kaplan Meier curve demonstrating group risk from the

QUASAR Stage II colon cancer patients treated with surgery alone.

[0033] Figure 13 is a risk profile plot (by Kaplan Meier curve) for risk of recurrence at five years and recurrence scores.

[0034] Figure 14 is a graph showing stromal group score (SGS) and cell cycle group score (CCGS) in tumor- associated stroma and tumor luminal areas.

[0035] Figure 15 is a graph showing results of analysis of stromal group score in tumor-associated stroma in six patients.

[0036] Figure 16 is a graph showing results of analysis of variability of stromal group and cell cycle group scores, GADD45B, and RS between tumor sections taken from 11 patient blocks.

[0037] Figure 17 is a graph showing the range of performance for multi-gene recurrence score models across all colon cancer studies

[0038] Figure 18: Performance of two gene model including a Stromal group gene

(BGN) and Cell cycle group gene (Ki-67)

[0039] Figure 19: Performance of three gene model including a Stromal group gene

(BGN), a Cell cycle group gene (Ki-67) and an Apoptosis group gene (BIK)

[0040] Figure 20: Comparative performance of ten-gene prognostic model (RS2) vs. seven-gene prognostic model (RS) in surgery-alone patients from the QUASAR study

[0041] Figure 21 is a variability plot for natural logarithm of stroma area for 444 colon cancer patients.

[0042] Figure 22 is a Kaplan-Meier plot for stage II colon cancer patients stratified by stroma risk group.

[0043] Figure 23 is a Kaplan-Meier plot for stage III colon cancer patients stratified by stroma risk group.

[0044] Figure 24 provides Kaplan- Meier estimates for stage II colon cancer patients stratified by stroma risk group and recurrence score risk group.

[0045] Figure 25 provides Kaplan-Meier survival curves for stage III colon cancer patients stratified by stroma risk group and recurrence score risk group.

[0046] Figure 26 is a graph showing the effects of diluting RNA concentration on

(non-normalized) gene expression (C_t) measurements of Ki-67. DETAILED DESCRIPTION DEFINITIONS

[0047] Unless defined otherwise, technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Singleton et al., Dictionary of Microbiology and Molecular Biology 2nd ed., J. Wiley & Sons (New York, NY 1994), and March, Advanced Organic Chemistry Reactions, Mechanisms and Structure 4th ed., John Wiley & Sons (New York, NY 1992), provide one skilled in the art with a general guide to many of the terms used in the present application.

[0048] One skilled in the art will recognize many methods and materials similar or equivalent to those described herein, which could be used in the practice of the present invention. Indeed, the present invention is in no way limited to the methods and materials described herein. For purposes of the invention, the following terms are defined below. [0049] The terms "tumor" and "lesion" as used herein, refer to all neoplastic cell growth and proliferation, whether malignant or benign, and all pre-cancerous and cancerous cells and tissues.

[0050] The terms "cancer" and "cancerous" refer to or describe the physiological condition in mammals that is typically characterized by unregulated cell growth. Examples of cancer in the present disclosure include cancer of the gastrointestinal tract, such as invasive colorectal cancer or Dukes B (stage II) or Dukes C (stage III) colorectal cancer. [0051] The "pathology" of cancer includes all phenomena that compromise the well- being of the patient. This includes, without limitation, abnormal or uncontrollable cell growth, metastasis, interference with the normal functioning of neighboring cells, release of cytokines or other secretory products at abnormal levels, suppression or aggravation of inflammatory or immunological response, neoplasia, premalignancy, malignancy, invasion of surrounding or distant tissues or organs, such as lymph nodes, etc. [0052] As used herein, the terms "colon cancer" and "colorectal cancer" are used interchangeably and in the broadest sense and refer to (1) all stages and all forms of cancer arising from epithelial cells of the large intestine and/or rectum and/or (2) all stages and all forms of cancer affecting the lining of the large intestine and/or rectum. In the staging systems used for classification of colorectal cancer, the colon and rectum are treated as one organ.

[0053] According to the tumor, node, metastatis (TNM) staging system of the

American Joint Committee on Cancer (AJCC) (Greene et al. (eds.), AJCC Cancer Staging Manual. 6th Ed. New York, NY: Springer; 2002), the various stages of colorectal cancer are defined as follows:

[0054] Tumor: Tl: tumor invades submucosal T2: tumor invades muscularis propria;

T3: tumor invades through the muscularis propria into the subserose, or into the pericolic or perirectal tissues; T4: tumor directly invades other organs or structures, and/or perforates.

[0055] Node: NO: no regional lymph node metastasis; Nl: metastasis in 1 to 3 regional lymph nodes; N2: metastasis in 4 or more regional lymph nodes.

[0056] Metastasis:M0: mp distant metastasis; Ml : distant metastasis present.

[0057] Stage groupings: Stage I: Tl NO MO; T2 NO MO; Stage II: T3 NO MO; T4

NO MO; Stage III: any T, Nl-2; MO; Stage IV: any T, any N, Ml.

[0058] According to the Modified Duke Staging System, the various stages of colorectal cancer are defined as follows:

[0059] Stage A: the tumor penetrates into the mucosa of the bowel wall but not further. Stage B: tumor penetrates into and through the muscularis propria of the bowel wall;

Stage C: tumor penetrates into but not through muscularis propria of the bowel wall, there is pathologic evidence of colorectal cancer in the lymph nodes; or tumor penetrates into and through the muscularis propria of the bowel wall, there is pathologic evidence of cancer in the lymph nodes; Stage D: tumor has spread beyond the confines of the lymph nodes, into other organs, such as the liver, lung or bone.

[0060] Prognostic factors are those variables related to the natural history of colorectal cancer, which influence the recurrence rates and outcome of patients once they have developed colorectal cancer. Clinical parameters that have been associated with a worse prognosis include, for example, lymph node involvement, and high grade tumors. Prognostic factors are frequently used to categorize patients into subgroups with different baseline relapse risks.

[0061] The term "prognosis" is used herein to refer to the prediction of the likelihood that a cancer patient will have a cancer- attributable death or progression, including recurrence, metastatic spread, and drug resistance, of a neoplastic disease, such as colon cancer.

[0062] The term "prognostic gene" is used herein to refer to a gene, the expression of which is correlated, positively or negatively, with a likelihood of cancer recurrence in a cancer patient treated with the standard of care. A gene may be both a prognostic and predictive gene, depending on the correlation of the gene expression level with the corresponding endpoint. For example, using a Cox proportional hazards model, if a gene is only prognostic, its hazard ratio (HR) does not change when measured in patients treated with the standard of care or in patients treated with a new intervention.

[0063] The term "prediction" is used herein to refer to the likelihood that a cancer patient will have a particular clinical response to treatment, whether positive ("beneficial response") or negative, following surgical removal of the primary tumor. For example, treatment could include chemotherapy.

[0064] The predictive methods of the present invention can be used clinically to make treatment decisions by choosing the most appropriate treatment modalities for any particular patient. The predictive methods of the present disclosure are valuable tools in predicting if a patient is likely to respond favorably ("beneficial response") to a treatment regimen, such as chemotherapy, surgical intervention, or both. Prediction may include prognostic factors. [0065] The terms "predictive gene" and "response indicator gene" are used interchangeably herein to refer to a gene, the expression level of which is correlated, positively or negatively, with likelihood of beneficial response to treatment with chemotherapy. A gene may be both a prognostic and predictive gene, and vice versa, depending on the correlation of the gene expression level with the corresponding endpoint (e.g., likelihood of survival without recurrence, likelihood of beneficial response to chemotherapy). A predictive gene can be identified using a Cox proportional hazards model to study the interaction effect between gene expression levels from patients treated with treatment A compared to patients who did not receive treatment A (but may have received standard of care, e.g. treatment B). The hazard ratio (HR) for a predictive gene will change when measured in untreated/standard of care patients versus patients treated with treatment A.

[0066] As used herein, the term "expression level" as applied to a gene refers to the normalized level of a gene product, e.g. the normalized value determined for the RNA expression level of a gene or for the polypeptide expression level of a gene. [0067] The term "gene product" or "expression product" are used herein to refer to the RNA transcription products (transcripts) of the gene, including mRNA, and the polypeptide translation products of such RNA transcripts. A gene product can be, for example, an unspliced RNA, an mRNA, a splice variant mRNA, a microRNA, a fragmented RNA, a polypeptide, a post-translationally modified polypeptide, a splice variant polypeptide, etc. [0068] The term "RNA transcript" as used herein refers to the RNA transcription products of a gene, including, for example, mRNA, an unspliced RNA, a splice variant mRNA, a microRNA, and a fragmented RNA.

[0069] Unless indicated otherwise, each gene name used herein corresponds to the

Official Symbol assigned to the gene and provided by Entrez Gene (URL: www.ncbi.nlm.nih.gov/sites/entrez) as of the filing date of this application. [0070] The terms "correlated" and "associated" are used interchangeably herein to refer to a strength of association between two measurements (or measured entities). The disclosure provides genes and gene subsets, the expression levels of which are associated with a particular outcome measure, such as for example between the expression level of a gene and the likelihood of beneficial response to treatment with a drug or micros atellite instability (MSI) phenotype status. For example, the increased expression level of a gene may be positively correlated (positively associated) with an increased likelihood of good clinical outcome for the patient, such as an increased likelihood of long-term survival without recurrence of the cancer and/or beneficial response to a chemotherapy, and the like. Such a positive correlation may be demonstrated statistically in various ways, e.g. by a low hazard ratio. In another example, the increased expression level of a gene may be negatively correlated (negatively associated) with an increased likelihood of good clinical outcome for the patient. In that case, for example, the patient may have a decreased likelihood of long- term survival without recurrence of the cancer and/or beneficial response to a chemotherapy, and the like. Such a negative correlation indicates that the patient likely has a poor prognosis or will respond poorly to a chemotherapy, and this may be demonstrated statistically in various ways, e.g., a high hazard ratio. "Correlated" is also used herein to refer to a strength of association between the expression levels of two different genes, such that expression level of a first gene can be substituted with an expression level of a second gene in a given algorithm in view of their correlation of expression. Such "correlated expression" of two genes that are substitutable in an algorithm usually gene expression levels that are positively correlated with one another, e.g., if increased expression of a first gene is positively correlated with an outcome (e.g., increased likelihood of good clinical outcome), then the second gene that is co-expressed and exhibits correlated expression with the first gene is also positively correlated with the same outcome.

[0071] A "positive clinical outcome" and "beneficial response" can be assessed using any endpoint indicating a benefit to the patient, including, without limitation, (1) inhibition, to some extent, of tumor growth, including slowing down and complete growth arrest; (2) reduction in the number of tumor cells; (3) reduction in tumor size; (4) inhibition (i.e., reduction, slowing down or complete stopping) of tumor cell infiltration into adjacent peripheral organs and/or tissues; (5) inhibition of metastasis; (6) enhancement of anti-tumor immune response, possibly resulting in regression or rejection of the tumor; (7) relief, to some extent, of one or more symptoms associated with the tumor; (8) increase in the length of survival following treatment; and/or (9) decreased mortality at a given point of time following treatment. Positive clinical response may also be expressed in terms of various measures of clinical outcome. Positive clinical outcome can also be considered in the context of an individual's outcome relative to an outcome of a population of patients having a comparable clinical diagnosis, and can be assessed using various endpoints such as an increase in the duration of Recurrence-Free interval (RFI), an increase in the time of survival as compared to Overall Survival (OS) in a population, an increase in the time of Disease-Free

Survival (DFS), an increase in the duration of Distant Recurrence-Free Interval (DRFI), and the like. An increase in the likelihood of positive clinical response corresponds to a decrease in the likelihood of cancer recurrence.

[0072] The term "risk classification" means a level of risk (or likelihood) that a subject will experience a particular clinical outcome. A subject may be classified into a risk group or classified at a level of risk based on the methods of the present disclosure, e.g. high, medium, or low risk. A "risk group" is a group of subjects or individuals with a similar level of risk for a particular clinical outcome.

[0073] The term "long-term" survival is used herein to refer to survival for a particular time period, e.g., for at least 3 years, more preferably for at least 5 years.

[0074] The term "Recurrence-Free Interval (RFI)" is used herein to refer to the time

(in years) from randomization to first colon cancer recurrence or death due to recurrence of colorectal cancer.

[0075] The term "Overall Survival (OS)" is used herein to refer to the time (in years) from randomization to death from any cause.

[0076] The term "Disease-Free Survival (DFS)" is used herein to refer to the time (in years) from randomization to first colon cancer recurrence or death from any cause.

[0077] The term "Distant Recurrence-Free Interval (DRFI)" is used herein to refer to the time (in years) from surgery to the first anatomically distant cancer recurrence.

[0078] The calculation of the measures listed above in practice may vary from study to study depending on the definition of events to be either censored or not considered. [0079] The term "tumor-associated stroma unit area" (or "sua") is used herein to refer to a measurement of the tumor-associated stroma area surrounding a tumor. Stroma is the framework or matrix of an organ providing support to the epithelia which includes components such as blood vessels, connective tissues and lymphoid cells. In the colon, tumor-associated stroma is interposed between normal stroma, epithelia, smooth muscle and malignant epithelial cells.

[0080] The term "tumor epithelial unit area" (or "cua") is used herein to refer to a measurement of the epithelial area of a tumor which comprises cancerous (e.g., malignant) epithelial cells. In the colon, the tumor associated epithelia cells are glandular in form, genomically clonal and are referred to as the adenocarcinoma.

[0081] The term "stromal area" as used herein, refers to the surface area of colon tumor-associated stroma in a biological sample obtained from a patient sample. The stromal area may be measured by any suitable method, such as by micrometer, or standard or digital microscopic assessment of a Hematoxylin and Eosin (H&E) section. [0082] The term "Stromal Risk," as used herein, refers to an estimate of recurrence risk of a patient with colon cancer based on stromal area. The amount of stromal area in a colon cancer tumor obtained from a patient is associated with the risk of recurrence of colon cancer for that patient. The greater the amount of stromal area present, the greater the risk of colon cancer recurrence. This estimate may be, for example, provided in the form of a Stromal Risk Score or Group that reflects the likelihood that a colon cancer patient will have a recurrence, such as a numeric range, descriptive categories (low, intermediate, high), etc. [0083] The term "microarray" refers to an ordered arrangement of hybridizable array elements, e.g. oligonucleotide or polynucleotide probes, on a substrate. [0084] The term "polynucleotide," when used in singular or plural, generally refers to any polyribonucleotide or polydeoxribonucleotide, which may be unmodified RNA or DNA or modified RNA or DNA. Thus, for instance, polynucleotides as defined herein include, without limitation, single- and double- stranded DNA, DNA including single- and double- stranded regions, single- and double-stranded RNA, and RNA including single- and double- stranded regions, hybrid molecules comprising DNA and RNA that may be single- stranded or, more typically, double-stranded or include single- and double- s.tranded regions. In addition, the term "polynucleotide" as used herein refers to triple- stranded regions comprising RNA or DNA or both RNA and DNA. The strands in such regions may be from the same molecule or from different molecules. The regions may include all of one or more of the molecules, but more typically involve only a region of some of the molecules. One of the molecules of a triple-helical region often is an oligonucleotide. The term "polynucleotide" specifically includes cDNAs. The term includes DNAs (including cDNAs) and RNAs that contain one or more modified bases. Thus, DNAs or RNAs with backbones modified for stability or for other reasons, are "polynucleotides" as that term is intended herein. Moreover, DNAs or RNAs comprising unusual bases, such as inosine, or modified bases, such as tritiated bases, are included within the term "polynucleotides" as defined herein. In general, the term "polynucleotide" embraces all chemically, enzymatically and/or metabolically modified forms of unmodified polynucleotides, as well as the chemical forms of DNA and RNA characteristic of viruses and cells, including simple and complex cells. [0085] The term "oligonucleotide" refers to a relatively short polynucleotide, including, without limitation, single-stranded deoxyribonucleotides, single- or double- stranded ribonucleotides, RNArDNA hybrids and double- stranded DNAs. Oligonucleotides, such as single-stranded DNA probe oligonucleotides, are often synthesized by chemical methods, for example using automated oligonucleotide synthesizers that are commercially available. However, oligonucleotides can be made by a variety of other methods, including in vitro recombinant DNA-mediated techniques and by expression of DNAs in cells and organisms.

[0086] As used herein, the term "expression level" as applied to a gene refers to the level of the expression product of a gene, e.g. the normalized value determined for the RNA expression product of a gene or for the polypeptide expression level of a gene. [0087] The term "C_t" as used herein refers to threshold cycle, the cycle number in quantitative polymerase chain reaction (qPCR) at which the fluorescence generated within a reaction well exceeds the defined threshold, i.e. the point during the reaction at which a sufficient number of amplicons have accumulated to meet the defined threshold. [0088] The terms "threshold" or "thresholding" refer to a procedure used to account for non-linear relationships between gene expression measurements and clinical response as well as to further reduce variation in reported patient scores. When thresholding is applied, all measurements below or above a threshold are set to that threshold value. Non-linear relationship between gene expression and outcome could be examined using smoothers or cubic splines to model gene expression in Cox PH regression on recurrence free interval or logistic regression on recurrence status. Variation in reported patient scores could be examined as a function of variability in gene expression at the limit of quantitation and/or detection for a particular gene. [0089] As used herein, the term "amplicon," refers to pieces of DNA that have been synthesized using amplification techniquest, such as polymerase chain reactions (PCR) and ligase chain reactions.

[0090] "Stringency" of hybridization reactions is readily determinable by one of ordinary skill in the art, and generally is an empirical calculation dependent upon probe length, washing temperature, and salt concentration. In general, longer probes require higher temperatures for proper annealing, while shorter probes need lower temperatures. Hybridization generally depends on the ability of denatured DNA to re-anneal when complementary strands are present in an environment below their melting temperature. The higher the degree of desired homology between the probe and hybridizable sequence, the higher the relative temperature which can be used. As a result, it follows that higher relative temperatures would tend to make the reaction conditions more stringent, while lower temperatures less so. For additional details and explanation of stringency of hybridization reactions, see Ausubel et al., Current Protocols in Molecular Biology, Wiley Interscience Publishers, (1995).

[0091] "Stringent conditions" or "high stringency conditions", as defined herein, typically: (1) employ low ionic strength and high temperature for washing, for example 0.015 M sodium chloride/0.0015 M sodium citrate/0.1% sodium dodecyl sulfate at 50⁰C; (2) employ during hybridization a denaturing agent, such as formamide, for example, 50% (v/v) formamide with 0.1% bovine serum albumin/0.1% Ficoll/0.1% polyvinylpyrrolidone/5 OmM sodium phosphate buffer at pH 6.5 with 750 mM sodium chloride, 75 mM sodium citrate at 42°C; or (3) employ 50% formamide, 5 x SSC (0.75 M NaCl, 0.075 M sodium citrate), 50 mM sodium phosphate (pH 6.8), 0.1% sodium pyrophosphate, 5 x Denhardt's solution, sonicated salmon sperm DNA (50 μg/ml), 0.1% SDS, and 10% dextran sulfate at 42°C, with washes at 42°C in 0.2 x SSC (sodium chloride/sodium citrate) and 50% formamide, followed by a high- stringency wash consisting of 0.1 x SSC containing EDTA at 55⁰C. [0092] "Moderately stringent conditions" may be identified as described by

Sambrook et al., Molecular Cloning: A Laboratory Manual, New York: Cold Spring Harbor Press, 1989, and include the use of washing solution and hybridization conditions (e.g., temperature, ionic strength and %SDS) less stringent that those described above. An example of moderately stringent conditions is overnight incubation at 37°C in a solution comprising: 20% formamide, 5 x SSC (150 mM NaCl, 15 mM trisodium citrate), 50 mM sodium phosphate (pH 7.6), 5 x Denhardt's solution, 10% dextran sulfate, and 20 mg/ml denatured sheared salmon sperm DNA, followed by washing the filters in 1 x SSC at about 37-50⁰C. The skilled artisan will recognize how to adjust the temperature, ionic strength, etc. as necessary to accommodate factors such as probe length and the like.

[0093] The terms "splicing" and "RNA splicing" are used interchangeably and refer to RNA processing that removes introns and joins exons to produce mature mRNA with continuous coding sequence that moves into the cytoplasm of an eukaryotic cell. [0094] As used herein, the term "exon" refers to any segment of an interrupted gene that is represented in the mature RNA product. As used herein, the term "intron" refers to any segment of DNA that is transcribed but removed from within the transcript by splicing together the exons on either side of it. "Intronic RNA" refers to mRNA derived from an intronic region of DNA. Operationally, exonic sequences occur in the mRNA sequence of a gene as defined by Ref. SEQ ID numbers. Operationally, intron sequences are the intervening sequences within the genomic DNA of a gene.

[0095] The term "co-expressed", as used herein, refers to a statistical correlation between the expression level of one gene and the expression level of another gene. Pairwise co-expression may be calculated by various methods known in the art, e.g., by calculating Pearson correlation coefficients or Spearman correlation coefficients. Co-expressed gene cliques may also be identified using a graph theory. An analysis of co-expression may be calculated using normalized expression data.

[0096] A "computer-based system" refers to a system of hardware, software, and data storage medium used to analyze information. The minimum hardware of a patient computer- based system comprises a central processing unit (CPU), and hardware for data input, data output (e.g., display), and data storage. An ordinarily skilled artisan can readily appreciate that any currently available computer-based systems and/or components thereof are suitable for use in connection with the methods of the present disclosure. The data storage medium may comprise any manufacture comprising a recording of the present information as described above, or a memory access device that can access such a manufacture. [0097] To "record" data, programming or other information on a computer readable medium refers to a process for storing information, using any such methods as known in the art. Any convenient data storage structure may be chosen, based on the means used to access the stored information. A variety of data processor programs and formats can be used for storage, e.g. word processing text file, database format, etc.

[0098] A "processor" or "computing means" references any hardware and/or software combination that will perform the functions required of it. For example, a suitable processor may be a programmable digital microprocessor such as available in the form of an electronic controller, mainframe, server or personal computer (desktop or portable). Where the processor is programmable, suitable programming can be communicated from a remote location to the processor, or previously saved in a computer program product (such as a portable or fixed computer readable storage medium, whether magnetic, optical or solid state device based). For example, a magnetic medium or optical disk may carry the programming, and can be read by a suitable reader communicating with each processor at its corresponding station.

[0099] As used herein, the term "surgery" applies to surgical methods undertaken for removal of cancerous tissue, including resection, laparotomy, colectomy (with or without lymphadenectomy), ablative therapy, endoscopic removal, excision, dissection, and tumor biopsy/removal. The tumor tissue or sections used for gene expression analysis may have been obtained from any of these methods.

[00100] As used herein, "graph theory" refers to a field of study in Computer Science and Mathematics in which situations are represented by a diagram containing a set of points and lines connecting some of those points. The diagram is referred to as a "graph", and the points and lines referred to as "vertices" and "edges" of the graph. In terms of gene co- expression analysis, a gene (or its equivalent identifier, e.g. an array probe) may be represented as a node or vertex in the graph. If the measures of similarity (e.g., correlation coefficient, mutual information, alternating conditional expectation) between two genes is higher than a significant threshold, the two genes are said to be co-expressed and an edge will be drawn in the graph. When co-expressed edges for all possible gene pairs for a given study have been drawn, all maximal cliques are computed. The resulting maximal clique is defined as a gene clique. A gene clique is a computed co-expressed gene group that meets predefined criteria.

[00101] As used herein, the terms "gene clique" and "clique" refer to a subgraph of a graph in which every vertex is connected by an edge to every other vertex of the subgraph. [00102] As used herein, a "maximal clique" is a clique in which no other vertex can be added and still be a clique.

[00103] Reference to "markers for prediction of response" with reference to 5- fluorouracil (5-FU), and like expressions, encompass within their meaning response to treatment comprising 5-FU as monotherapy, or in combination with other agents, or as prodrugs, or together with local therapies such as surgery and radiation, or as adjuvant or neoadjuvant chemotherapy, or as part of a multimodal approach to the treatment of neoplastic disease. [00104] As used herein, the terms "5-FU-based therapy", "5-FU based treatment", and

"5-FU therapy" are used interchangeably to refer to encompass administration of 5-FU or a prodrug thereof and further encompasses administration of 5-FU combination or 5-FU combination therapy.

[00105] "5-FU combination" or "5-FU combination therapy" refers to a combination of 5-FU and another agent. A number of agents have been combined with 5-FU to enhance the cytotoxic activity through biochemical modulation. Addition of exogenous folate in the form of 5-formyl-tetrahydrofolate (leucovorin) sustains inhibition of thymidylate synthase. Methotrexate, by inhibiting purine synthesis and increasing cellular pools of certain substrates for reactivity with 5-FU, enhances the activation of 5-FU. The combination of cisplatin and 5-FU increases the antitumor activity of 5-FU. Oxaliplatin is commonly used with 5-FU and leucovorin for treating colorectal cancer, and it may inhibit catabolism of 5- FU, perhaps by inhibiting dihydropyrimidine dehydrogenase (the enzyme that is responsible for the catabolism of 5-FU), and may also inhibit expression of thymidylate synthase. The combination of 5-FU and irinotecan, a topoisomerase- 1 inhibitor, is a treatment that combines 5-FU with an agent that has a different mechanism of action. Eniluracil, which is an inactivator of dihydropyrimidine dehydrogenase, leads to another strategy for improving the efficacy of 5-FU.

[00106] "5-FU prodrug" refers to drugs that, following administration to a patient, provide for activity of 5-FU. A number of 5-FU prodrugs have been developed. For example, capecitabine (N4-pentoxycarbonyl-5'-deoxy-5-fluorcytidine) is an orally administered agent that is approved by the FDA for certain treatments including colorectal cancer. Another fluoropyrimidine that acts as a prodrug for 5-FU is florafur.

ALGORITHM-BASED METHODS AND GENE SUBSETS

[00107] The present disclosure provides an algorithm-based molecular diagnostic assay for determining an expected clinical outcome (prognostic) and/or the likelihood that a patient with cancer will have a clinically beneficial response to chemotherapy (predictive). For example, the expression levels of the prognostic genes may be used to calculate a likelihood of colorectal cancer recurrence. The expression levels of the predictive genes, and in some cases the predictive and prognostic genes, may be used to calculate the likelihood that a patient with colorectal cancer will have a clinically beneficial response to chemotherapy. The cancer can be, for example, Stage II and/or Stage III colorectal cancer. The chemotherapy can be, for example, a 5-FU-based chemotherapy. [00108] The present disclosure provides methods to classify a tumor based on the likelihood of cancer recurrence for a patient. The likelihood of recurrence is calculated based on expression levels of prognostic genes from particular gene subsets, wherein gene subsets include at least one gene each from a stromal group and a cell cycle group. Prognostic gene subsets may also include at least one gene from a cell signaling group, an apoptosis group, and/or a transcription factor group.

[00109] The present disclosure provides methods of classifying a tumor according to the likelihood that a patient with cancer will have a beneficial response to chemotherapy based on expression levels of predictive genes. The likelihood of a beneficial response is calculated based on expression levels of predictive genes from particular gene subsets, wherein the gene subsets include at least one gene from each of a stromal group, an apoptosis group, and a MSI group. Predictive gene subsets can also include at least one gene from a transcription factor group and/or a cell cycle group.

[00110] The gene subset identified herein as the "stromal group" includes genes that are synthesized predominantly by stromal cells and are involved in stromal response and genes that co-express with stromal group genes. "Stromal cells" are defined herein as connective tissue cells that make up the support structure of biological tissues. Stromal cells include fibroblasts, immune cells, pericytes, endothelial cells, and inflammatory cells. "Stromal response" refers to a desmoplastic response of the host tissues at the site of a primary tumor or invasion. See, e.g., E. Rubin, J. Farber, Pathology, 985-986 (2^nd Ed. 1994). The stromal group includes, for example, BGN, FAP, INHBA, and genes that are co- expressed with BGN, FAP, or INHBA, wherein a gene is said to be co-expressed with a stromal gene when the expression level of the gene exhibits a Pearson correlation coefficient greater than or equal to 0.6. For example, the stromal group includes the genes and/or gene cliques shown in Tables 4, 5 and 6 (provided in specification just prior to claims). The combination of genes used from within the stromal group can vary with the method of analysis for which expression is to be evaluated. For example, the stromal group for classifying a tumor according to the likelihood of colorectal cancer recurrence includes BGN, FAP and INHBA. The gene subset herein identified as the "cell cycle group" includes genes that are involved with cell cycle functions and genes that co-express with cell cycle group genes. "Cell cycle functions" are defined herein as cell proliferation and cell cycle control, e.g. checkpoint/Gl to S phase transition. The cell cycle group thus includes genes that (1) are involved in biological pathways associated with cell cycle functions; and (2) co-express with Ki-67, cMYC, MYBL2, MAD2L1, or HSPEl, with a Pearson correlation coefficient greater than or equal to 0.4. Exemplary co-expressed genes and/or gene cliques for Ki-67, cMYC, MYB L2, MAD2L1, and HSPEl are provided in Tables 5 and 6. The combination of genes used from within the cell cycle group can vary with the method of analysis for which expression is to be evaluated. For example, the cell cycle group for classifying a tumor according to the likelihood of colorectal cancer recurrence includes Ki-67, cMYC, MYB L2, MAD2L1, and HSPEl. The cell cycle group for classifying a tumor according to likelihood that a patient will have a beneficial response to chemotherapy includes MAD2L1 and HSPEl. [00111] This specification discloses data demonstrating that genes associated with the stroma of a tumor are associated with an increased risk of recurrence, whereas cell cycle genes are correlated with a decreased risk of recurrence. In addition, the present disclosure provides prognostic and predictive methods that take into account the observation that expression levels for certain genes vary with respect to the regions of a tumor. [00112] Specifically, the present disclosure provides evidence that there are higher expression levels of (1) the stromal genes in the tumor-associated stroma; and (2) the cell cycle genes in the luminal part of the tumor. The ratios of expression levels to tumor region areas vary from patient to patient. This ratio of expression between tumor-associated stroma and the luminal part of the tumor can be exploited in the prognostic and predictive methods disclosed herein.

[00113] In exemplary embodiments, expression values of stromal genes may be calculated using stromal gene expression per stroma unit area, and expression values of cell cycle genes may be calculated using cell cycle gene expression per epithelial unit area. Thus, the area of the tumor- associated stroma and the area of the tumor-luminal regions may be taken into account by the prognostic and predictive algorithms in order to increase reproducibility and accuracy of RFI prediction and prediction of response to therapy, respectively. One skilled in the art would recognize that there are many conventional methods available to capture percent stroma and percent epithelia. For example, such ratios could be obtained by examining the H&E slide immediately adjacent to the tissue sections to be analyzed. This could be performed by either a pathologist (to get a gross measurement) or by digital image analysis (to obtain a more precise measurement).

[00114] In addition, the present disclosure provides evidence that measurement of the stroma area has prognostic value to colon cancer patients. Specifically, the stromal surface area of the tumor- associate stromal region of a tumor is positively correlated with increase risk of recurrence. This risk of recurrence may be reported in the form of a Stromal Risk score, or combined with risk information obtained from other sources, such as a Recurrence Score

[00115] The gene subset herein identified as the "angiogenesis group" includes genes that regulate new blood capillary formation or that otherwise participate in "wound healing." The angiogenesis group includes genes that (1) are involved in biological pathways associated with wound healing functions; and (2) co-express with EFNB2 with a Peason correlation coefficient greater than or equal to 0.6.

[00116] The gene subset defined herein as the "apoptosis group" includes genes which are involved in apoptosis functions and genes that co-express with apoptosis group genes. "Apoptosis functions" are defined herein as a series of cellular signaling intended to positively or negatively induce apoptosis, or programmed cell death. The apoptosis group includes BIK and genes that co-express with BIK with a Pearson correlation coefficient greater than or equal to 0.6. The gene subset defined herein as the "cell signaling group" includes genes which are involved with signaling pathways impacting cell growth and apoptosis and genes that co-express with cell signaling group genes. The cell signaling group includes GADD45B and genes that co-express with GADD45B, with a Pearson correlation coefficient greater than or equal to 0.6. Exemplary genes that co-express with GADD45B are provided in Tables 4 and 5. Table 4 provides genes for which expression is highly correlated with validated prognostic and/or predictive genes (by rank and Pearson co-expresion coefficient). Table 5 provides the results of identification of genes through gene module/clique analysis of validated gene biomarkers.

[00117] The gene subset herein defined as the "transcription factor group" includes genes which are involved with transcription factor functions and genes that co-express with transcription factor group genes. "Transcription factor functions" are defined herein as the binding of specific DNA sequences to facilitate the transcription of DNA to RNA, either alone or as part of a complex. The transcription factor group includes RUNXl and genes that co-express with RUNXl with a Pearson correlation coefficient greater than or equal to 0.6. Exemplary co-expressed genes and/or gene cliques encompassed by the transcription factor group are provided in Tables 5 and 6.

[00118] The gene subset defined herein as the "MSI group" includes genes which are known to have a statistically significant correlation with microsatellite instability high (MSI- H) status and genes that co-express with MSI group genes. Practice guidelines indicate that MSI-H histology is one factor to consider in making cancer screening recommendations for colorectal cancer patients. (See, e.g., NCCN Practice Guidelines in Oncology, v.2.2008.) The MSI group includes AXIN2 and genes that are (1) significantly associated with MSI-H status; or (2) co-express with AXIN2 with a correlation coefficient greater than or equal to 0.4. Exemplary co-expressed genes and/or gene cliques encompassed by the MSI group are provided in Table 5.

[00119] The present disclosure also provides methods to determine a threshold expression level for a particular gene. A threshold expression level may be calculated for a prognostic or predictive gene. A threshold expression level for a gene may be based on a normalized expression level. In one example, a C_t threshold expression level may be calculated by assessing functional forms using logistic regression.

[00120] The disclosure further provides methods to determine genes that co-express with particular target genes identified by quantitative RT-PCR (qRT-PCR), e.g. validated biomarkers relevant to a particular type of cancer. The co-expressed genes are themselves useful biomarkers. The co-expressed genes may be substituted for the prognostic or predictive gene marker with which they co-express. The methods can include identifying gene cliques from microarray data, normalizing the microarray data, computing a pairwise Spearman correlation matrix for the array probes, filtering out significant co-expressed probes across different studies, building a graph, mapping the probe to genes, and generating a gene clique report. For example, the expression levels of one or more genes of a prognostic and/or predictive gene clique may be used to calculate the likelihood that a patient with colorectal cancer will experience a recurrence and/or respond to chemotherapy. A "prognostic gene clique", as used herein, refers to a gene clique that includes a prognostic gene. A "predictive gene clique", as used herein, refers to a gene clique that includes a predictive gene.

[00121] Various technological approaches for determination of expression levels of the disclosed genes are set forth in this specification, including, without limitation, RT-PCR, microarrays, high-throughput sequencing, serial analysis of gene expression (SAGE) and Digital Gene Expression (DGE), which will be discussed in detail below. In particular aspects, the expression level of each gene may be determined in relation to various features of the expression products of the gene including exons, introns, protein epitopes and protein activity. One or more of the prognostic and/or predictive genes, or their expression products, may be analyzed for micros atellite instability (MSI) status.

[00122] The expression levels of prognostic and/or predictive genes may be measured in tumor tissue. For example, the tumor tissue is obtained upon surgical removal or resection of the tumor, or by tumor biopsy. The expression level of prognostic and/or predictive genes may also be measured in tumor cells recovered from sites distant from the tumor, for example circulating tumor cells, body fluid (e.g., urine, blood, blood fraction, etc.). [00123] The expression product that is assayed can be, for example, RNA or a polypeptide. The expression product may be fragmented. For example, the assay may use primers that are complementary to target sequences of an expression product and could thus measure full transcripts as well as those fragmented expression products containing the target sequence. Further information is provided in Tables A and B (inserted in specification prior to claims).

[00124] The RNA expression product may be assayed directly or by detection of a cDNA product resulting from a PCR-based amplification method, e.g., quantitative reverse transcription polymerase chain reaction (qRT-PCR). (See e.g., U.S. Pub. No. US2006- 0008809A1.) Polypeptide expression product may be assayed using immunohistochemistry (IHC). Further, both RNA and polypeptide expression products may also be is assayed using microarrays.

CLINICAL UTILITY

[00125] The algorithm-based assay and associated information provided by the practice of the methods disclosed herein facilitates physicians in making more well-informed treatment decisions, and to customize the treatment of colorectal cancer to the needs of individual patients, thereby maximizing the benefit of treatment and minimizing the exposure of patients to unnecessary treatments which may provide little or no significant benefits and often carry serious risks due to toxic side-effects.

[00126] Multi-analyte gene expression tests can be used measure the expression level of one or more genes involved in each of several relevant physiologic processes or component cellular characteristics.

[00127] The algorithm used to calculate such a score in a method disclosed herein may group the expression level values of genes. The grouping of genes may be performed at least in part based on knowledge of the contribution of the genes according to physiologic functions or component cellular characteristics, such as in the groups discussed above. The formation of groups, in addition, can facilitate the mathematical weighting of the contribution of various expression levels to the recurrence and/or treatment scores. The weighting of a gene group representing a physiological process or component cellular characteristic can reflect the contribution of that process or characteristic to the pathology of the cancer and clinical outcome. Accordingly, the present disclosure provides subsets of the prognostic and predictive genes identified herein for use in the methods disclosed herein. [00128] Based on the determination of a recurrence and/or treatment score, patients can be partitioned into subgroups (e.g., tertiles or quartiles) based on a selected value(s) of the recurrence and/or treatment score(s), where all patients with values in a given range can be classified as belonging to a particular risk group or treatment benefit group. Thus, the values chosen will define subgroups of patients with respectively greater or lesser risk and/or greater or lesser benefit.

[00129] The utility of a gene marker in predicting colorectal cancer outcome and/or response to chemotherapy may not be unique to that marker. An alternative marker having an expression pattern that is parallel to that of a selected marker gene may be substituted for, or used in addition to, a test marker. Due to the co-expression of such genes, substitution of expression level values should have little impact on the overall prognostic and/or predictive utility of the test. The closely similar expression patterns of two genes may result from involvement of both genes in the same process and/or being under common regulatory control in colon tumor cells. The present disclosure thus contemplates the use of such co- expressed genes or gene sets as substitutes for, or in addition to, prognostic and/or predictive methods of the present disclosure.

[00130] The present methods can provide for identification of colorectal cancer patients are likely to recur after surgery, and who will benefit from adjuvant chemotherapy. Such methods can be used alone or in combination with other clinical methods for patient stratification, e.g., using pathologic (tumor grade and histology) or molecular markers (e.g., levels of expression of genes such as thymidine synthase, thymidine phosphorylase (TP), dihydropyrimidine dehydrogenase (DPD), or microsatellite instability (MSI) status). [00131] The algorithm-based molecular assay and associated information provided by the methods disclosed herein for predicting the clinical outcome in Stage II and/or Stage III cancers of the colon and/or rectum have utility in many areas, including in the development and appropriate use of drugs to treat Stage II and/or Stage III cancers of the colon and/or rectum, to stratify cancer patients for inclusion in (or exclusion from) clinical studies, to assist patients and physicians in making treatment decisions, provide economic benefits by targeting treatment based on personalized genomic profile, and the like. For example, the recurrence score may be used on samples collected from patients in a clinical trial and the results of the test used in conjunction with patient outcomes in order to determine whether subgroups of patients are more or less likely to show a response to a new drug than the whole group or other subgroups. Further, such methods can be used to identify from clinical data subsets of patients who can benefit from therapy. Additionally, a patient is more likely to be included in a clinical trial if the results of the test indicate a higher likelihood that the patient will have a poor clinical outcome if treated with surgery alone and a patient is less likely to be included in a clinical trial if the results of the test indicate a lower likelihood that the patient will have a poor clinical outcome if treated with surgery alone. [00132] Staging of rectal tumors can be carried out based on similar criteria as for colon tumor staging, although there are some differences resulting, for example, from differences in the arrangement of the draining lymph nodes. As a result, Stage Will rectal tumors bear a reasonable correlation to Stage II/III colon tumors as to their state of progression. As noted above, the rate of local recurrence and other aspects of prognosis differ between rectal cancer and colon cancer, and these differences may arise from difficulties in accomplishing total resection of rectal tumors. Nevertheless, there is no compelling evidence that there is a difference between colon cancer and rectal cancer as to the molecular characteristics of the respective tumors. Tests able to predict chemotherapy treatment benefit for rectal cancer patients have utility similar in nature as described for colon cancer tests and the same markers might well have utility in both cancer types.

[00133] Tests that identify patients more likely to be those that fail to respond to standard-of-care are useful in drug development, for example in identifying patients for inclusion in clinical trials testing the efficacy of alternative drugs. For example, 30-35% of Stage III colon cancer patients fail to survive five years when treated with fluorouracil-based chemotherapy after surgical resection of tumor. Preferential inclusion of these patients in a clinical trial for a new Stage III colon cancer treatment could substantially improve the efficiency and reduce the costs of such a clinical trial.

METHODS OF ASSAYING EXPRESSION LEVELS OF A GENE PRODUCT [00134] The methods and compositions of the present disclosure will employ, unless otherwise indicated, conventional techniques of molecular biology (including recombinant techniques), microbiology, cell biology, and biochemistry, which are within the skill of the art. Exemplary techniques are explained fully in the literature, such as, "Molecular Cloning: A Laboratory Manual", 2^nd edition (Sambrook et al., 1989); "Oligonucleotide Synthesis"

(MJ. Gait, ed., 1984); "Animal Cell Culture" (R.I. Freshney, ed., 1987); "Methods in Enzymology" (Academic Press, Inc.); "Handbook of Experimental Immunology", 4^th edition (D.M. Weir & CC. Blackwell, eds., Blackwell Science Inc., 1987); "Gene Transfer Vectors for Mammalian Cells" (J.M. Miller & M.P. Calos, eds., 1987); "Current Protocols in Molecular Biology" (F.M. Ausubel et al., eds., 1987); and "PCR: The Polymerase Chain Reaction", (Mullis et al., eds., 1994).

[00135] Methods of gene expression profiling include methods based on hybridization analysis of polynucleotides, methods based on sequencing of polynucleotides, and proteomics- based methods. Exemplary methods known in the art for the quantification of mRNA expression in a sample include northern blotting and in situ hybridization (Parker & Barnes, Methods in Molecular Biology 106:247-283 (1999)); RNAse protection assays (Hod, Biotechniques 13:852-854 (1992)); and PCR-based methods, such as reverse transcription PCT (RT-PCR) (Weis et al., Trends in Genetics 8:263-264 (1992)). Antibodies may be employed that can recognize sequence-specific duplexes, including DNA duplexes, RNA duplexes, and DNA-RNA hybrid duplexes or DNA-protein duplexes. Representative methods for sequencing-based gene expression analysis include Serial Analysis of Gene Expression (SAGE), and gene expression analysis by massively parallel signature sequencing (MPSS).

Reverse Transcriptase PCR (RT-PCR)

[00136] Typically, mRNA is isolated from a test sample. The starting material is typically total RNA isolated from a human tumor, usually from a primary tumor. Optionally, normal tissues from the same patient can be used as an internal control. mRNA can be extracted from a tissue sample, e.g., from a sample that is fresh, frozen (e.g. fresh frozen), or paraffin-embedded and fixed (e.g. formalin-fixed).

[00137] General methods for mRNA extraction are well known in the art and are disclosed in standard textbooks of molecular biology, including Ausubel et al., Current Protocols of Molecular Biology, John Wiley and Sons (1997). Methods for RNA extraction from paraffin embedded tissues are disclosed, for example, in Rupp and Locker, Lab Invest. 56:A67 (1987), and De Andres et al, BioTechniques 18:42044 (1995). In particular, RNA isolation can be performed using a purification kit, buffer set and protease from commercial manufacturers, such as Qiagen, according to the manufacturer's instructions. For example, total RNA from cells in culture can be isolated using Qiagen RNeasy mini-columns. Other commercially available RNA isolation kits include MasterPure™ Complete DNA and RNA Purification Kit (EPICENTRE®, Madison, WI), and Paraffin Block RNA Isolation Kit (Ambion, Inc.). Total RNA from tissue samples can be isolated using RNA Stat-60 (Tel- Test). RNA prepared from tumor can be isolated, for example, by cesium chloride density gradient centrifugation. [00138] The sample containing the RNA is then subjected to reverse transcription to produce cDNA from the RNA template, followed by exponential amplification in a PCR reaction. The two most commonly used reverse transcriptases are avilo myeloblastosis virus reverse transcriptase (AMV-RT) and Moloney murine leukemia virus reverse transcriptase (MMLV-RT). The reverse transcription step is typically primed using specific primers, random hexamers, or oligo-dT primers, depending on the circumstances and the goal of expression profiling. For example, extracted RNA can be reverse-transcribed using a Gene Amp RNA PCR kit (Perkin Elmer, CA, USA), following the manufacturer's instructions. The derived cDNA can then be used as a template in the subsequent PCR reaction.

[00139] PCR-based methods use a thermostable DNA-dependent DNA polymerase, such as a Taq DNA polymerase. For example, TaqMan® PCR typically utilizes the 5'- nuclease activity of Taq or Tth polymerase to hydrolyze a hybridization probe bound to its target amplicon, but any enzyme with equivalent 5' nuclease activity can be used. Two oligonucleotide primers are used to generate an amplicon typical of a PCR reaction product. A third oligonucleotide, or probe, can be designed to facilitate detection of a nucleotide sequence of the amplicon located between the hybridization sites the two PCR primers. The probe can be detectably labeled, e.g., with a reporter dye, and can further be provided with both a fluorescent dye, and a quencher fluorescent dye, as in a Taqman® probe configuration. Where a Taqman® probe is used, during the amplification reaction, the Taq DNA polymerase enzyme cleaves the probe in a template-dependent manner. The resultant probe fragments disassociate in solution, and signal from the released reporter dye is free from the quenching effect of the second fluorophore. One molecule of reporter dye is liberated for each new molecule synthesized, and detection of the unquenched reporter dye provides the basis for quantitative interpretation of the data.

[00140] TaqMan® RT-PCR can be performed using commercially available equipment, such as, for example, ABI PRISM 7700™ Sequence Detection System™ (Perkin- Elmer- Applied Biosystems, Foster City, CA, USA), or Lightcycler (Roche Molecular Biochemicals, Mannheim, Germany). In a preferred embodiment, the 5' nuclease procedure is run on a real-time quantitative PCR device such as the ABI PRISM 7700™ Sequence Detection System™. The system consists of a thermocycler, laser, charge-coupled device (CCD), camera and computer. The system amplifies samples in a 384-well format on a thermocycler. The RT-PCR may be performed in triplicate wells with an equivalent of 2ng RNA input per 10 μL-reaction volume. During amplification, laser-induced fluorescent signal is collected in real-time through fiber optics cables for all wells, and detected at the CCD. The system includes software for running the instrument and for analyzing the data. [00141] 5'-Nuclease assay data are initially expressed as a threshold cycle ("C_t")-

Fluorescence values are recorded during every cycle and represent the amount of product amplified to that point in the amplification reaction. The threshold cycle (C_t) is generally described as the point when the fluorescent signal is first recorded as statistically significant. [00142] To minimize errors and the effect of sample-to-sample variation, RT-PCR is usually performed using an internal standard. The ideal internal standard gene (also referred to as a reference gene) is expressed at a constant level among cancerous and non-cancerous tissue of the same origin (i.e., a level that is not significantly different among normal and cancerous tissues), and is not significantly unaffected by the experimental treatment (i.e., does not exhibit a significant difference in expression level in the relevant tissue as a result of exposure to chemotherapy). For example, reference genes useful in the methods disclosed herein should not exhibit significantly different expression levels in cancerous colon as compared to normal colon tissue. RNAs most frequently used to normalize patterns of gene expression are mRNAs for the housekeeping genes glyceraldehyde-3-phosphate- dehydrogenase (GAPDH) and β-actin. Exemplary reference genes used for normalization comprise one or more of the following genes: ATP5E, GPXl, PGKl, UBB, and VDAC2. Gene expression measurements can be normalized relative to the mean of one or more (e.g., 2, 3, 4, 5, or more) reference genes. Reference-normalized expression measurements can range from 0 to 15, where a one unit increase generally reflects a 2-fold increase in RNA quantity.

[00143] Real time PCR is compatible both with quantitative competitive PCR, where internal competitor for each target sequence is used for normalization, and with quantitative comparative PCR using a normalization gene contained within the sample, or a housekeeping gene for RT-PCR. For further details see, e.g. Held et al, Genome Research 6:986-994 (1996).

[00144] The steps of a representative protocol for use in the methods of the present disclosure use fixed, paraffin-embedded tissues as the RNA source. mRNA isolation, purification, primer extension and amplification can be preformed according to methods available in the art. (see, e.g., Godfrey et al. J. Molec. Diagnostics 2: 84-91 (2000); Specht et al., Am. J. Pathol. 158: 419-29 (2001)). Briefly, a representative process starts with cutting about 10 μm thick sections of paraffin-embedded tumor tissue samples. The RNA is then extracted, and protein and DNA depleted from the RNA-containing sample. After analysis of the RNA concentration, RNA is reverse transcribed using gene specific primers followed by RT-PCR to provide for cDNA amplification products. .

Design of Intron-Based PCR Primers and Probes

[00145] PCR primers and probes can be designed based upon exon or intron sequences present in the mRNA transcript of the gene of interest. Primer/probe design can be performed using publicly available software, such as the DNA BLAT software developed by Kent, WJ., Genome Res. 12(4):656-64 (2002), or by the BLAST software including its variations. [00146] Where necessary or desired, repetitive sequences of the target sequence can be masked to mitigate non-specific signals . Exemplary tools to accomplish this include the Repeat Masker program available on-line through the Baylor College of Medicine, which screens DNA sequences against a library of repetitive elements and returns a query sequence in which the repetitive elements are masked. The masked intron sequences can then be used to design primer and probe sequences using any commercially or otherwise publicly available primer/probe design packages, such as Primer Express (Applied Biosystems); MGB assay- by-design (Applied Biosystems); Primer3 (Steve Rozen and Helen J. Skaletsky (2000) Primer3 on the WWW for general users and for biologist programmers. In: Rrawetz S, Misener S (eds) Bioinformatics Methods and Protocols: Methods in Molecular Biology. Humana Press, Totowa, NJ, pp 365-386).

[00147] Other factors that can influence PCR primer design include primer length, melting temperature (Tm), and G/C content, specificity, complementary primer sequences, and 3 '-end sequence. In general, optimal PCR primers are generally 17-30 bases in length, and contain about 20-80%, such as, for example, about 50-60% G+C bases, and exhibit Tm's between 50 and 80 ⁰C, e.g. about 50 to 70 ⁰C.

[00148] For further guidelines for PCR primer and probe design see, e.g. Dieffenbach,

CW. et al, "General Concepts for PCR Primer Design" in: PCR Primer, A Laboratory Manual, Cold Spring Harbor Laboratory Press, New York, 1995, pp. 133-155; Innis and Gelfand, "Optimization of PCRs" in: PCR Protocols, A Guide to Methods and Applications, CRC Press, London, 1994, pp. 5-11; and Plasterer, T.N. Primerselect: Primer and probe design. Methods MoI. Biol. 70:520-527 (1997), the entire disclosures of which are hereby expressly incorporated by reference.

[00149] Tables A and B provide further information concerning the primer, probe, and amplicon sequences associated with the Examples disclosed herein. MassARRAY® System

[00150] In MassARRAY-based methods, such as the exemplary method developed by

Sequenom, Inc. (San Diego, CA) following the isolation of RNA and reverse transcription, the obtained cDNA is spiked with a synthetic DNA molecule (competitor), which matches the targeted cDNA region in all positions, except a single base, and serves as an internal standard. The cDNA/competitor mixture is PCR amplified and is subjected to a post-PCR shrimp alkaline phosphatase (SAP) enzyme treatment, which results in the dephosphorylation of the remaining nucleotides. After inactivarion of the alkaline phosphatase, the PCR products from the competitor and cDNA are subjected to primer extension, which generates distinct mass signals for the competitor- and cDNA-derives PCR products. After purification, these products are dispensed on a chip array, which is pre-loaded with components needed for analysis with matrix- assisted laser desorption ionization time-of- flight mass spectrometry (MALDI-TOF MS) analysis. The cDNA present in the reaction is then quantified by analyzing the ratios of the peak areas in the mass spectrum generated. For further details see, e.g. Ding and Cantor, Proc. Natl. Acad. Sci. USA 100:3059-3064 (2003).

Other PCR-based Methods

[00151] Further PCR-based techniques that can find use in the methods disclosed herein include, for example, BeadArray® technology (Illumina, San Diego, CA; Oliphant et al., Discovery of Markers for Disease (Supplement to Biotechniques), June 2002; Ferguson et al., Analytical Chemistry 72:5618 (2000)); BeadsArray for Detection of Gene Expression® (BADGE), using the commercially available LuminexlOO LabMAP® system and multiple color-coded microspheres (Luminex Corp., Austin, TX) in a rapid assay for gene expression (Yang et al., Genome Res. 11:1888-1898 (2001)); and high coverage expression profiling (HiCEP) analysis (Fukumura et al., Nucl. Acids. Res. 31(16) e94 (2003).

Microarrays

[00152] Expression levels of a gene of interest can also be assessed using the microarray technique. In this method, polynucleotide sequences of interest (including cDNAs and oligonucleotides) are arrayed on a substrate. The arrayed sequences are then contacted under conditions suitable for specific hybridization with detectably labeled cDNA generated from mRNA of a test sample. As in the RT-PCR method, the source of mRNA typically is total RNA isolated from a tumor sample, and optionally from normal tissue of the same patient as an internal control or cell lines. mRNA can be extracted, for example, from frozen or archived paraffin-embedded and fixed (e.g. formalin-fixed) tissue samples. [00153] For example, PCR amplified inserts of cDNA clones of a gene to be assayed are applied to a substrate in a dense array. Usually at least 10,000 nucleotide sequences are applied to the substrate. For example, the microarrayed genes, immobilized on the microchip at 10,000 elements each, are suitable for hybridization under stringent conditions. Fluorescently labeled cDNA probes may be generated through incorporation of fluorescent nucleotides by reverse transcription of RNA extracted from tissues of interest. Labeled cDNA probes applied to the chip hybridize with specificity to each spot of DNA on the array. After washing under stringent conditions to remove non-specifically bound probes, the chip is scanned by confocal laser microscopy or by another detection method, such as a CCD camera. Quantitation of hybridization of each arrayed element allows for assessment of corresponding mRNA abundance.

[00154] With dual color fluorescence, separately labeled cDNA probes generated from two sources of RNA are hybridized pair wise to the array. The relative abundance of the transcripts from the two sources corresponding to each specified gene is thus determined simultaneously. The miniaturized scale of the hybridization affords a convenient and rapid evaluation of the expression pattern for large numbers of genes. Such methods have been shown to have the sensitivity required to detect rare transcripts, which are expressed at a few copies per cell, and to reproducibly detect at least approximately two-fold differences in the expression levels (Schena et at, Proc. Natl. Acad. ScL USA 93(2): 106-149 (1996)). Microarray analysis can be performed by commercially available equipment, following manufacturer's protocols, such as by using the Affymetrix GenChip® technology, or Incyte's microarray technology.

Serial Analysis of Gene Expression (SAGE)

[00155] Serial analysis of gene expression (SAGE) is a method that allows the simultaneous and quantitative analysis of a large number of gene transcripts, without the need of providing an individual hybridization probe for each transcript. First, a short sequence tag (about 10-14 bp) is generated that contains sufficient information to uniquely identify a transcript, provided that the tag is obtained from a unique position within each transcript. Then, many transcripts are linked together to form long serial molecules, that can be sequenced, revealing the identity of the multiple tags simultaneously. The expression pattern of any population of transcripts can be quantitatively evaluated by determining the abundance of individual tags, and identifying the gene corresponding to each tag. For more details see, e.g. Velculescu et al., Science 270:484-487 (1995); and Velculescu et al., Cell 88:243-51 (1997). Gene Expression Analysis by Nucleic Acid Sequencing

Nucleic acid sequencing technologies are suitable methods for analysis of gene expression. The principle underlying these methods is that the number of times a cDNA sequence is detected in a sample is directly related to the relative expression of the mRNA corresponding to that sequence. These methods are sometimes referred to by the term Digital Gene Expression (DGE) to reflect the discrete numeric property of the resulting data. Early methods applying this principle were Serial Analysis of Gene Expression (SAGE) and Massively Parallel Signature Sequencing (MPSS). See, e.g., S. Brenner, et al., Nature Biotechnology 18(6):630-634 (2000). More recently, the advent of "next-generation" sequencing technologies has made DGE simpler, higher throughput, and more affordable. As a result, more laboratories are able to utilize DGE to screen the expression of more genes in more individual patient samples than previously possible. See, e.g., J. Marioni, Genome Research 18(9):1509-1517 (2008); R. Morin, Genome Research 18(4):610-621 (2008); A. Mortazavi, Nature Methods 5(7):621-628 (2008); N. Cloonan, Nature Methods 5(7):613-619 (2008).

Isolating RNA from Body Fluids

[00156] Methods of isolating RNA for expression analysis from blood, plasma and serum (See for example, Tsui NB et al. (2002) 48,1647-53 and references cited therein) and from urine (See for example, Boom R et al. (1990) J Clin Microbiol. 28, 495-503 and reference cited therein) have been described.

Immunohistochemistrv

[00157] Immunohistochemistry methods are also suitable for detecting the expression levels of genes and applied to the method disclosed herein. Antibodies (e.g., monoclonal antibodies) that specifically bind a gene product of a gene of interest can be used in such methods. The antibodies can be detected by direct labeling of the antibodies themselves, for example, with radioactive labels, fluorescent labels, hapten' labels such as, biotin, or an enzyme such as horse radish peroxidase or alkaline phosphatase. Alternatively, unlabeled primary antibody can be used in conjunction with a labeled secondary antibody specific for the primary antibody. Immunohistochemistry protocols and kits are well known in the art and are commercially available.

Proteomics

[00158] The term "proteome" is defined as the totality of the proteins present in a sample (e.g. tissue, organism, or cell culture) at a certain point of time. Proteomics includes, among other things, study of the global changes of protein expression in a sample (also referred to as "expression proteomics"). Proteomics typically includes the following steps: (1) separation of individual proteins in a sample by 2-D gel electrophoresis (2-D PAGE); (2) identification of the individual proteins recovered from the gel, e.g. my mass spectrometry or N- terminal sequencing, and (3) analysis of the data using bioinformatics.

General Description of the mRNA Isolation, Purification and Amplification [00159] The steps of a representative protocol for profiling gene expression using fixed, paraffin-embedded tissues as the RNA source, including mRNA isolation, purification, primer extension and amplification are provided in various published journal articles. (See, e.g., T.E. Godfrey et al,. J. Molec. Diagnostics 2: 84-91 (2000); K. Specht et al., Am. J. Pathol. 158: 419-29 (2001), M. Cronin, et al., Am J Pathol 164:35-42 (2004)). Briefly, a representative process starts with cutting a tissue sample section (e.g.about 10 μm thick sections of a paraffin-embedded tumor tissue sample). The RNA is then extracted, and protein and DNA are removed. After analysis of the RNA concentration, RNA repair is performed if desired. The sample can then be subjected to analysis, e.g., by reverse transcribed using gene specific promoters followed by RT-PCR.

STATISTICAL ANALYSIS OF GENE EXPRESSION LEVELS IN IDENTIFICATION OF MARKER GENES FOR USE IN PROGNOSTIC AND/OR PREDICTIVE METHODS [00160] One skilled in the art will recognize that there are many statistical methods that may be used to determine whether there is a significant relationship between an outcome of interest (e.g., likelihood of survival, likelihood of response to chemotherapy) and and expression levels of a marker gene as described here. This relationship can be presented as a continuous recurrence score (RS), or patients may stratified into risk groups (e.g., low, intermediate, high). For example, a Cox proportional hazards regression model may fit to a particular clinical endpoint (e.g., RFI, DFS, OS). One assumption of the Cox proportional hazards regression model is the proportional hazards assumption, i.e. the assumption that effect parameters multiply the underlying hazard. Assessments of model adequacy may be performed including, but not limited to, examination of the cumulative sum of martingale residuals. One skilled in the art would recognize that there are numerous statistical methods that may be used (e.g., Royston and Parmer (2002), smoothing spline, etc.) to fit a flexible parametric model using the hazard scale and the Weibull distribution with natural spline smoothing of the log cumulative hazards function, with effects for treatment (chemotherapy or observation) and RS allowed to be time-dependent. (See, P. Royston, M. Parmer, Statistics in Medicine 21(15:2175-2197 (2002).) The relationship between recurrence risk and (1) recurrence risk groups; and (2) clinical/pathologic covariates (e.g., number of nodes examined, pathological T stage, tumor grade, MSI status, lymphatic or vascular invasion, etc.) may also be tested for significance.

[00161] Many statistical methods may be used to determine if there is a significant interaction between expression levels of predictive genes and beneficial response to treatment ("treatment benefit"). For example, this relationship can be presented as a continuous treatment score (TS), or patients may stratified into benefit groups (e.g., low, intermediate, high). The interaction studied may vary, e.g. standard of care vs. new treatment, or surgery alone vs. surgery followed by chemotherapy. For example, a Cox proportional hazards regression could be used to model the follow-up data, i.e. censoring time to recurrence at a certain time (e.g., 3 years) after randomization for patients who have not experienced a recurrence before that time, to determine if the TS is associated with the magnitude of chemotherapy benefit. One might use the likelihood ratio test to compare the reduced model with RS, TS and the treatment main effect, with the full model that includes RS, TS, the treatment main effect, and the interaction of treatment and TS. A pre-determined p- value cutoff (e.g., p < 0.05) may be used to determine significance.

[00162] Alternatively, the method of Royston and Parmer (2002) can be used to fit a flexible parametric model using the hazard scale and the Weibull distribution with natural spline smoothing of the log cumulative hazards function, with effects for treatment (chemotherapy or observation), RS, TS and the interaction of TS with treatment, allowing the effects of RS, TS and TS interaction with treatment to be time dependent. To assess relative chemotherapy benefit across the benefit groups, pre-specified cut-points for the RS and TS may be used to define low, intermediate, and high chemotherapy benefit groups. The relationship between treatment and (1) benefit groups; and (2) clinical/pathologic covariates may also be tested for significance. For example, one skilled in the art could identify significant trends in absolute chemotherapy benefit for recurrence at 3 years across the low, intermediate, and high chemotherapy benefit groups for surgery alone or surgery followed by chemotherapy groups. An absolute benefit of at least 3-6% in the high chemotherapy benefit group would be considered clinically significant.

[00163] In an exemplary embodiment, power calculations were carried for the Cox proportional hazards model with a single non-binary covariate using the method proposed by F. Hsieh and P. Lavori, Control Clin Trials 21:552-560 (2000) as implemented in PASS 2008. COEXPRESSION ANALYSIS

[00164] The present disclosure provides genes that co-express with particular prognostic and/or predictive gene that has been identified as having a significant correlation to recurrence and/or treatment benefit. To perform particular biological processes, genes often work together in a concerted way, i.e. they are co-expressed. Co-expressed gene groups identified for a disease process like cancer can serve as biomarkers for disease progression and response to treatment. Such co-expressed genes can be assayed in lieu of, or in addition to, assaying of the prognostic and/or predictive gene with which they are co-expressed. [00165] One skilled in the art will recognize that many co-expression analysis methods now known or later developed will fall within the scope and spirit of the present invention. These methods may incorporate, for example, correlation coefficients, co-expression network analysis, clique analysis, etc., and may be based on expression data from RT-PCR, microarrays, sequencing, and other similar technologies. For example, gene expression clusters can be identified using pair- wise analysis of correlation based on Pearson or Spearman correlation coefficients. (See, e.g., Pearson K. and Lee A., Biometrika 2, 357 (1902); C. Spearman, Amer. J. Psychol 15:72-101 (1904); J. Myers, A. Well, Research Design and Statistical Analysis, p. 508 (2^nd Ed., 2003).) In general, a correlation coefficient of equal to or greater than 0.3 is considered to be statistically significant in a sample size of at least 20. (See, e.g., G. Norman, D. Streiner, Biostatistics: The Bare Essentials, 137-138 (3rd Ed. 2007).)

GENERAL DESCRIPTION OF EXEMPLARY EMBODIMENTS

[00166] This disclosure provides a method to determine a patient's likelihood of experiencing a cancer recurrence by assaying expression levels of certain prognostic genes from a tumor sample obtained from the patient. Such methods involve use of gene subsets that are created based on similar functions of gene products. For example, prognostic methods disclosed herein involve assaying expression levels of gene subsets that include at least one gene each from each of a stromal group and a cell cycle group, and calculating a recurrence score (RS) for the patient by weighting the expression levels of each of the gene subsets by their respective contributions to cancer recurrence. The weighting may be different for each gene subset, and may be either positive or negative. For example, the stromal group score could be weighted by multiplying by a factor of 0.15, the cell cycle group score by a factor of -0.3, the cell signaling group score by a factor of 0.15, and so on. Gene subsets in such prognostic methods can further include at least one gene from a cell signaling group, apoptosis group, or transcription factor group.

[00167] For example, the weights assigned to each gene subset in the exemplary embodiments is set forth below:

RSl = W_s x Stromal Group Score

+ W_z x Angiogenesis Group Score

- W_cc x Cell Cycle Group Score

+ W_cs x Cell Signaling Group Score

- W_a x Apoptosis Group Score

Where: Stromal Group Score = (SGl + ...SGn)/n (SG = Stromal gene normalized expression level (NEL))

Cell Cycle Group Score = (CCGl + ...CCGn)Ai (CCG = Cell cycle gene NEL) Cell Signaling Group Score = (CSGl + ...CSGn) (CSG = Cell signaling gene

NEL)

Apoptosis Group Score = (AGl + ... AGn)Ai (AG = Apoptosis gene) Angiogenesis Group Score = (AgGl + ... AgGn)Ai (AgG = Angiogenesis gene) W_x = weighting factor for each gene subset

[00168] Alternatively, the genes within each gene subset may be weighted individually. Assuming standardized expression, the weights assigned to each gene subset in the exemplary embodiment is set forth below:

[00169] Stromal Group Score₂ = +BGN score + FAP score + INHBA score

[00170] Cell Cycle Group Score₂ = -2[Ki-67 score + MAD2L1 score + .75(cMYC score) + .25(MYBL2 score)]

[00171] Apoptosis Group Score₂ = -2(BIK score)

[00172] Cell Signaling Group Score₂ = +.33(GADD45B score)

[00173] Angiogenesis Group Score₂ = +EFNB2 score

[00174] To translate the RS2 model into non-standardized expression, the weights may be divided by gene standard deviation. For example, assuming non- standardized expression, the weights assigned to each gene subset in the exemplary embodiment is set forth below: [00175] Stromal Group Score_ns = +1.06(BGN score) + 1.38(FAP score) + 1.14(INHBA score)

[00176] Angiogenesis Group Score_ns = +1.34(EFNB2)

[00177] Cell Signaling Group Score_ns = +0.44GADD45B

[00178] Cell Cycle Group Score_ns = -2[ 1.85(Ki-67 score) + 1.32(MAD2L1 +

0.83(cMYC score) + 0.45(MYB L2 score)]

[00179] Apoptosis Group Score_ns = -2(BIK score)

[00180] In exemplary embodiments, RS is calculated using expression levels of one or more of BGN, FAP, INHBA, EFNB2, MYB L2, Ki-67, cMYC, MAD2L1, HSPEl,

GADD45B, BIK, and RUNXl. The disclosure provides substitute prognostic genes, the expression levels of which may similarly be used to calculate RS. These substitute predictive genes include genes that co-express with BGN, FAP, INHBA, EFNB2, MYB L2, Ki-67, cMYC, MAD2L1, HSPEl, GADD45B, BIK, or RUNXl

[00181] The RS_U (recurrence score unsealed) may be rescaled, for example to be between 0 and 100. More particularly, the RS_U may be rescaled as follows:

0 if 44X(RS_1J +0.82) < 0

RS = 44X(RS₁₁ + 0.82) if O < 44X(RS₁₁ + 0.82) < 100

100 if 44X(RS_1J + 0.82) > 100

[00182] The RS may be used to determine a recurrence risk group for each patient. For example, recurrence scores may be divided into three risk classification groups using predefined cut-points. The cut-points between the low, intermediate, and high recurrence risk groups may be defined, for example, as in Table 1.

Table 1: Recurrence Risk Stratification

[00183] The RS may be rounded to the nearest integer before the cut-points defining recurrence risk groups are applied. [00184] The disclosure also provides methods to determine the likelihood that a patient with colorectal cancer will have a beneficial response to chemotherapy including assaying expression levels of predictive genes, where the expression levels are used in an algorithm based on gene subsets that include at least one gene each from a growth factor receptor group, an apoptosis group, and a MSI group, and calculating a treatment score (TS) for the patient by weighting the expression levels of each of the gene subsets by their respective contributions to response to chemotherapy. The weighting may be different for each gene subset, and may be either positive or negative. For example, the stromal group could be weighted by multiplying by a factor of -0.3, the transcription factor by a factor of -0.04, the apoptosis group by a factor of 0.3, the cell cycle group by a factor of 0.1, and the MSI group by a factor of 0.1. The gene subsets may additionally comprise at least one gene from a transcription factor group and/or a cell cycle group.

[00185] In the exemplary embodiments, the weights assigned to each gene subset is set forth below:

TS = - W_s x Stromal Group Score

- W_tf x Transcription Factor Group Score

+ W_a x Apoptosis Group Score

+ W_cc x Cell Cycle Group Score

+ W_msi x MSI Group Score Where: Stromal Group Score = (SGl + ...SGn) (SG = stromal gene normalized expression level

(NEL)) Transcription Factor Group Score = (TFGl +

...TFGn) (TFG = transcription factor gene

NEL) Apoptosis Group Score = (AGl +...AGn) (AG

= apoptosis gene NEL) Cell Cycle Group Score = (CCGl + ...CCGn)

(CCG = cell cycle gene NEL) MSI Group Score = (MGl + ...MGn) (MG =

MSI gene NEL) W_x = weighting factor for each gene subset

[00186] In exemplary embodiments, TS is calculated using expression levels for

AXIN2, BIK, EFNB2, HSPEl, MAD2L1, and RUNXl.

[00187] The disclosure provides other predictive genes, the expression levels of which may similarly be used to calculate a TS. These substitute predictive genes include RANB P2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2, and gene that co-express with said substitute predictive genes with a Pearson correlation coefficient of at least 0.60.

[00188] The TS_U (Treatment Score unsealed) may be rescaled, for example it may be rescaled to be between 0 and 100. More particularly, TS_U may be rescaled as follows:

0 If STx(TS_1J -I) < 0

TS = 37X(TS₁₁ -I) if O < 37X(TS₁₁ -I) < 100

100 if 37X(TS_1J -I) > 100

[00189] In addition, the TS may be used to determine a "benefit score" for each patient. For example, the patient may be classified as one who is expected to have a low, medium, or high benefit from chemotherapy. In a particular example, the RS, TS, and predefined cut-points can be used to determine a benefit score for each patient. The low, intermediate, and high benefit scores or groups may be defined as in Table 2. Table 2: Beneficial Response to Chemotherapy Stratification

Data aggregation

[00190] The expression data may be aggregated. The purpose of data aggregation is to combine information across replicate qRT-PCR wells for individual genes. For example, during qRT-PCR, triplicate wells may be run for each gene and sample. Valid triplicate wells for each gene may be aggregated into a single weighted average C_t value. The resulting weighted average C_t effectively down weights the influence of outlier observations. The data aggregation module may include the following steps for each gene and sample:

(1) Retrieve calculated C_t values and status data.

(2) Aggregate plate level statistics and record module version, date and time of processing.

(3) Aggregate C_t values for each gene and store statistics using all wells (valid and invalid).

(4) Compute gene validity based on the number of valid wells.

(5) Compute the weighted average of the valid wells for each gene. Normalization of Expression Levels

[00191] The expression data used in the methods disclosed herein can be normalized.

Normalization refers to a process to correct for (normalize away), for example, differences in the amount of RNA assayed and variability in the quality of the RNA used, to remove unwanted sources of systematic variation in C_t measurements, and the like. With respect to RT-PCR experiments involving archived fixed paraffin embedded tissue samples, sources of systematic variation are known to include the degree of RNA degradation relative to the age of the patient sample and the type of fixative used to store the sample. Other sources of systematic variation are attributable to laboratory processing conditions. [00192] Assays can provide for normalization by incorporating the expression of certain normalizing genes, which genes do not significantly differ in expression levels under the relevant conditions. Exemplary normalization genes include housekeeping genes such as PGKl and UBB. (See, e.g., E. Eisenberg, et al., Trends in Genetics 19(7):362-365 (2003).) Normalization can be based on the mean or median signal (C_T) of all of the assayed genes or a large subset thereof (global normalization approach). In general, the normalizing genes, also referred to as reference genes should be genes that are known not to exhibit significantly different expression in colorectal cancer as compared to non-cancerous colorectal tissue, and are not significantly affected by various sample and process conditions, thus provide for normalizing away extraneous effects.

[00193] Unless noted otherwise, normalized expression levels for each mRNA/tested tumor/patient will be expressed as a percentage of the expression level measured in the reference set. A reference set of a sufficiently high number (e.g. 40) of tumors yields a distribution of normalized levels of each mRNA species. The level measured in a particular tumor sample to be analyzed falls at some percentile within this range, which can be determined by methods well known in the art.

[00194] In exemplary embodiments, one or more of the following genes are used as references by which the expression data is normalized: ATP5E, GPXl, PGKl, UBB, and VDAC2. The calibrated weighted average C_t measurements for each of the prognostic and predictive genes may be normalized relative to the mean of five or more reference genes. [00195] Those skilled in the art will recognize that normalization may be achieved in numerous ways, and the techniques described above are intended only to be exemplary, not exhaustive. Bridging Expression Measurements and Calibration

[00196] An oligonucleotide set represents a forward primer, reverse primer, and probe that are used to build a primer and probe (P3) pool and gene specific primer (GSP) pool. Systematic differences in RT-PCR cycle threshold (Ct) measurements can result between different oligonucleotide sets due to inherent variations oligonucleotide syntheses. For example, differences in oligonucleotide sets may exist between development, production (used for validation), and future production nucleotide sets. Thus, use of statistical calibration procedures to adjust for systematic differences in oligonucleotide sets resulting in translation in the gene coefficients used in calculating RS and TS may be desirable. For example, for each of the genes assayed for use in an algorithm, one may use a scatterplot of C_t measurements for production oligonucleotide sets versus C_t measurements from a corresponding sample used in different oligonucleotide set to create linear regression model that treats the effect of lot-to-lot differences as a random effect. Examination of such a plot will reveal that the variance of C_t measurements increases exponentially as a function of the mean C_t. The random effects linear regression model can be evaluated with log-linear variance, to obtain a linear calibration equation. A calculated mean squared error (MSE) for the scores can be compared to the MSE if no calibration scheme is used at all. [00197] As another example, a latent variable measurement of C_t (e.g. first principle component) may be derived from various oligonucleotide sets. The latent variable is a reasonable measure of the "true" underlying C_t measurement. Similar to the method described above, a linear regression model may be fit to the sample pairs treating the effects of differences as a random effect, and the weighted average C_t value adjusted to a calibrated

Centering and Data Compression/Scaling

[00198] Systematic differences in the distribution of patient RS and TS due to analytical or sample differences may exist between early development, clinical validation and commercial samples. A constant centering tuning parameter may be used in the algorithm to account for such difference.

[00199] Data compression is a procedure used to reduce the variability in observed normalized C_t values beyond the limit of quantitation (LOQ) of the assay. Specifically, for each of the colon cancer assay genes, variance in C_t measurements increase exponentially as the normalized C_t for a gene extends beyond the LOQ of the assay. To reduce such variation, normalized C_t values for each gene may be compressed towards the LOQ of the assay. Additionally, normalized C_t values may be rescaled. For example, normalized C_t values of the prognostic, predictive, and reference genes may be rescaled to a range of 0 to 15, where a one-unit increase generally reflects a 2-fold increase in RNA quantity.

Threshold Values

[00200] The present invention describes a method to determine a threshold value for expression of a cancer-related gene, comprising measuring an expression level of a gene, or its expression product, in a tumor section obtained from a cancer patient, normalizing the expression level to obtain a normalized expression level, calculating a threshold value for the normalized expression level, and determining a score based on the likelihood of recurrence or clinically beneficial response to treatment, wherein if the normalized expression level is less than the threshold value, the threshold value is used to determine the score, and wherein if the normalized expression level is greater or equal to the threshold value, the normalized expression level is used to determine the score.

[00201] For example, a threshold value for each cancer-related gene may be determined through examination of the functional form of relationship between gene expression and outcome. Examples of such analyses are presented for Cox PH regression on recurrence free interval where gene expression is modeled using natural splines and for logistic regression on recurrence status where gene expression is modeled using lowess smoother-(See, e.g., Figures 6-10.)

[00202] Thresholded C_t values for each prognostic, predictive, and reference genes can be used to calculate RS and TS. Exemplary thresholded C_t values for the 18-gene assay described herein are set forth in Table 3. [00203] Table 3: Gene expression panel and threshold values

Gene Accession Threshold Gene Accession Threshold

Number Number

ATP5E NM_006886 None MYBL2 NM_002466 6

GPXl NM_000581 None Ki-67 NM_002417 6

PGKl NM_000291 None GADD45B NM_015675 4.5

UBB NM_018955 None EFNB2 NM_004093 4

VDAC2 NM_003375 None RUNXl NMJ)01754 4.5

BGN NM_001711 None BIK NMJ)Ol 197 4.5

FAP NM_004460 6 MAD2L1 NMJ)02358 $

INHBA NM_002192 None HSPEl NMJ)02157 None cMYC NM 002467 None AXIN2 NM 004655 None

Thresholded C_t values for each gene are calculated according to the formula: [if Normalized C_x < Threshold Thesholded C_x = Threshold [if Normalized C_x > Threshold Thresholded C_x = Normalized C_x

[00204] It will be appreciated by one of ordinary skill in the art that a purpose of thresholding is to address non-linear functional forms for gene expression measurements. However, it will be readily appreciated that other nonlinear transforms other than thresholding can be used to accomplish the same effect.

BUILDING GENE CLIQUES FROM VALIDATED BIOMARKERS

[00205] This disclosure contemplates using co-expressed genes and/or gene cliques, identified with respect to prognostic and/or predictive genes, as substitutes for, or for analysis with, the prognostic and/or predictive genes disclosed herein. One method disclosed to analyze gene cliques that co-express with a target gene (i.e, a gene of interest) involves normalizing microarray gene expression data for cancer tumor samples based on array probes, calculating a correlation coefficient (e.g., using Spearman or Pearson correlation coefficients) based on gene expression levels for every unique pair of array probes, determining significant probe pairs, wherein significant probe pairs are a target gene probe and an array probe with a correlation co-efficient greater than a significant threshold value (e.g., a Spearman correlation co-efficient > 0.5), mapping the target gene to its corresponding target gene probe, selecting a candidate probe set, wherein each candidate probe is part of a significant probe pair, and identifying an official gene symbol for each candidate probe (e.g., Entrez Gene Symbol). For example, Table 6 lists the gene cliques associated with FAP, INHBA, Ki-67, HSPEl, MAD2L1, and RUNXl. KITS OF THE INVENTION

[00206] The materials for use in the methods of the present invention are suited for preparation of kits produced in accordance with well known procedures. The present disclosure thus provides kits comprising agents, which may include gene-specific or gene- selective probes and/or primers, for quantitating the expression of the disclosed genes for predicting prognostic outcome or response to treatment. Such kits may optionally contain reagents for the extraction of RNA from tumor samples, in particular fixed paraffin- embedded tissue samples and/or reagents for RNA amplification. In addition, the kits may optionally comprise the reagent(s) with an identifying description or label or instructions relating to their use in the methods of the present invention. The kits may comprise containers (including microliter plates suitable for use in an automated implementation of the method), each with one or more of the various reagents (typically in concentrated form) utilized in the methods, including, for example, pre-fabricated microarrays, buffers, the appropriate nucleotide triphosphates (e.g., dATP, dCTP, dGTP and dTTP; or rATP, rCTP, rGTP and UTP), reverse transcriptase, DNA polymerase, RNA polymerase, and one or more probes and primers of the present invention (e.g., appropriate length poly(T) or random primers linked to a promoter reactive with the RNA polymerase). Mathematical algorithms used to estimate or quantify prognostic or predictive information are also properly potential components of kits.

REPORTS

[00207] The methods of this invention, when practiced for commercial diagnostic purposes, generally produce a report or summary of information obtained from the herein- described methods. For example, a report may include information concerning expression levels of prognostic and/or predictive genes, a prediction of the predicted clinical outcome or response to chemotherapy for a particular patient, or gene cliques or thresholds. The methods and reports of this invention can further include storing the report in a database. The method can create a record in a database for the subject and populate the record with data. The report may be a paper report, an auditory report, or an electronic record. The report may be displayed and/or stored on a computing device (e.g., handheld device, desktop computer, smart device, website, etc.). It is contemplated that the report is provided to a physician and/or the patient. The receiving of the report can further include establishing a network connection to a server computer that includes the data and report and requesting the data and report from the server computer. COMPUTER PROGRAM

[00208] The values from the assays described above, such as expression data, recurrence score, treatment score and/or benefit score, can be calculated and stored manually. Alternatively, the above-described steps can be completely or partially performed by a computer program product. The present invention thus provides a computer program product including a computer readable storage medium having a computer program stored on it. The program can, when read by a computer, execute relevant calculations based on values obtained from analysis of one or more biological sample from an individual (e.g., gene expression levels, normalization, thresholding, and conversion of values from assays to a score and/or graphical depiction of likelihood of recurrence/response to chemotherapy, gene co-expression or clique analysis, and the like). The computer program product has stored therein a computer program for performing the calculation.

[00209] The present disclosure provides systems for executing the program described above, which system generally includes: a) a central computing environment; b) an input device, operatively connected to the computing environment, to receive patient data, wherein the patient data can include, for example, expression level or other value obtained from an assay using a biological sample from the patient, or microarray data, as described in detail above; c) an output device, connected to the computing environment, to provide information to a user (e.g., medical personnel); and d) an algorithm executed by the central computing environment (e.g., a processor), where the algorithm is executed based on the data received by the input device, and wherein the algorithm calculates a RS, TS, risk or benefit group classification, gene co-expression analysis, thresholding, or other functions described herein. The methods provided by the present invention may also be automated in whole or in part.

[00210] All aspects of the present invention may also be practiced such that a limited number of additional genes that are co-expressed with the' disclosed genes, for example as evidenced by statistically meaningful Pearson and/or Spearman correlation coefficients, are included in a prognostic or predictive test in addition to and/or in place of disclosed genes. [00211] Having described the invention, the same will be more readily understood through reference to the following Examples, which are provided by way of illustration, and are not intended to limit the invention in any way. EXAMPLE 1: GENE EXPRESSION ANALYSIS FOR COLON CANCER RECURRENCE Methods and Materials: Patients and Samples

[00212] Tumor tissue samples were from two cohorts of patients with stage II or stage

III colon cancer treated with surgery alone form the basis for this report. Further details concerning the NSABP protocols C-Ol, C-02, C-03, and C-04 are available in C. Allegra, J Clin Oncology 21(2):241-250 (2003) and related U.S. Application Serial Nos. 11/653,102 and 12/075,813, the contents of which are incorporated herein by reference. [00213] The first cohort pooled available patient samples from NSABP protocols C-Ol or C-02 in which patients were randomly assigned to receive either colon resection alone or resection + bacillus Calmette-Guerin ("BCG") immunotherapy. The second cohort (CCF) included stage II and stage III colon cancer patients treated with surgery alone at CCF between the years 1981 and 2000. None of the patients in either group received adjuvant chemotherapy. In both cohorts, gene expression measurements were obtained from archived, formalin-fixed, paraffin-embedded (FPE) colon tumor tissue.

Differential Expression Data:

[00214] The final number of evaluable FPE blocks was 270 in the NSABP cohort and

765 in the CCF cohort (n = 1035). The primary reasons for exclusion were failure to meet minimum RNA yield (10% of samples in NSABP and 8% in CCF) and failure to meet quality control criteria for RT-qPCR (7% in NSABP and 2% in CCF).

[00215] The primary analysis in both studies investigated the relationship between the expression of 761 genes and RFI. This analysis identified sixty-five genes were found to be nominally significant in both studies. (See Figure 1.) The high level of agreement was observed between the univariate hazard ratios for 63 (97%) of 65 genes significantly related to RFI in both studies. Of the genes found to be significantly related to RFI in either study, the majority were also related to both DFS and OS within the same study. [00216] In both cohorts, the relationship between the expression of each gene and RFI was investigated, controlling for study and baseline characteristics. Any of the baseline clinical characteristics or study design attributes that had at least a modest association (p<0.2) with RFI were included in the multivariate analysis. Sixty-one (43%) of the 143 genes significant in univariate analyses in the NSABP cohort were statistically significant after controlling for nodal status, tumor location, tumor grade, mucinous tumor type, study protocol (C-Ol vs. C-02), treatment assignment (BCG vs. none), and year of surgery. Eighty- eight (74%) of the 119 genes significant in univariate analysis in the CCF cohort retained significance after adjustment for age, nodal status, number of lymph nodes examined, tumor grade, mucinous tumor type, fixative, surgery year and T stage. There was agreement between the multivariate hazard ratios for the 65 genes significantly related to RFI in both studies. The hazard ratios were concordant for 63 of 65 genes. The consistency of hazard ratio estimates from the uni- and multivariate Cox regression analyses indicates that expression levels of these genes provide prognostic information which is relatively independent of traditional clinical predictors.

[00217] These 65 genes represent pathways that would be expected to be important in colon cancer recurrence. To identify genes that were co-expressed and therefore possibly members of the same functional gene family, hierarchical cluster analysis and forest plots were created using the genes that were significantly related to RFI in that study (not shown) as well as for the genes significantly related to RFI in both studies. Cluster analysis identified that the majority (48) of the prognostic genes fell into two relatively distinct gene groups: a stromal gene group (containing several subgroups) and a cell cycle gene group. The stromal group contained genes which, when highly expressed, were associated with a worse outcome and increased likelihood of recurrence, such as BGN, FAP, INHBA, and EFNB2. The cell cycle group contained genes which, when highly expressed, were associated with a better outcome and decreased likelihood of recurrence, such as cMYC, MYB L2, Ki-67, MAD2L1, and HSPEl.

EXAMPLE 2: GENE EXPRESSION ANALYSIS FOR PROGNOSTIC AND PREDICTIVE GENES [00218] A study was conducted to assay gene expression levels in tumor samples obtained from patients with stage II or III colon cancer treated with surgery and 5FU/LV and perform analysis across four independent studies to identify genes that quantitate both the individual risk of recurrence in patients treated with surgery alone (prognosis) and the individual treatment benefit of 5-FU/LV adjuvant chemotherapy (prediction). Further information about these studies can be found in related U.S. Application Serial Nos. 11/653,102 and 12/075,813, the contents of which are incorporated herein by reference.

Methods and Materials

Patients and Samples

[00219] Tissue samples were obtained from two cohorts of patients with stage II or stage III colon cancer treated with surgery and 5FU/LV. The first cohort included available patient samples from the 5FU/LV arm of NSABP Study C-04 in which patients were randomly assigned to receive either 5FU/LV, 5FU + levamisole or 5FU/LV + levamisole. (See, N. Wolmark, et al., J Clin Oncol 17:3553-3559 (1999). The second cohort included available patient samples from the 5FU/LV arm of NSABP Study C-06 in which patients were randomly assigned to receive 5FU/LV or oral uracil/tegafur plus leucovorin. (See, B. Lembersky, et al., J Clin Oncol 24:2059-2064 (2006). The 5FU/LV regimen was the same in both studies. In both cohorts, gene expression measurements were obtained from archived, formalin-fixed, paraffin-embedded (FPE) colon tumor tissue.

[00220] Based on treatment assignment and eligibility in the original NSABP studies,

691 C-04 patients and 792 C-06 patients qualified for this study. Available formalin- fixed paraffin-embedded (FPE) blocks for patients enrolled in C-04 (n=360) and C-06 (n=573) were assayed. After applying pre-specified exclusion criteria, the final number of evaluable patients was 308 in the C-04 cohort and 508 in the C-06 cohort. The primary reasons for exclusion were failure to satisfy pathology requirements (8.6% in C-04 and 1.7% in C-06) and failure to meet clinical eligibility criteria (1.7% in C-04 and 7.5% in C-06).

Analysis methods

[00221] The primary analysis in both studies investigated the relationship between the expression of each gene and RFI. This analysis identified 143 (19%) of the 761 genes as being significantly related to RFI in the C-04 cohort compared to 169 (45%) of the 375 genes in the C-06 cohort. Seventy-five genes were found to be nominally significant in both studies. The hazard ratios were concordant (i.e. in similar direction) for 73 (97%) of these 75 genes. Of the genes found to be significantly related to RFI in either study, the majority were also related to both DFS and OS within the same study. Seventy-one (50%) of 143 genes significantly associated with RFI in univariate analyses in the C-04 study were statistically significant after controlling for nodal status and age. One hundred thirty-seven (81%) of the 169 genes significant in univariate analyses in the C-06 study were statistically significant after controlling for nodal involvement and mucinous tumor type. A high level of agreement between the univariate and multivariate hazard ratios for genes significantly related to RFI in both studies was observed.

[00222] To identify prognostic genes across the four colon development studies, the focus was on the genes which significantly and consistently associated with RFI in both surgery only (C-Ol/C-02 and CCF studies described in Example 1) and surgery + 5FU/LV- treated (C-04 and C-06) patients since prognostic genes are expected to have a similar relationship (i.e. similar direction and magnitude of the HR' s) with outcome when measured in patients treated with the standard of care or in patients treated with a new intervention. A total of 48 (13%) of 375 genes studied in all four development studies were significantly (p<0.05) associated with RFI in both surgery only studies and at least one surgery + 5FU/LV study. Due to type II error considerations, genes were not required to be significant in all four studies. The univariate hazard ratios and associated confidence intervals for the 48 genes in each of the four colon development studies are presented in Figure 2. Cluster analysis identified two relatively distinct gene groups among the 48 prognostic genes: a stromal activation gene group (containing several subgroups) and a cell cycle gene group. The stromal group contained genes which, when highly expressed, were associated with a worse outcome and increased likelihood of recurrence, such as BGN, FAP, INHBA, and EFNB2. The cell cycle group contained genes which, when highly expressed, were associated with a better outcome and decreased likelihood of recurrence, such as cMYC, MYB L2, Ki-67, MAD2L1, and HSPEl.

[00223] In contrast to prognostic genes, the predictive genes are expected to exhibit a different relationship with outcome (i.e. different HR' s) in patients treated with surgery only as compared to patients treated with surgery + 5FU/LV. To identify predictive genes, multivariate Cox proportional hazards models were examined, including main effects of gene and treatment and an interaction of gene and treatment for each of the 375 genes pooling the data across the four colon development studies. A total of 66 (18%) of 375 genes studied in all four development studies had an interaction of gene expression and treatment significant at 0.10 level. Only 4 of these 66 genes had significant association with RFI in the two independent surgery alone studies and at least one of the surgery+5 FU/LV study (i.e. were included in the set of 48 prognostic genes), indicating that a small minority of predictive genes are both prognostic and predictive. Fifty-nine of the 66 genes were not associated with RFI in both surgery only studies, indicating that the majority of predictive genes are not also prognostic genes.

[00224] These 66 genes represent pathways that would be expected to be important in response to chemotherapy. Cluster analysis identified two relatively distinct gene groups among 66 potentially predictive genes. One group contains a large number of cell cycle related genes such as centromere and spindle associated proteins (CENPA, KIFCl, KIF22, STK15, MAD2L1, AURKB), checkpoint regulation (CDC2, BUBl), and a DNA topoisomerase (TOP2A). The second group contains genes which represent several different biological pathways, including a tight group of stromal activation genes (BGN, SPARC, COLlAl, CDHI l, MMP2, and TIMPl), and genes associated with apoptosis (BIK), 5FU metabolism (UPP), and B-catenin/wnt signaling (AXIN2, LEF). It is of note that the two mismatch repair genes (MSH2 and MSH3) and several hypoxia/stress response genes (NR4A1, RhoB, HIFlA, CREBBP, PKR2, EPASl) were also associated with response to 5- FU/LV chemotherapy.

[00225] Preliminary prognostic models were built using subsets of the 48 prognostic genes. The results from a representative model containing 10 prognostic genes are shown in Figures 3a and 3b for stage II and stage III patients, respectively, treated with surgery only (C-Ol/C-02 and CCF cohorts). Patients were divided into three equally sized groups based on the calculated Recurrence Score. This model separated the 628 Stage II patients into groups with low, intermediate and high risk of recurrence: the lowest tertile had a 5% (95%CI 3%, 9%) risk of recurrence at 3 years vs. 14% (10%, 20%) and 22% (16%, 28%), respectively, for the middle and highest tertiles. (See, Figure 4a.) For 395 Stage III patients, the two lowest tertiles had 26% (19%, 35%) and 26% (19%, 34%) risk of recurrence at 3 years vs. a 47% (39%, 56%) risk for the highest tertile. (See Figure 4b.) For comparison, the overall 3-year risks of recurrence of Stage II and Stage III patients were 13% and 33%, respectively. When bootstrap was applied, the average Kaplan-Meier estimates (and associated 95% confidence intervals) of recurrence rates at 3 years for stage II patients were 5% (2%,9%), 12% (8%, 17%) and 22% (18%, 27%) for the 1^st, 2^nd and 3^rd tertile, respectively. For stage III patients, the corresponding estimates were 23% (16%, 30%), 28% (19%, 37%) and 48% (40%, 56%), respectively.

EXAMPLE 3: VALIDATION OF ALGORITHM-BASED MOLECULAR DIAGNOSTIC ASSAY [00226] After the 65 prognostic and 66 predictive gene candidates were identified, the genes were examined further for consistency in association between gene expression and RFI (prognosis) and differential relationship between with RFI in treated vs. untreated patients (prediction) across the four colon development studies using univariate and multivariate Cox proportional hazards models. Representation of the relevant biologic pathways, distribution of gene expression, functional form of the relationship between gene expression, and RFI and analytical performance of individual genes were also taken into account. [00227] Forest plots for the predictive genes (after thresholding) were reviewed and genes were identified that (1) displayed predictive effects either in both Stage II and Stage III colorectal cancer, or in Stage III only; (2) had significant (e.g., p<0.10) gene by treatment interaction in a model of gene (n = 9) or median C_t < 4 (n = 2); and (3) had significant (p < 0.10) gene by treatment interaction after RS_U and TRT were forced into the model. Genes with consistent univariate hazard ratios (HRs) were preferred. In addition, forest plots for the predictive genes were examined qualitatively and genes displaying predictive effects either in both Stage II and Stage III colorectal cancer, or in Stage III only were identified. Through this analysis the following additional 10 predictive gene candidates were identified (in addition to the 6 predictive genes in the final algorithm): RANBP2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2. Based on these results, multi-gene models were designed and analyzed across all four studies. Those analyses, together with a methodical evaluation of analytical performance of each candidate gene, led to the design of a multi-gene RT-PCR-based clinical assay to predict recurrence risk and treatment benefit from 5FU/L_V. The genes represent biological categories that are important in colon cancer: stromal group (BGN, FAP, INHBA, EFNB2), cell cycle group (Ki-67, MYB L2, cMYC, MAD2L1, HSPEl), cell signaling (GADD45B), apoptosis group (BIK), transcription factor group (RUNXl), and MSI group (AXIN2), as well as 5 reference genes (ATP5E, GPXl, PGKl, UBB, VDAC2) for normalization of gene expression.

Methods and Materials [00228] Patients and Samples

[00229] The developed algorithm may be validated using samples obtained from the

QUASAR study. The QUASAR Collaborative Group trial is the largest reported single randomized study of observation versus adjuvant chemotherapy in patients with resected stage II colon cancer. (See, Lancet 370:2020-2029 (2007).) In that study, patients with resected stage II and III colon and rectal cancer were assigned by treating physicians to one of two arms of the study based on either a "clear" or "uncertain" indication for adjuvant therapy. In the "clear" arm, all patients (n= 4320) received adjuvant 5-FU/leucovorin (LV) chemotherapy with or without levamisole. In the "uncertain" arm, patients (n=3239) were randomized to either observation (n=1617) or adjuvant 5-FU/LV chemotherapy (n=1622). As expected, the "clear" arm enrolled primarily stage III patients (70%), and the "uncertain" arm enrolled a high proportion of stage II patients (91% stage II, 71% colon cancer). [00230] These results from QUASAR demonstrate that adjuvant 5-FU/LV treatment benefits a small but significant subset of stage II colon cancer patients. (See, e.g., Figure 5.) Nevertheless, the physician managing stage II colon cancer still faces considerable challenges, including the fact that the majority of such patients are cured with surgery alone and that adjuvant 5-FU/LV chemotherapy carries potential toxicities of leucopenia, stomatitis, and diarrhea. Clearly, the decision to administer adjuvant 5-FU/LV chemotherapy would be greatly aided by the ability to identify reliably: 1) patients who are likely to be cured with surgery alone and 2) patients who are at substantial risk of recurrence following surgery and have a significant likelihood of clinical benefit with adjuvant treatment. With regard to the latter, it is worth emphasizing that the clinically relevant information for patients and oncologists includes not only the magnitude of the baseline risk of recurrence but also the magnitude of potential benefit (i.e. the absolute clinical benefit) associated with adjuvant 5-FU/LV treatment.

[00231] The validation study entailed the use of a pre-specified RT-PCR-based 18- gene clinical assay (see genes listed in Table 3) applied to archival paraffin-embedded tumor tissue specimens from colon cancer patients studied in QUASAR. The study considered the relationship between (1) a continuous RS and recurrence risk in patients randomized to surgery alone, and compared to that of patients randomized to surgery followed by adjuvant 5-FU/LV chemotherapy (controlling for simultaneous prognostic effects of clinical and pathological covariates); and (2) a continuous TS and chemotherapy benefit in patients randomized to surgery alone or surgery followed by adjuvant 5-FU/LV chemotherapy. The study comparied the risk of recurrence between the high and low recurrence risk groups based on pre-specified cut-points for RS. A two-fold higher recurrence risk at 3 years in the high recurrence risk group compared to the low recurrence risk group was considered clinically significant. Alternative clinical endpoints, including RFI, DFS and OS, were considered. The study also looked for a significant (1) trend in absolute chemotherapy benefit for recurrence at 3 years across the low, intermediate, and high chemotherapy benefit groups; (2) interaction between the continuous TS and treatment relative to alternative clinical endpoints, including RFI, OS and DFS; (3) interaction between MMR status and treatment after controlling for the prognostic effects of the continuous RS and prognostic covariates. [00232] Fixed paraffin-embedded colon tumor tissue from approximately 1,500 patients from QUASAR with stage II colon cancer. The RNA was extracted from the tumor tissue and RT-PCR analysis was conducted to determine expression levels of 13 cancer- related and 5 reference genes (Table 3). A prospectively-defined algorithm was used to calculate a RS and TS for each patient. Patients were classified into low, intermediate, and high recurrence risk groups using the RS and pre-specified cut-points (Table 1). Similarly, patients were classified into low, intermediate, and high chemotherapy benefit groups based on the combination of the RS and TS and on pre-specified cut-points (Table T). These cut- points define the boundaries between low and intermediate benefit groups and between intermediate and high benefit groups.

[00233] The specimens were also assessed by pathology to determine: tumor type, tumor grade, presence of lymphatic and/or vascular invasion, number of nodes examined, depth of invasion (pathologic T stage), MMR status, and other QC metrics. This information was used to determine whether there was a significant relationship between risk of recurrence and individual and pathologic covariates.

[00234] Expression levels of 13 cancer-related genes used in the calculation of the RS and TS were reported as values from the RT-PCR assay. Gene expression measurements were normalized relative to the mean of five reference genes (ATP5E, GPXl, PGKl, UBB, VDAC2). For each cancer-related gene, a cycle threshold (C_T) measurement was obtained by RT-PCR, and then normalized relative to a set of five reference genes. Reference-normalized expression measurements typically range from 0 to 15, where a one unit increase generally reflects a 2-fold increase in RNA quantity. [00235] Analysis methods:

[00236] Unless otherwise stated, all significance tests were conducted at the 0.05 significance level, and two-sided p- values and confidence intervals will be reported. To preserve the overall family-wise error rate for testing the primary objectives at the 0.05 significance level, the analysis applied conditional fixed sequential testing. A Cox proportional hazards regression model was fit to the clinical endpoint RFI for the patients who were randomized to surgery alone and a likelihood ratio test used to determine if the RS is significantly associated with the risk of recurrence (i.e. if the hazard ratio associated with the RS is significantly different from 1).

[00237] A Cox proportional hazards regression was used to model the first 3 years of follow-up data, that is, censoring time to recurrence at 3 years after randomization for patients who have not experienced a recurrence before that time, to determine if the TS is associated with the magnitude of chemotherapy benefit. The likelihood ratio test was used to compare the reduced model with RS, TS and the treatment main effect, with the full model that includes RS, TS, the treatment main effect, and the interaction of treatment and TS. In addition, we will use the method of Royston and Parmar (2002) to fit a flexible parametric model to RFI using all available follow-up data. The method will model the hazard of recurrence using the Weibull distribution with natural spline smoothing of the log cumulative hazards function, with effects for treatment (chemotherapy or observation), RS, TS and the interaction of TS with treatment, allowing the effects of treatment, RS, TS and TS interaction with treatment to be time dependent. The predicted effect of chemotherapy as a function of TS will be estimated at follow-up times of 2, 3, and 5 years.

[00238] Power calculations were carried for the Cox proportional hazards model with a single non-binary covariate using the method proposed by Hsieh and Lavori (2000) as implemented in PASS 2008. One skilled in the art would recognize that power at alpha 0.01- 0.05 alpha would be sufficient to control for type I error.

[00239] For example, a test comparing a reduced Cox proportional hazards regression model of gene expression and treatment to a full model containing gene expression, treatment and interaction of gene expression and treatment indicated an association of chemotherapy benefit and expression of RUNXl (p = 0.030, Interaction HR = 0.59, HR 95% CI (0.37, 0.95) and FAP (p= 0.065, Interaction HR = 0.66, HR 95% CI (0.42, 1.03). [00240] The association of gene expression and recurrence risk in surgery alone patients was examined for the 13 cancer-related genes. Multivariate Cox proportional hazards regression model allows estimation of recurrence risk adjusted for a specific distribution of clinical covariates. Recurrence risk estimates were produced from this multivariate model, adjusting for distribution of clinical covariates, differences in distribution in various study populations (if any), and baseline survival.

[00241] Table 8 presents the results of the univariate Cox proportional hazards regression models of gene expression on RFI. Figure 12 demonstrates the group risk (by Kaplan Meier curve) for Stage II colon cancer patients following surgery based on risk of recurrence at three years and recurrence score (including stromal and cell cycle group genes). Figure 13 demonstrates the risk profile plot (by Kaplan Meier curve) for risk of recurrence at five years (QUASAR- surgery only) and recurrence score (including stromal, cell cycle, and (for RS2) apoptosis genes).

[00242] In addition, the analyses combining the results from the four colon development studies and the QUASAR validation study were carried out to assess the performance of the 13 cancer-related genes across more than 3000 patients. Two different analysis methods were applied to combine the results across studies: (1) meta-analysis treating inter-study variation as random using the method of Paule and Mandel (1982) as implemented by DerSimonian and Kacker (2007); and (2) Cox proportional hazards regression model stratified by study, stage and treatment. Table 9 presents the results of these analyses. As can be observed, all but AXIN were shown to be associated with risk of recurrence in colon cancer (i.e. 95% CI did not include 1). (See, e.g., R. Paule, J. Mandel, Journal of Research of the National Bureau of Standards 87:377-385 (1982); R. der Simonian and R. Kacker, Cotemp. Clin Trials 28:105-144 (2007), both incorporated herein by reference.) EXAMPLE 4: ALTERNATIVE ALGORITHM-BASED ASSAY

[00243] Further analysis of data from the studies outlined in the Examples above suggested that incorporating additional genes into the Recurrence Score gene panel may yield improved performance. For example, BIK and EFNB2 were significantly associated with recurrence risk in both surgery alone and 5FU-treated patients. Statistical modeling was conducted to explore the strength of association between several multi-gene modules and recurrence of colon cancer. Table 10 and Figures 17-19 demonstrate comparative prognostic performance of selected multi-gene models.

Table 10: Multi-gene models based on standardized gene expression.

[00244] Based on the statistical modeling, it was determined that a multi-gene model using BGN and Ki-67, or BGN, Ki-67 and BIK, can provide minimal prognostic information to colon cancer patients. See Figure 18-19. However, a model consisting of ten prognostic genes (BGN, FAP, INHBA, EFNB2, GADD45B, Ki-67, MAD2L1, BIK, cMYC, MYB L2), plus reference genes ("RS2"), provided a highly accurate assessment of risk of recurrence in colon cancer. See Figure 20.

EXAMPLE 5: IDENTIFYING CO-EXPRESSED GENES AND GENE CLIQUES [00245] Gene cliques that co-express with the validated prognostic and predictive genes are set forth in Tables 4-6. These gene cliques were identified using the method described herein.

Materials and Methods:

[00246] Microarray data for colon tumor samples may be obtained internally, or derived from a public database, such as Gene Expression Omnibus (GEO). Microarray data was normalized and a pairwise Spearman correlation matrix computed for all array probes. Significant co-expressed probes across different studies was filtered out, and a graph built to compute probe cliques, map the probes to genes, and generate the gene cliques.

Download colon cancer microarray datasets

[00247] Five datasets from the Gene Expression Omnibus (GEO) database were used to compute the colon cliques. These datasets were identified as colon tumor expression experiments using the Affymetrix® HG-U133A microarray chip (Affymetrix Inc., Santa Clara, CA). Detailed information can be found at GEO website http://www.ncbi.nlm.nih.gov/geo/. Table 7 provides the accession number for the GEO datasets and the number of tumor samples in each datasets.

Array data normalization

[00248] The array data from GEO may be normalized using appropriate software, e.g.

Affymetrix MAS5.0, or an open source RMA software like bioconductor package. (See, e.g., http://rafalab.jhsph.edu/).

[00249] If the sample array data are of MAS5.0 type, they are normalized with the following steps:

1. Expression level is changed to "10" if the value is <10.

2. Expression level is then log transformed.

3. Median is computed on the log transformed values for the whole array probes. 4. Each probe value subtracts the median and the resulting value will be defined as normalized value

[00250] If the sample array data are of RMA type, they are normalized with the following steps:

1. Median is computed on the RMA generated values for the whole array probes.

2. Each probe value subtracts the median and the resulting value will be defined as normalized value

Array probe co-expression pair generation

[00251] The Spearman's rank correlation coefficient (r_s) was calculated for every unique pair of probes in the dataset (22283 probes resulting 248,254,903 unique pairs for each dataset). These pairs were then filtered by a significant threshold value T; any probe pair which has an r_s >= T was considered significant. Significant correlation pairs (had Spearman correlation values above threshold) were generated for each GEO dataset. For a given seeding gene probe, if the significant pairs involving the seeding probe or its directly connected probes existed across all five GEO datasets, they were placed in a graph and used to calculate maximal cliques.

Array probe clique generation

[00252] The Bron-Kerbosch algorithm was used to generate the maximal cliques from a graph of significant probe pairs generated from the above step. First, three "sets" of nodes were created. The first set, compsub, was the set to be extended or shrunk on traveling along a branch of the backtracking tree. The second set, candidates, was the set of all points that will be added to compsub. The third set, not, was the set of nodes already added to compsub. The recursive mechanism for generating cliques is as follows:

1. Selection of a candidate node.

2. Adding the selected candidate node to compsub.

3. Creating new sets candidates and not from the old sets by removing all nodes not connected to the selected candidate, keeping the old sets in tact.

4. Calling the extension operator to operate on the sets just formed.

5. Upon return, removal of the selected candidate from compsub and its addition to the old set not.

[00253] If after the extension operator, the candidates and not sets were empty, then the nodes on compsub were a clique and the mechanism starts over with a new candidate node. (See Figure 11.) Gene clique reporting

[00254] After the probe cliques had been computed, each probe in the cliques was mapped to genes as identified by Entrez Gene Symbol (Official Gene Symbol). Table 6 lists the report for the cliques associated with FAP, INHBA, Ki-67, HSPEl, MAD2L1, and RUNXl.

[00255] Certain probes have multiple mapping to Genes. They are listed as the same

AffyProbeID within a SeedingGene but have multiple ambiguous map to Official Genes (listed as CliquedGene column). Certain CliquedGenes are listed as "— " in Table 6. That means the AffyProbes do not map to any current Official Genes. The weight column list out the weight as we merged cliques. It is essentially is the number of clique evidence for coexpression with the seeding gene.

EXAMPLE 6: USE OF THRESHOLDING

[00256] Thresholding can be used to improve the reproducibility in recurrence score

(RS) and treatment score (TS) reporting by accounting for significant losses in precision as gene expression measurements approach the limit of quantitation (LOQ) of the assay. The LOQ of an assay represents the lowest concentration of RNA at which results can reliably be reported and have been estimated for each of the 18 colon cancer genes. [00257] As an example, Figure 26 shows the effects of diluting RNA concentration on

(non-normalized) gene expression (C_t) measurements of Ki-67. The variance in C_t measurement clearly increases as RNA concentration decreases. In fact, it may be shown that the log variance in C_t measurement is roughly proportional to the mean C_t measurement for a gene. As a consequence, the variability in RS and TS may be further reduced by truncating gene expression measurements at or near the LOQ, thereby reducing the potential for noise being introduced into RS and TS estimation.

EXAMPLE 7: CALCULATING GENE EXPRESSION: TUMOR REGION RATIOS

[00258] The clinical development studies in stage II/III colon cancer described above illustrated that genes which are frequently associated with stroma are correlated with increased risk of recurrence, whereas cell cycle genes are correlated with decreased risk of recurrence. This fact may account for the variability of RS/TS scores, and could be taken into account if the algorithm described herein considered the amount of stroma and luminal area, as well as localized gene expression in these regions. For example, an algorithm taking into account the ratios of stromal gene expression values per stroma area unit, and cell cycle gene expression values per epithelial area unit, would increase the precision and reproducibility of a recurrence risk prediction by decreasing heterogeneity within tumor blocks for a given patient.

[00259] A study was conducted to clarify the impact of variable tumor region areas and stromal/cell cycle gene expression on recurrence risk. RNA was extracted from different regions of colon tumors - the luminal part of the tumor and the tumor-associated stroma. Figure 14 shows that there are higher expression levels of the stromal genes in the tumor- associated stroma and higher expression levels of the cell cycle genes in the luminal part of the tumor. It is therefore likely that the stroma is contributing significantly to the stromal group score (SG or SGS) and the epithelia is significantly contributing to the cell cycle gene score (CCG or CCGS). Given these assumptions, the area of stroma within the sample contributes to the variability of the SG (within and between blocks) and therefore the score(s). Similarly, the area of epithelia within the sample analyzed could contribute to the variability of the CCG (within and between blocks) and therefore the score(s). [00260] Gene expression within tumor epithelia cells and stroma varies from patient to patient. For example, Figure 15 demonstrates that some patients may have higher levels of gene expression in their tumor-associated stroma for stromal genes than do other patients. Thus, some patients can have large amounts of stroma but low activity, whereas other patients can have smaller amounts of stroma but high activity. In addition, gene expression levels for the same patient can vary depending on the location of the tumor (e.g., within and between tumor blocks). This variability can impact reproducibility of recurrence and treatment scores for a patient. For example, Figure 16 demonstrates the variability, by tissue section of the same tumor block, of stromal group score (SG), cell cycle group score CCG), cell signaling group (CSG or GADD45B), and recurrence score (RS). This analysis was done on multiple sections from the same tumor block, and included data from 11 patients. [00261] Therefore, taking into account the area of the tumor-associated stroma and the area of the tumor-luminal regions in calculating the RS algorithm and in calculating the TS algorithm can enhance the reproducibility of the RS and TS, respectively, thus leading to greater accuracy of recurrence risk prediction.

[00262] For example, the expression level of stromal group genes can be provided as a ratio of the expression level of one or more stromal group genes to the tumor-associated stroma unit area ("sua") assayed. In another example, the expression level of cell cycle group genes can be provided as a ratio of the expression level of one or more cell cycle group genes to the tumor epithelial unit area ("cua") are assayed. The RS algorithm could be modified in the following form: RS = [(SG x sua coefficient) ± (CCG x sua coefficient)] + [(SG x cua coefficient) ± (CCG x cua coefficient)] ± (repeat analysis for other gene groups, e.g., CSG,

AG, and/or TFG). Similarly, the TS algorithm could be modified in the following form: TS =

[(SG x sua coefficient) ± (CCG x sua coefficient)] + [(SG x cua coefficient) ± (CCG x cua coefficient)] ± (repeat analysis for other gene groups, e.g., AG, TFG, and/or MG.)

[00263] In addition, the following exemplary algorithm provides a method to analyze and remove variability associated with gene expression in different portions of the block. For example, for cell cycle and stromal gene expression in different portions of a tumor block one could calculate: SGSij = SGi + SBij (Stromal gene group value for subject i block j is sum of a Gene effect and a Block effect) and CCGSij = CCGi + CCBij (Cell cycle gene group value for subject i block j is sum of a Gene effect and a Block effect).

[00264] SGS and CCGS are not correlated across subjects: SGS and CCGS variability is mostly from SG and CCG, the gene expression factor, and these are not correlated.

[00265] SGS and CCGS are correlated within subjects: There is a common effect underlying CCB and SB. Calculate: SGSrij = SGSij - SGSi = SBij - SBi.

[00266] CCGSrij = CCGSij - CCGSi. = CCBij - CCBi.

[00267] Correlation between SGSrij and CCGSrij can be thought of as a within subjects correlation pooled across subjects, i.e. an average within patient correlation. An informal approach to estimating p in (Yij, Xij ) ~ N((μyi, μxi), [σy, pyx // pyx, σx]).

Alternatively could assume Yij = αi + βXij + εij

[00268] If %Stroma correlates with the SGS within subject, it could provide a means of removing this source of variability in the RS and/or TS values.

EXAMPLE 8: STROMAL RISK ANALYSIS

Methods and Materials

[00269] A study involving 444 patients from a subset of the Cleveland Clinic

Foundation (CCF) cohort described in Example 1 was conducted to clarify how the amount of tumor-associated stroma ("stroma area") in a colon cancer tumor sample impacts the recurrence risk for stage II/III colon cancer patients ("Stromal Risk"). Specifically, a subset of the CCF cohort (cohort-sampling study design) involving all 148 recurrences from the CCF cohort and a random sample of approximately twice as many (i.e., 296) non-recurrences was used, resulting in 444 patients treated by resection of the colon. [00270] Inclusion criteria included: o Either stage II or stage III colon cancer patient. o Patient treated with colon resection (surgery) at CCF between the years of 1981 and 2000.

[00271] Exclusion criteria included: o No tumor block available from initial diagnosis in the CCF archive. o No tumor or very little tumor (<5% of the area occupied by invasive cancer cells compared to the area occupied by other epithelial elements, such as normal epithelium, or lymphatic) in block as assessed by examination of the H&E slide by the CCF and Genomic Health Pathologist. o Patients diagnosed with stage II or stage III signet ring colon cancer (WHO classification) o Insufficient RNA (<586 ng) for RT-PCR analysis. o Average non-normalized CT for the 5 reference genes > 35.

[00272] The full CCF cohort included a total of 886 FPE tumor tissue blocks. Of these,

108 were excluded due to failure to satisfy pathology and/or laboratory requirements described below. An additional 13 patients were excluded after the laboratory, pathology and clinical data were merged because of failure to satisfy all study inclusion and exclusion criteria, leaving 765 evaluable patients. The initial histological assessment by a Genomic Health pathologist was to evaluate the slide for the quantity of tumor and, where necessary, mark for manual micro dissection to enrich the tumor region. In this initial pathology review 8 cases were found to have insufficient tumor tissue (<5% tumor tissue) and thus failed the initial pathology review. The samples then underwent full histology review. Grade was captured by CCF and Genomic Health pathologists and each pathology read was analyzed separately (i.e. no attempt was made to create a 'combined' pathology score). An additional 11 cases failed this full pathology review due to the presence of a signet ring morphology comprising greater than 50% of the invasive component, lack of sufficient invasive tumor tissue (<5% cancer cells) or tissue type other than colon. Patient and sample disposition from the CCF study are summarized in Table 11.

Table 11 Patient Disposition from CCF Study

[00273] All 444 evaluable samples underwent both standard and digital pathology assessments. Using the 120-slide capacity ScanScope XT system, automated scanning of all study H&E slides were conducted at 2OX scanning magnification with autopopulation of patient identification fields with barcode data using the Spectrum information management system. The 2OX scanning magnification was selected because this magnification gives superior optimization of image quality and scanning speed.

[00274] Digital H&E scans were obtained from the Aperio® Digital Pathology

System. Two different software systems - the Aperio® Genie Digital Pathology Image Analysis software and the Definiens® Digital Pathology Image Analysis software - were used to generate digital H&E measurements. The Definiens image analysis software, based on the Definiens Cognition Network Technology®, examines pixels in context and builds up a picture iteratively, recognizing groups of pixels as objects.

[00275] The pathologist and assistant trained the image analysis applications to detect regions of interest (e.g., mucin, tumor glands and tumor stroma) using previously captured digital images of the entire enriched tumor portion. These training slides were representative of the slides to be assessed by the Aperio system. Several variations of the two image analysis algorithms were developed for low and high grade carcinomas and mucinous carcinomas. These were developed by identification of regions of interest, and then having the programs "learn" from the training slides. The resulting algorithms were applied to the entire patient cohort, analyzing the enriched tumor portions of the patient samples. The patient samples were batched into three digital study sets (i.e., low grade, high grade and mucinous carcinomas) as determined by the GHI pathologist and all images were processed using batch processing.

Findings and Statistical Analysis

[00276] The surface area of tumor-associated stroma varies from patient to patient.

For example, Figure 21 provides a variability plot for natural logarithm of stroma area, as measured by the Aperio digital pathology system, for the 444 patients under study, stratified by recurrence-free interval status.

[00277] Statistical analyses were performed to determine if there was a significant relationship between stroma area and recurrence-free interval (RFI). Specifically, we compared the (reduced) Weighted Cox Proportional Hazards model for RFI based on the main effect for tumor stage (Stage II and Stage III), versus the (full) Weighted Cox Proportional Hazards model for RFI based on the main effects of tumor stage and stroma area as measured by the Aperio digital image analysis system. Weighted Pseudo Partial Likelihood approach was used to accommodate the use of a case-cohort sampling study design. A WaId test for the hypothesis that the hazard ratio for stroma area is 1 versus the 2- sided alternative hypothesis that the hazard ratio is not 1 was performed. The resulting WaId χ² = 15.64 with 1 degree of freedom resulting in a 2-sided p-value < 0.001, indicating that stroma area is prognostic of disease recurrence (beyond tumor stage alone) in colon cancer patients treated with colon resection. The resulting standardized hazard ratio for stroma area is 1.45, indicating that there is a 45% increase in the relative risk for disease recurrence for each standard deviation increase in stroma area.

Table 12: Proportional Hazard Regression for Recurrence-Free Interval: Stage and Stroma Area Alone

[00278] In addition to testing if stroma area is prognostic of disease recurrence, statistical analyses were performed to determine if stroma area provides additional prognostic information beyond both stage and Recurrence Score. Specifically, we compared the (reduced) Weighted Cox Proportional Hazards model for RFI based on the main effect for stage (Stage II and Stage III) and Recurrence Score, versus the (full) Weighted Cox Proportional Hazards model for RFI based on the main effects of tumor stage, Recurrence Score and stroma area as measured by the Aperio digital image analysis system. A WaId test for the hypothesis that the hazard ratio for stroma area is 1 versus the 2- sided alternative hypothesis that the hazard ratio is not 1 was performed. The resulting WaId ratio χ² = 13.17 with 1 degree of freedom resulting in a 2-sided p-value < 0.001, indicating that stroma area is prognostic of disease recurrence beyond tumor stage and Recurrence Score. The resulting standardized hazard ratio for stroma area is 1.41, indicating that there is a 41% increase in the relative risk for disease recurrence for each standard deviation increase in stroma area.

Table 13: Proportional Hazard Regression for Recurrence-Free Interval: Stage, Stroma Area and Recurrence Score

[00279] Similar analyses were performed to test if stroma area provides additional prognostic information beyond both stage and RS2. Specifically, we compared the (reduced) Weighted Cox Proportional Hazards model for RFI based on the main effect for stage (Stage II and Stage III) and RS2, versus the (full) Weighted Cox Proportional Hazards model for RFI based on the main effects of tumor stage, RS2 and stroma area as measured by the Aperio digital image analysis system. A WaId test for the hypothesis that the hazard ratio for stroma area is 1 versus the 2-sided alternative hypothesis that the hazard ratio is not 1 was performed. The resulting WaId ratio χ² = 14.86 with 1 degree of freedom resulting in a 2- sided p- value < 0.001, indicating that stroma area is prognostic of disease recurrence beyond tumor stage and RS2. The resulting standardized hazard ratio for stroma area is 1.44, indicating that there is a 44% increase in the relative risk for disease recurrence for each standard deviation increase in stroma area.

Table 14: Proportional Hazard Regression for Recurrence-Free Interval: Stage, Stroma Area and RS2

[00280] For analysis purposes, stroma area can be stratified into low and high Stroma

Risk Groups. Specifically, we define low risk (stroma score < 0) and high risk (stroma score > 0) where stroma score = (stroma area - mean)/standard deviation. Kaplan-Meier Plots for Stage II and Stage III patients stratified by Stroma Risk Group, provided in Figures 22 and 23 respectively, clearly show separation between risk groups (Logrank p- value < 0.01). Similarly, Kaplan-Meier Plots for Stage II and Stage III patients stratified by both Stroma Risk Group and Recurrence Score Risk Group, provided in Figures 24 and 25 respectively, show even greater separation between risk groups (Logrank p-value < 0.01).

Conclusion

[00281] These analyses show that stroma area is independently prognostic of disease recurrence in stage II and stage III patients and that RS, stromal area, and nodal status all provide important prognostic information in stage II and III colon cancer. The discovery that it is surface area of tumor-associated stroma that is most strongly associated with risk of recurrence, rather that proportional measurements of tumor regions, was an unexpected result of this study.

Oi

σ>

OO

CD

N) O

N)

N) N)

N) CO

N)

N) Ol

N)

N)

N) OO

N) CD

o

co

N)

GO

Oi

OO

CD

O

N)

-^ -^

Oi

-^

^l

OO

CD

Oi O

Oi

Oi N)

Oi

Oi

Oi Oi

Oi

Oi

Oi 00

Oi CD

o

σ>

N)

Table 7: Datasets used for ene cli ue anal sis of ro nostic and redictive genes

Table 8. Association of gene expression and risk of recurrence in surgery alone patients from the QUASAR study

Table 9. Results of the meta analysis and stratified Cox models

Claims

CLAIMSWhat is claimed is:

1. A method for determining a likelihood of cancer recurrence after surgery in a patient with colorectal cancer, comprising: measuring expression levels of prognostic genes in a tumor sample obtained from the patient; calculating a recurrence score (RS) for the patient using the measured gene expression levels, wherein the RS is calculated by assigning the measured expression levels to gene subsets of an RS algorithm, wherein the gene subsets comprise at least one gene each from a stromal group and a cell cycle group; and generating a report based on the RS.

2. The method of claim 1, wherein the expression level of each gene subset is weighted according to a contribution of that gene subset to cancer recurrence.

3. The method of claim 1, wherein the expression level of each gene subset is further calculated as a ratio of expression level to tumor region, wherein the ratio comprises tumor-associated stroma unit area or tumor epithelia unit area.

4. The method of claim 1, further comprising: measuring a stromal area of the tumor sample to obtain a Stromal Risk Score, wherein increased stromal area is positively correlated to increased risk of recurrence of cancer for said patient; and wherein the report is further based on the Stromal Risk Score.

5. The method of claim 1, wherein the stromal group comprises BGN, FAP, INHBA, and genes that co-express with BGN, FAP, or INHBA.

6. The method of claim 1, wherein the cell cycle group comprises MYB L2, Ki- 67, cMYC, MAD2L1, and genes that co-express with MYB L2, Ki-67, cMYC, or MAD2L1.

7. The method of claim 1, wherein said gene subsets further comprise a cell signaling group including GADD45B, and genes that co-express with GADD45B.

8. The method of claim 1, wherein said gene subsets further comprise an apoptosis group including 9 BIK and genes that co-express with BIK.

9. The method of claim 1, wherein said gene subsets further comprise an angiogenesis group including

EFNB2 and genes that co-express with EFNB2.

10. A method of determining a likelihood that a patient with colorectal cancer will have a beneficial response to chemotherapy, comprising: measuring expression levels of predictive genes in a tumor sample obtained from the patient; calculating a treatment score (TS) for the patient using the measured gene expression levels, wherein the TS is calculated by assigning the measured expression levels to gene subsets of a TS algorithm, wherein the gene subsets comprise at least one gene each from a MSI group, an apoptosis group, and a stromal group; and creating a report based on the TS.

11. The method of claim 10, wherein the expression level for each gene subset is weighted according to a contribution of the gene subset to response to chemotherapy.

12. The method of claim 10, wherein the MSI group comprises AXIN2 and genes that co-express with AXIN2.

13. The method of claim 10, wherein the apoptosis group comprises BIK and genes that co-express with BIK.

14. The method of claim 10, wherein the stromal group comprises EFNB2 and genes that co-express with EFNB2.

15. The method of claim 10, wherein said gene subsets further comprise a transcription factor group including

RUNXl and genes that co-express with RUNXl.

16. The method of claim 10, wherein said gene subsets further comprise a cell cycle group including

MAD2L1 and HSPEl, and genes that co-express with MAD2L1 and HSPEl.

17. The method of claim 10, wherein at least one gene from said gene subsets may be replaced by a substitute gene from the group consisting of RANB P2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2, and genes that co- express with RANBP2, BUBl, TOP2A, C20_ORF1, CENPF, STK15, AURKB, HIFlA, UBE2C, and MSH2.

18. A method to use a threshold value for expression values used in an algorithm- based gene expression analysis, comprising: measuring an expression level of a gene in a tissue section obtained from a patient; and comparing the measured expression level to a threshold value for said gene; wherein if the threshold value is less than the expression level of said gene, the expression value is used in an expression algorithm, and wherein if the expression level of said gene is greater than or equal to the threshold value, the expression level is used in an expression algorithm.

19. The method of claim 18, wherein the threshold value is based on a C_t value.

20. The method of claim 18, wherein the threshold value is one or more from those listed in Table 3.

21. A gene expression analysis method to identify a gene that is co-expressed with a target gene comprising: normalizing microarray gene expression data for cancer tumor samples based on array probes; calculating a correlation coefficient based on gene expression levels for every unique pair of array probes; determining significant probe pairs, wherein significant probe pairs are a target gene probe and an array probe with a correlation co-efficient greater than a significant threshold value; mapping the target gene to its corresponding target gene probe, selecting a candidate probe set, wherein each candidate probe is part of a significant probe pair; and identifying a gene associated with each candidate probe; wherein said gene associated with each candidate probe is a co-expressed gene.

22. A method of assessing gene expression, the method comprising: measuring a normalized expression level of a gene in a cancer tumor sample obtained from a patient; calculating a ratio of normalized expression of the gene to a tissue unit area in the colorectal sample, wherein the tissue unit area is a tumor-associated stroma unit area or a tumor epithelial unit area; and calculating a recurrence score (RS) or a treatment score (TS) for the patient using the ratio.

23. The method of claim 22, wherein the gene is a stromal group gene.

24. The method of claim 23, wherein the tissue unit area is a tumor-associated stroma unit area.

25. The method of claim 22, wherein the gene is a cell cycle group gene.

26. The method of claim 22, wherein the gene is an apoptosis group gene.

27. The method of claim 22, wherein the tissue unit area is a tumor epithelial unit area unit area.

28. A method of determining a prognosis for a cancer patient, comprising; measuring a stromal area of a tumor sample obtained from the cancer patient to obtain a Stromal Risk Score, wherein increased stromal area of the tumor sample is positively correlated with an increased risk of recurrence of cancer for said cancer patient, and generating a report based on the Stromal Risk Score.

29. The method of claim 28, wherein the tumor sample is a colorectal cancer tumor.