WO2012009382A2 - Molecular indicators of bladder cancer prognosis and prediction of treatment response - Google Patents

Abstract

Disclosed are biomarkers, methods and assay systems for the identification of bladder cancer patients who are predicted to benefit, or not benefit from the therapeutic administration of neoadjuvant therapy. Particularly, the invention provides a diagnostic paradigm based on each of these tests and combinations of these tests to select bladder cancer patients who will or will not benefit from neoadjuvant chemotherapy, as well as a diagnostic paradigm to identify bladder cancer patients likely to suffer recurrence of cancer following cystectomy.

Description

MOLECULAR INDICATORS OF BLADDER CANCER PROGNOSIS AND PREDICTION OF TREATMENT RESPONSE

CROSS REFERENCE TO RELATED APPLICATION

This application claims the benefit of priority under 35 U.S.C. § 119(e) to U.S. Provisional Patent Application Serial No. 61/363,338 filed July 12, 2010, which is incorporated herein by reference.

GOVERNMENT INTEREST

This invention was made in part with United States Government support under Grant No. CA 104106, awarded by The National Institutes of Health. The United States Government has certain rights in the invention.

TECHNICAL FIELD

The invention relates to genes the expression of which is important in the treatment and prognosis of bladder cancer.

BACKGROUND OF INVENTION

Bladder cancer is the fifth most common cancer in the United States, with almost 60,000 cases diagnosed each year, and more than 12,000 patients dying from the disease each year. Men, Caucasians and smokers have twice the risk of bladder cancer than the general population. When diagnosed and treated in a localized stage, superficial bladder cancer is very treatable, with a five-year cancer-specific survival rate approaching 95%. Up to eighty percent of all bladder cancers are diagnosed as superficial bladder cancers. Once positively diagnosed and biopsied, many superficial bladder cancers (SBC) can be surgically removed by a procedure known as a transurethral resection (TUR), although bladder cancer can come back after this surgery, so repeat TURs are sometimes needed.

Unfortunately, nearly a third of patients with bladder cancer present with muscle- invasive disease (stage >T2), which when clinically confined to the bladder (stage N0M0) is treated by radical cystectomy (or bladder removal) and pelvic lymphadenectomy (removal of the surrounding lymph nodes) or chemoradiation. Despite this treatment, up to 50% of such patients develop metastatic disease, which is nearly always fatal. For such invasive bladder cancer, cystectomy is necessary to stop its progression to metastatic bladder cancer. Pathological assessment of the regional lymph nodes removed during cystectomy leads to the classification of nodal involvement (pNO denoting the absence of regional lymph node metastasis, and pNl-3 denoting higher levels of lymph node(s) involvement).

In instances of lymph node involvement, neoadjuvant therapy has been proposed before cystectomy, and evidence from randomized controlled trials (HB Grossman, RB Natale and CM Tangen et al. , Neoadjuvant chemotherapy plus cystectomy compared with cystectomy alone for locally advanced bladder cancer, N Engl J Med 349 (2003), pp. 859- 866; RR Hall, Updated results of a randomized trial of neoadjuvant cisplatin (C) methotrexate (M) and vinblastine (V) chemotherapy for muscle-invasive bladder cancer, Proc Am Soc Clin Oncol 21 (2002), p. 178a (abstr 710)) and a meta-analysis thereof (Advanced Bladder Cancer Meta-analysis Collaboration, Neoadjuvant chemotherapy for invasive bladder cancer, Cochrane Database Syst Rev 2 (2005) CD005246) support the use of neoadjuvant, platinum-based combination chemotherapy in improving survival. Despite such evidence, neoadjuvant chemotherapy has not been widely adopted. In a series that encompassed about 60% of patients in the USA diagnosed with muscle- invasive bladder cancer, less than 2% of patients received neoadjuvant chemotherapy. Recent reviews have identified barriers to its implementation that included concerns for delay of surgery and risk of disease progression.

A study of cancer recurrence after cystectomy in more than 9000 patients

(untreated by chemotherapy or radiotherapy) shows that about 80% of cases with pathological node-positive disease recurred, whereas only about 30%> of extravesical, pathological node-negative or about 20%> of organ-confined cases recurred. By use of the 10% disease-free survival benefit at 5 years from neoadjuvant chemotherapy for bladder cancer calculated in the clinical trial meta-analysis, only two of 100 node-negative, organ- confined patients would benefit from neoadjuvant chemotherapy. By contrast, for patients with nodal positive disease (pNl-3), about eight of 100 patients could benefit from this approach. Therefore, there is a need for a reliable method of predicting pathological nodal status before surgery in order to identify high-risk patients with a greater likelihood of benefiting from neoadjuvant therapy.

SUMMARY OF INVENTION

The present invention provides a gene expression model (GEM), evaluable on the primary tumor that predicts pNl-3 at cystectomy in clinically node-negative patients. Use of this GEM allows selection of patients most likely to benefit from neoadjuvant therapy while avoiding overtreatment and delay. One embodiment of the invention provides to a method that includes detecting a level of gene expression of a marker gene or plurality of marker genes in a sample of bladder tumor cells from a patient. The marker gene(s) are selected from a marker gene having at least 95% sequence identity with a sequence selected from the genes listed in Table A, or homologs or variants thereof.

Table A. Identified Markers

In a related embodiment, the detection of the level of expression of the marker gene(s) may be conducted by detection of polypeptides encoded by the marker genes of 1- 21 above, and/or fragments of polypeptides of the marker genes of 1-21 above, and/or a polynucleotide which is fully complementary to at least a portion of a marker gene of 1-21 above. In each of these embodiments, the detection of an elevated gene expression of the plurality of markers is indicative of a bladder cancer patient predicted to benefit from neoadjuvant chemotherapy.

In a preferred embodiment, the genes detected in these methods share 100% sequence identity with the corresponding marker genes.

In each of these embodiments, the levels of at least one of the plurality of markers may be determined and compared to a standard level or reference range of gene expression, that may be determined according to a statistical procedure for risk prediction.

In one embodiment of this method, the presence of the polypeptides may be detected using a reagent that specifically binds to the polypeptide, or a fragment thereof. In a preferred embodiment, the reagent is selected from the group consisting of an antibody, an antibody derivative, and an antibody fragment.

In another embodiment of this method, the presence of the marker is determined by obtaining R A from the bladder cancer tissue sample; generating cDNA from the R A; amplifying the cDNA with probes or primers for marker genes; obtaining from the amplified cDNA the expression levels of the genes or gene expression products in the sample.

These methods may include comparing the expression level of the marker gene or plurality of marker genes, in the tumor cell sample to a control level of the marker gene(s) including: a control level of the marker gene that has been correlated with beneficial response to the administration of neoadjuvant chemotherapy, and/or a control level of the bio marker that has been correlated with lack of beneficial response to neoadjuvant chemotherapy. In these embodiments, the patient is predicted to respond to the

administration of neoadjuvant chemotherapy if the expression level of the marker gene in the patient's bladder tumor cells is statistically similar to, or greater than, the control level of expression of the marker gene that has been correlated with sensitivity to the administration of neoadjuvant chemotherapy. Alternatively, the patient is predicted to not respond to neoadjuvant chemotherapy, if the level of the marker gene in the patient's bladder tumor cells is statistically less than the control level of the marker gene that has been correlated with beneficial response to the administration of neoadjuvant

chemotherapy.

Additionally, or as an alternative, these embodiments may include comparing the expression level of the marker gene or plurality of marker genes, in the tumor cell sample to a level of the marker gene(s) in a second patient predicted to not respond to the administration of neoadjuvant chemotherapy. In this embodiment, the patient is predicted to respond to the administration of neoadjuvant chemotherapy, if the expression level of the marker gene in the patient's bladder tumor cells is greater than the level of expression of the marker gene(s) in the second patient. Alternatively, the patient is predicted to not respond to the administration of neoadjuvant chemotherapy, if the level of the marker gene in the patient's bladder tumor cells is less than or equal to the level of expression of the marker gene(s) in the second patient.

A preferred embodiment of these methods of determining if a patient is predicted to respond to the administration of neoadjuvant chemotherapy includes detecting a level of gene expression of a gene having at least 95% sequence identity with each of TOX3, SLC 1 1A2, FAM36A, LIMCH1 , RAB 15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGA10, MT1E, MAP4K4, SLC 16A1 , BST2, MMP14, IFI27, NCLN, HLA-G, RRBP1 , ICAM1 , or homo logs or variants thereof, in a sample of bladder tumor cells from a patient. In a preferred embodiment, the genes detected preferably share 100% sequence identity with the corresponding marker genes. The method may also be conducted by detecting a level of polypeptides encoded by the genes, and/or fragments of polypeptides, and/or a polynucleotide that is fully complementary to the genes. In this embodiment, an elevated level of expression of the plurality of markers is indicative of whether a patient that will respond to treatment with neoadjuvant chemotherapy.

Another embodiment of the invention is a method for identifying a bladder cancer patient predicted to suffer recurrence of the cancer following cystectomy by detecting in a sample of bladder tumor cells from the patient, a level of gene expression of a marker gene or plurality of marker genes selected from the group consisting of a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC 1 1A2, FAM36A, LIMCH1 , RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGA10, MT1E, MAP4K4, SLC 16A1 , BST2, MMP14, IFI27, NCLN, HLA-G, RRBP1 , ICAM1 , or homo logs or variants thereof; or polypeptides encoded by these marker genes; or fragments of polypeptides of ii) or a

polynucleotide that is fully complementary to at least a portion of these markers, wherein the expression of the plurality of markers is indicative of whether the cancer is likely to recur in the patient following cystectomy.

Another embodiment of the invention is a method of monitoring the progression of bladder cancer in a subject by measuring the expression level of a plurality of marker genes in a first biological sample obtained from the subject, measuring the level of the plurality of markers in a second biological sample obtained from the subject, and comparing the level of the marker measured in the first sample with the level of the marker measured in the second sample. In this embodiment, the plurality of marker gene(s) are selected from a marker gene having at least 95% sequence identity with a sequence selected from TOX3, SLCl 1A2, FAM36A, LIMCHl, RAB15, AVL9,

PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBPl, ICAMl or homologs or variants thereof. Preferably, the second biological sample is obtained from the subject at a time later than the first biological sample is obtained. Alternatively, the first biological sample and the second biological sample are obtained from the subject more than once, over a range of times.

In a related embodiment, the detection of the level of expression of the marker gene(s) may be conducted by detection of polypeptides encoded by the marker genes, and/or fragments of polypeptides of the marker genes, and/or a polynucleotide which is fully complementary to at least a portion of the marker genes. In a preferred embodiment, the genes detected in these methods share 100% sequence identity with the corresponding marker genes.

Another embodiment of the present invention is an assay system for predicting bladder cancer patient response or outcome to neoadjuvant chemotherapy. The assay system includes a means to detect the expression of a marker gene or plurality of marker genes having at least 95% sequence identity with a sequences selected from TOX3, SLCl 1A2, FAM36A, LIMCHl, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBPl, ICAMl, or homologs or variants thereof.

In one embodiment of the assay system, the means to detect includes a nucleic acid probe having at least 10 to 50 contiguous nucleic acids of the marker gene(s), or complementary nucleic acid sequences thereof. In another embodiment of the assay system, the means to detect includes binding ligands that specifically detect polypeptides encoded by the marker genes. These binding ligands may include antibodies or binding fragments thereof. The nucleic acid probes and/or binding ligands are preferably disposed on an assay surface, which may include a chip, array, or fluidity card.

The assay system preferably includes a control selected from information containing a predetermined control level of the marker gene that has been correlated with response or outcome to neoadjuvant chemotherapy, and/or information containing a predetermined control level of the marker gene that has been correlated with a lack of response or outcome to neoadjuvant chemotherapy.

In any one of the embodiments of the present invention, the step of detecting can include, but is not limited to, using a nucleotide probe that hybridizes to at least one of the marker gene(s). In one aspect, the probe may be a chimeric probe (e.g., that hybridizes to more than one of the biomarker genes). In another aspect, the step of detecting can include detecting the number of copies of the biomarkers genes per tumor cell in one or more tumor cells in the sample, and/or detecting marker gene amplification per tumor cell in one or more tumor cells in the sample. In a specific embodiment, the step of detecting gene expression is performed by TaqMan® Gene Signature Array, as described in U.S. Patent Nos. 6,514,750 and 6,942,837 and 7,211,443 and 7,235,406, each ofwhich is incorporated by reference in its entirety.

This Summary of the Invention is neither intended nor should it be construed as being representative of the full extent and scope of the present invention. Moreover, references made herein to "the present invention," or aspects thereof, should be understood to mean certain embodiments of the present invention and should not necessarily be construed as limiting all embodiments to a particular description. The present invention is set forth in various levels of detail in the Summary of the Invention as well as in the attached drawings and the Description of Embodiments and no limitation as to the scope of the present invention is intended by either the inclusion or non-inclusion of elements, components, etc. in this Summary of the Invention. Additional aspects of the present invention will become more readily apparent from the Description of Embodiments, particularly when taken together with the drawings.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 shows a hypothetical example illustrating how a test sample is analyzed in reference to training samples with positive correlation.

Figure 2 shows the derivation of the WNN algorithm in detail. As shown, Bayes' equation serves as the basis of the derivation of this classifier.

Figure 3 shows the construction of the post-test probabilities used to calculate the probability of nodal involvement in a given stratum from the WNN classifier score.

Figure 4 shows the development of a GEM predictor of pathological nodal status at cystectomy.

Figure 5 shows the risk cutoff development for 20-gene model with training cohorts. Figure 6 shows the performance of the 20-gene model on a prospectively collected independent test cohort.

DESCRIPTION OF EMBODIMENTS

The present invention is directed to methods that identify high-risk bladder cancer patients, who can then be administered additional, appropriate therapy while avoiding the overtreatment of low-risk bladder cancer patients. The inventors have developed a test that can predict a powerful determinant of prognosis after cystectomy: node-positive disease, and have shown that such molecular intelligence, for which no other molecular marker exists, provides a technique that allows more effective and frequent use of neoadjuvant therapy, particularly neoadjuvant chemotherapy.

According to one definition, a biological marker is "a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacological responses to therapeutic interventions." NIH Biomarker Definitions Working Group (1998). Biological markers can also include patterns or ensembles of characteristics indicative of particular biological processes ("panel of markers"). The marker measurement can be increased or decreased to indicate a particular biological event or process. In addition, if a marker measurement typically changes in the absence of a particular biological process, a constant measurement can indicate occurrence of that process.

Marker measurements may be of the absolute values (e.g., the molar concentration of a molecule in a biological sample) or relative values (e.g., the relative concentration of two molecules in a biological sample). The quotient or product of two or more measurements also may be used as a marker. For example, some physicians use the total blood cholesterol as a marker of the risk of developing coronary artery disease, while others use the ratio of total cholesterol to HDL cholesterol.

In the invention, the markers are primarily used for stratification purposes (e.g., to group patients into any number of "subsets" for evaluation) and prognostic purposes. However they may also be used for diagnostic, therapeutic, and drug screening as well as other purposes described herein, including evaluating the effectiveness of a bladder cancer therapeutic.

The practice of the invention employs, unless otherwise indicated, conventional methods of analytical biochemistry, microbiology, molecular biology and recombinant DNA generally known techniques within the skill of the art. Such techniques are explained fully in the literature. (See, e.g., Sambrook et al. Molecular Cloning: A Laboratory Manual. 3rd, ed., Cold Spring Harbor Laboratory, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 2000; DNA Cloning: A Practical Approach, Vol. I & II (Glover, ed.); Oligonucleotide Synthesis (Gait, ed., Current Edition); Nucleic Acid Hybridization (Hames & Higgins, eds., Current Edition); Transcription and

Translation (Hames & Higgins, eds., Current Edition); CRC Handbook of Parvoviruses, Vol. I & II (Tijessen, ed.); Fundamental Virology, 2nd Edition, Vol. I & II (Fields and Knipe, eds.)).

The terminology used herein is for describing particular embodiments and is not intended to be limiting. As used herein, the singular forms "a," "and" and "the" include plural referents unless the content and context clearly dictate otherwise. Thus, for example, a reference to "a marker" includes a combination of two or more such markers. Unless defined otherwise, all scientific and technical terms are to be understood as having the same meaning as commonly used in the art to which they pertain. For the purposes of the present invention, the following terms are defined below.

As used herein, the term "marker" includes polypeptide markers and

polynucleotide markers. For clarity of disclosure, aspects of the invention will be described with respect to "polypeptide markers" and "polynucleotide markers." However, statements made herein with respect to "polypeptide markers" are intended to apply to other polypeptides of the invention. Likewise, statements made herein with respect to "polynucleotide" markers are intended to apply to other polynucleotides of the invention, respectively. Thus, for example, a polynucleotide described as encoding a "polypeptide marker" is intended to include a polynucleotide that encodes: a polypeptide marker, a polypeptide that has substantial sequence identity to a polypeptide marker, modified polypeptide markers, fragments of a polypeptide marker, precursors of a polypeptide marker and successors of a polypeptide marker, and molecules that comprise a polypeptide marker, homologous polypeptide, a modified polypeptide marker or a fragment, precursor or successor of a polypeptide marker (e.g., a fusion protein).

As used herein, the term "polypeptide" refers to a polymer of amino acid residues that has at least 5 contiguous amino acid residues, e.g., 5, 6, 7, 8, 9, 10, 11 or 12 or more amino acids long, including each integer up to the full length of the polypeptide. A polypeptide may be composed of two or more polypeptide chains. A polypeptide includes a protein, a peptide, an oligopeptide, and an amino acid. A polypeptide can be linear or branched. A polypeptide can comprise modified amino acid residues, amino acid analogs or non-naturally occurring amino acid residues and can be interrupted by non-amino acid residues. Included within the definition are amino acid polymers that have been modified, whether naturally or by intervention, e.g., formation of a disulfide bond, glycosylation, lipidation, methylation, acetylation, phosphorylation, or by manipulation, such as conjugation with a labeling component. Also included are antibodies produced by a subject in response to overexpressed polypeptide markers.

As used herein, a "fragment" of a polypeptide refers to a single amino acid or a plurality of amino acid residues comprising an amino acid sequence that has at least 5 contiguous amino acid residues, at least 10 contiguous amino acid residues, at least 20 contiguous amino acid residues or at least 30 contiguous amino acid residues of a sequence of the polypeptide. As used herein, a "fragment" of polynucleotide refers to a single nucleic acid or to a polymer of nucleic acid residues comprising a nucleic acid sequence that has at least 15 contiguous nucleic acid residues, at least 30 contiguous nucleic acid residues, at least 60 contiguous nucleic acid residues, or at least 90% of a sequence of the polynucleotide. In some embodiment, the fragment is an antigenic fragment, and the size of the fragment will depend upon factors such as whether the epitope recognized by an antibody is a linear epitope or a conformational epitope. Thus, some antigenic fragments will consist of longer segments while others will consist of shorter segments, (e.g. 5, 6, 7, 8, 9, 10, 11 or 12 or more amino acids long, including each integer up to the full length of the polypeptide). Those skilled in the art are well versed in methods for selecting antigenic fragments of proteins.

In some embodiments, a polypeptide marker is a member of a biological pathway. As used herein, the term "precursor" or "successor" refers to molecules that precede or follow the polypeptide marker or polynucleotide marker in the biological pathway. Thus, once a polypeptide marker or polynucleotide marker is identified as a member of one or more biological pathways, the present invention can include additional precursor or successor members of the biological pathway. Such identification of biological pathways and their members is within the skill of one in the art.

As used herein, the term "polynucleotide" refers to a single nucleotide or a polymer of nucleic acid residues of any length. The polynucleotide may contain deoxyribonucleotides, ribonucleotides, and/or their analogs and may be double-stranded or single stranded. A polynucleotide can comprise modified nucleic acids (e.g., methylated), nucleic acid analogs or non-naturally occurring nucleic acids and can be interrupted by non-nucleic acid residues. For example a polynucleotide includes a gene, a gene fragment, cDNA, isolated DNA, mR A, tRNA, rR A, isolated R A of any sequence, recombinant polynucleotides, primers, probes, plasmids, and vectors. Included within the definition are nucleic acid polymers that have been modified, whether naturally or by intervention.

As used herein, a component (e.g., a marker) is referred to as "differentially expressed" in one sample as compared to another sample when the method used for detecting the component provides a different level or activity when applied to the two samples. A component is referred to as "increased" in the first sample if the method for detecting the component indicates that the level or activity of the component is higher in the first sample than in the second sample (or if the component is detectable in the first sample but not in the second sample). Conversely, a component is referred to as

"decreased" in the first sample if the method for detecting the component indicates that the level or activity of the component is lower in the first sample than in the second sample (or if the component is detectable in the second sample but not in the first sample). In particular, marker is referred to as "increased" or "decreased" in a sample (or set of samples) obtained from a bladder cancer subject (or a subject who is suspected of having bladder cancer, or is at risk of developing bladder cancer) if the level or activity of the marker is higher or lower, respectively, compared to the level of the marker in a sample (or set of samples) obtained from a non-bladder cancer subject, or a reference value or range.

The markers identified as being indicative of the value and patient response to neoadjuvant therapy and prognosis for survival and recurrence in bladder cancer are of significant biologic interest. Neoadjuvant chemotherapy before cystectomy confers a survival benefit in bladder cancer, but it has not been widely adopted since most patients do not benefit and it is not presently possible to predict those patients that do. Since the most important predictor of recurrence of cancer after cystectomy is pathologically positive nodes, the present inventors developed a gene expression model (GEM) to predict the pathological node status in primary tumor tissue from three independent cohorts of patients who were clinically node negative. From a subset of transcripts detected faithfully by microarrays from both paired frozen and formalin- fixed tissues (32 pairs), a GEM assay was developed, including cutoffs that identified patient strata with raised risk of nodal involvement by use of two separate training cohorts (90 and 66 patients). These assays were then assessed to predict node-positive disease in tissues from a phase 3 trial cohort (AUO-AB-05/95; 185 patients). A 20-gene GEM was developed for prediction of nodal disease at cystectomy. The cutoff system identified patients with high relative risk and low relative risk of node-positive disease. Multivariate logistic regression showed the GEM predictor was independent of age, sex, pathological stage, and lymphovascular space invasion. Use of these assays to select patients for neoadjuvant chemotherapy on the basis of risk of node-positive disease may benefit high-risk patients while sparing other patients toxic effects and delay to cystectomy.

In addition to the discovery of biomarkers that can be used individually or in any combination in assays and kits for the diagnosis of, prognosis of, or other evaluation or study of bladder cancer, the biomarkers not previously recognized to play a role in the disease process of bladder cancer can now be studied in more detail and/or be used as targets for the discovery of other modulators of disease or therapeutic agents. Table A provides polynucleotide markers that were found at significantly different levels in debulked samples obtained from patients with bladder cancer and that were found significantly correlated with overall survival of these patients.

All information associated with the publicly-available identifiers and accession numbers in any of the tables described herein, including the nucleic acid sequences of the associated genes, is incorporated herein by reference in its entirety. Given the name of the protein (also referred to herein as the "full protein"; indicated as "Protein"), other peptide fragments of such measured proteins may be obtained (by whatever means), and such other peptide fragments are included within the scope of the invention. The methods of the present invention may be used to evaluate fragments of the listed molecules as well as molecules that contain an entire listed molecule, or at least a significant portion thereof (e.g., measured unique epitope), and modified versions of the markers. Accordingly, such fragments, larger molecules and modified versions are included within the scope of the invention.

Homo logs and alleles of the polypeptide markers of the invention can be identified by conventional techniques. As used herein, a homolog to a polypeptide is a polypeptide from a human or other animal that has a high degree of structural similarity to the identified polypeptides. Identification of human and other organism homo logs of polypeptide markers identified herein will be familiar to those of skill in the art. In general, nucleic acid hybridization is a suitable method for identification of homologous sequences of another species (e.g., human, cow, sheep), which correspond to a known sequence. Standard nucleic acid hybridization procedures can be used to identify related nucleic acid sequences of selected percent identity. For example, one can construct a library of cDNAs reverse transcribed from the mRNA of a selected tissue (e.g., colon) and use the nucleic acids that encode polypeptides identified herein to screen the library for related nucleotide sequences. The screening preferably is performed using high- stringency conditions (described elsewhere herein) to identify those sequences that are closely related by sequence identity. Nucleic acids so identified can be translated into polypeptides and the polypeptides can be tested for activity.

Additionally, the present invention includes polypeptides that have substantially similar sequence identity to the polypeptides of the present invention. As used herein, two polypeptides have "substantial sequence identity" when there is at least about 70% sequence identity, at least about 80% sequence identity, at least about 90%> sequence identity, at least about 95% sequence identity, at least about 99% sequence identity, and preferably 100% sequence identity between their amino acid sequences, or when polynucleotides encoding the polypeptides are capable of forming a stable duplex with each other under stringent hybridization conditions. For example, conservative amino acid substitutions may be made in polypeptides to provide functionally equivalent variants of the foregoing polypeptides, i.e., the variants retain the functional capabilities of the polypeptides. As used herein, a "conservative amino acid substitution" refers to an amino acid substitution that does not alter the relative charge or size characteristics of the protein in which the amino acid substitution is made. Variants can be prepared according to methods for altering polypeptide sequence known to one of ordinary skill in the art such as are found in references that compile such methods. For example, upon determining that a peptide is a bladder cancer-associated polypeptide, one can make conservative amino acid substitutions to the amino acid sequence of the peptide, and still have the polypeptide retain its specific antibody-binding characteristics. Additionally, one skilled in the art will realize that allelic variants and SNPs will give rise to substantially similar polypeptides and the same or substantially similar polypeptide fragments.

A number of comparison studies were performed to identify the polypeptide markers listed using various groups of node positive bladder cancer and node negative bladder cancer patients. The tables list markers that were found to be differentially expressed with statistical significance. Accordingly, it is believed that these bio markers and the expression levels thereof are indicative of nodep positive bladder cancer.

Accordingly, in one aspect, the invention provides polypeptide bio markers of bladder cancer. In one embodiment, the invention provides an isolated component listed in Table A. In another embodiment, the invention provides a polypeptide having substantial sequence identity with a component set forth in Table A. In another embodiment, the invention provides a molecule that comprises a foregoing polypeptide. As used herein, a compound is referred to as "isolated" when it has been separated from at least one component with which it is naturally associated. For example, a polypeptide can be considered isolated if it is separated from contaminants including metabolites,

polynucleotides and other polypeptides. Isolated molecules can be either prepared synthetically or purified from their natural environment. Standard quantification methodologies known in the art can be employed to obtain and isolate the molecules of the invention.

Some variation is inherent in the measurements of the physical and chemical characteristics of the markers. The magnitude of the variation depends to some extent on the reproductivity of the separation means and the specificity and sensitivity of the detection means used to make the measurement. Preferably, the method and technique used to measure the markers is sensitive and reproducible.

When a sample is processed and analyzed as described in the Example, the retention time of the marker is about the value stated for the marker; that is, within about 10% of the value stated, within about 5% of the value stated, or within about 1% of the value stated, and the marker has a mass to charge ratio of about the value stated for the marker; that is, within about 10% of the value stated, within about 5% of the value stated, or within about 1% of the value stated. Accordingly, in another embodiment, the invention provides a polypeptide having (i) a mass-to-charge value and (ii) an RT value of about the values stated, respectively, for a component listed in Table A. In another embodiment, the invention provides a molecule that comprises a foregoing polypeptide.

Polypeptides corresponding to the markers identified in Table A reflect a single polypeptide appearing in a database for which the component was a match. In general, the polypeptide is the largest polypeptide found in the database. But such a selection is not meant to limit the polypeptide to those corresponding to the markers disclosed in Table A. Accordingly, in another embodiment, the invention provides a polypeptide that is a fragment, precursor, successor or modified version of a marker described in Table A. In another embodiment, the invention includes a molecule that comprises a foregoing fragment, precursor, successor or modified polypeptide.

Another embodiment of the present invention relates to an assay system including a plurality of antibodies, or antigen binding fragments thereof, or aptamers for the detection of the expression of bio markers differentially expressed in patients with bladder cancer. The plurality of antibodies, or antigen binding fragments thereof, or aptamers consists of antibodies, or antigen binding fragments thereof, or aptamers that selectively bind to proteins differentially expressed in patients with node positive bladder cancer, and that can be detected as protein products using antibodies or aptamers. In addition, the plurality of antibodies, or antigen binding fragments thereof, or aptamers comprises antibodies, or antigen binding fragments thereof, or aptamers that selectively bind to proteins or portions thereof (peptides) encoded by any of the genes from Table A.

Certain embodiments of the present invention utilize a plurality of bio markers that have been identified herein as being differentially expressed in subjects with bladder cancer. As used herein, the terms "patient," "subject" and "a subject who has bladder cancer" and "bladder cancer patient" are intended to refer to subjects who have been diagnosed with bladder cancer. The terms "non-subject" and "a subject who does not have bladder cancer" are intended to refer to a subject who has not been diagnosed with bladder cancer, or who is cancer- free as a result of surgery to remove the diseased bladder. A non- bladder cancer subject may be healthy and have no other disease, or they may have a disease other than bladder cancer.

The plurality of biomarkers within the above-limitation includes at least two or more biomarkers (e.g., at least 2, 3, 4, 5, 6, and so on, in whole integer increments, up to all of the possible biomarkers) identified by the present invention, and includes any combination of such biomarkers. Such biomarkers are selected from any of the polypeptides listed in the tables provided herein, and polynucleotides encoding any of the polypeptides listed in the Tables. In a preferred embodiment, the plurality of biomarkers used in the present invention includes all of the biomarkers in the gene signature that has been demonstrated to be predictive of benefit from the therapeutic administration of neoadjuvant therapy, in a bladder cancer patient.

The polypeptide and polynucleotide markers of the invention are useful in methods for diagnosing bladder cancer, determining the extent and/or severity of the disease, monitoring progression of the disease and/or response to therapy. Such methods can be performed in human and non- human subjects. The markers are also useful in methods for treating bladder cancer and for evaluating the efficacy of treatment for the disease. Such methods can be performed in human and non-human subjects. The markers may also be used as pharmaceutical compositions or in kits. The markers may also be used to screen candidate compounds that modulate their expression. The markers may also be used to screen candidate drugs for treatment of bladder cancer. Such screening methods can be performed in human and non-human subjects. Polypeptide markers may be isolated by any suitable method known in the art. Native polypeptide markers can be purified from natural sources by standard methods known in the art (e.g., chromatography, centrifugation, differential solubility,

immunoassay). In one embodiment, polypeptide markers may be isolated from a sample by contacting the sample with substrate-bound antibodies or aptamers that specifically bind to the marker.

The present invention also included polynucleotide markers related to the polypeptide markers of the present invention. In one aspect, the invention provides polynucleotides that encode the polypeptides of the invention. The polynucleotide may be genomic DNA, cDNA, or mRNA transcripts that encode the polypeptides of the invention. In one embodiment, the invention provides polynucleotides that encode a polypeptide described in Table A, or a molecule that comprises such a polypeptide.

In another embodiment, the invention provides polynucleotides that encode a polypeptide having substantial sequence identity with a component set forth in Table A, or a molecule that comprises such a polypeptide.

In another embodiment, the invention provides polynucleotides that encode a polypeptide that is a fragment, precursor, successor or modified version of a marker described in Table A, or a molecule that comprises such polypeptide.

In another embodiment, the invention provides polynucleotides that have substantial sequence similarity to a polynucleotide that encodes a polypeptide that is a fragment, precursor, successor or modified version of a marker described in Table A, or a molecule that comprises such polypeptide. Two polynucleotides have "substantial sequence identity" when there is at least about 70% sequence identity, at least about 80%> sequence identity, at least about 90% sequence identity, at least about 95% sequence identity or at least 99% sequence identity between their amino acid sequences or when the polynucleotides are capable of forming a stable duplex with each other under stringent hybridization conditions. Such conditions are described elsewhere herein. As described above with respect to polypeptides, the invention includes polynucleotides that are allelic variants, the result of SNPs, or that in alternative codons to those present in the native materials as inherent in the degeneracy of the genetic code.

In some embodiments, the polynucleotides described may be used as surrogate markers of bladder cancer or node positive bladder cancer. Thus, for example, if the level of a polypeptide marker is increased in bladder cancer-patients, an increase in the mRNA that encodes the polypeptide marker may be interrogated rather than the polypeptide marker (e.g., to diagnose bladder cancer in a subject).

Polynucleotide markers may be isolated by any suitable method known in the art. Native polynucleotide markers may be purified from natural sources by standard methods known in the art (e.g., chromatography, centrifugation, differential solubility,

immunoassay). In one embodiment, a polynucleotide marker may be isolated from a mixture by contacting the mixture with substrate bound probes that are complementary to the polynucleotide marker under hybridization conditions.

Alternatively, polynucleotide markers may be synthesized by any suitable chemical or recombinant method known in the art. In one embodiment, for example, the makers can be synthesized using the methods and techniques of organic chemistry. In another embodiment, a polynucleotide marker can be produced by polymerase chain reaction (PCR).

The present invention also encompasses molecules that specifically bind the polypeptide or polynucleotide markers of the present invention. In one aspect, the invention provides molecules that specifically bind to a polypeptide marker or a polynucleotide marker. As used herein, the term "specifically binding," refers to the interaction between binding pairs (e.g., an antibody and an antigen or aptamer and its target). In some embodiments, the interaction has an affinity constant of at most 10^~6 moles/liter, at most 10^~7 moles/liter, or at most 10^~8 moles/liter. In other embodiments, the phrase "specifically binds" refers to the specific binding of one protein to another (e.g., an antibody, fragment thereof, or binding partner to an antigen), wherein the level of binding, as measured by any standard assay (e.g., an immunoassay), is statistically significantly higher than the background control for the assay. For example, when performing an immunoassay, controls typically include a reaction well/tube that contain antibody or antigen binding fragment alone (i.e., in the absence of antigen), wherein an amount of reactivity (e.g., non-specific binding to the well) by the antibody or antigen binding fragment thereof in the absence of the antigen is considered to be background. Binding can be measured using a variety of methods standard in the art including enzyme immunoassays (e.g., ELISA), immunoblot assays, etc.).

The binding molecules include antibodies, aptamers and antibody fragments. As used herein, the term "antibody" refers to an immunoglobulin molecule capable of binding an epitope present on an antigen. The term is intended to encompasses not only intact immunoglobulin molecules such as monoclonal and polyclonal antibodies, but also bi- specific antibodies, humanized antibodies, chimeric antibodies, anti-idiopathic (anti-ID) antibodies, single-chain antibodies, Fab fragments, F(ab') fragments, fusion proteins and any modifications of the foregoing that comprise an antigen recognition site of the required specificity. As used herein, an aptamer is a non-naturally occurring nucleic acid having a desirable action on a target. A desirable action includes, but is not limited to, binding of the target, catalytically changing the target, reacting with the target in a way which modifies/alters the target or the functional activity of the target, covalently attaching to the target as in a suicide inhibitor, facilitating the reaction between the target and another molecule, in the preferred embodiment, the action is specific binding affinity for a target molecule, such target molecule being a three dimensional chemical structure other than a polynucleotide that binds to the nucleic acid ligand through a mechanism which predominantly depends on Watson/Crick base pairing or triple helix binding, wherein the nucleic acid ligand is not a nucleic acid having the known physiological function of being bound by the target molecule.

In one aspect, the invention provides antibodies or aptamers that specifically bind to a component listed in Table A, or to a molecule that comprises a foregoing component (e.g., a protein comprising a polypeptide identified in a table of the invention).

In another embodiment, the invention provides antibodies or aptamers that specifically bind to a polypeptide having substantial sequence identity with a component set forth in Table A, or to a molecule that comprises a foregoing polypeptide.

In another embodiment, the invention provides antibodies or aptamers that specifically bind to a component that is a fragment, modification, precursor or successor of a marker described in Table A, or to a molecule that comprises a foregoing component.

In another embodiment, the invention provides antibodies or aptamers that specifically bind to a polypeptide marker or a polynucleotide marker that is structurally different from a component specifically identified in Table A but has the same (or nearly the same) function or properties, or to a molecule that comprises a foregoing component.

Another embodiment of the present invention relates to a plurality of aptamers, antibodies, or antigen binding fragments thereof, for the detection of the expression of bio markers differentially expressed in patients with bladder cancer. The plurality of aptamers, antibodies, or antigen binding fragments thereof, consists of antibodies, or antigen binding fragments thereof, that selectively bind to proteins differentially expressed in patients with bladder cancer, and that can be detected as protein products using antibodies. In addition, the plurality of aptamers, antibodies, or antigen binding fragments thereof, comprises antibodies, or antigen binding fragments thereof, that selectively bind to proteins or portions thereof (peptides) encoded by any of the genes from the tables provided herein.

According to the present invention, a plurality of aptamers, antibodies, or antigen binding fragments thereof, refers to at least 2, and more preferably at least 3, and more preferably at least 4, and more preferably at least 5, and more preferably at least 6, and more preferably at least 7, and more preferably at least 8, and more preferably at least 9, and more preferably at least 10, and so on, in increments of one, up to any suitable number of antibodies, or antigen binding fragments thereof, including, in a preferred embodiment, antibodies representing all of the bio markers described herein, or antigen binding fragments thereof.

Certain antibodies that specifically bind polypeptide markers polynucleotide markers of the invention already may be known and/or available for purchase from commercial sources. In any event, the antibodies of the invention may be prepared by any suitable means known in the art. For example, antibodies may be prepared by immunizing an animal host with a marker or an immunogenic fragment thereof (conjugated to a carrier, if necessary). Adjuvants (e.g., Freund's adjuvant) optionally may be used to increase the immunological response. Sera containing polyclonal antibodies with high affinity for the antigenic determinant can then be isolated from the immunized animal and purified.

Alternatively, antibody-producing tissue from the immunized host can be harvested and a cellular homogenate prepared from the organ can be fused to cultured cancer cells. Hybrid cells which produce monoclonal antibodies specific for a marker can be selected. Alternatively, the antibodies of the invention can be produced by chemical synthesis or by recombinant expression. For example, a polynucleotide that encodes the antibody can be used to construct an expression vector for the production of the antibody. The antibodies of the present invention can also be generated using various phage display methods known in the art.

Antibodies or aptamers that specifically bind markers of the invention can be used, for example, in methods for detecting components listed in Table A using methods and techniques well-known in the art. In some embodiments, for example, the antibodies are conjugated to a detection molecule or moiety (e.g., a dye, and enzyme) and can be used in ELISA or sandwich assays to detect markers of the invention.

In another embodiment, antibodies or aptamers against a polypeptide marker or polynucleotide marker of the invention can be used to assay a tissue sample (e.g., a thin cortical slice) for the marker. The antibodies or aptamers can specifically bind to the marker, if any, present in the tissue sections and allow the localization of the marker in the tissue. Similarly, antibodies or aptamers labeled with a radioisotope may be used for in vivo imaging or treatment applications.

Another aspect of the invention provides compositions comprising a polypeptide or polynucleotide marker of the invention, a binding molecule that is specific for a polypeptide or polynucleotide marker (e.g., an antibody or an aptamer), an inhibitor of a polypeptide or polynucleotide marker, or other molecule that can increase or decrease the level or activity of a polypeptide marker or polynucleotide marker. Such compositions may be pharmaceutical compositions formulated for use as a therapeutic.

Alternatively, the invention provides a composition that comprises a component that is a fragment, modification, precursor or successor of a marker described in Table A, or to a molecule that comprises a foregoing component.

In another embodiment, the invention provides a composition that comprises a polynucleotide that binds to a polypeptide or a molecule that comprises a foregoing polynucleotide.

In another embodiment, the invention provides a composition that comprises an antibody or aptamer that specifically binds to a polypeptide or a molecule that comprises a foregoing antibody or aptamer.

The present invention also provides methods of detecting the biomarkers of the present invention. The practice of the present invention employs, unless otherwise indicated, conventional methods of analytical biochemistry, microbiology, molecular biology and recombinant DNA techniques within the skill of the art. Such techniques are explained fully in the literature. (See, e.g., Sambrook, J. et al. Molecular Cloning: A Laboratory Manual. 3rd, ed., Cold Spring Harbor Laboratory, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 2000; DNA Cloning: A Practical Approach, Vol. I & II (D. Glover, ed.); Oligonucleotide Synthesis (N. Gait, ed., Current Edition); Nucleic Acid Hybridization (B. Hames & S. Higgins, eds., Current Edition); Transcription and Translation (B. Hames & S. Higgins, eds., Current Edition); CRC Handbook of Parvoviruses, Vol. I & II (P. Tijessen, ed.); Fundamental Virology, 2nd Edition,

Vol. I & II (B. N. Fields and D. M. Knipe, eds.)).

The markers of the invention may be detected by any method known to those of skill in the art, including without limitation LC-MS, GC-MS, immunoassays,

immunohistochemistry, hybridization, microarray and enzyme assays. The detection may be quantitative or qualitative. A wide variety of conventional techniques are available, including mass spectrometry, chromatographic separations, 2-D gel separations, binding assays (e.g., immunoassays), competitive inhibition assays, and so on. Any effective method in the art for measuring the presence/absence, level or activity of a polypeptide or polynucleotide is included in the invention. It is within the ability of one of ordinary skill in the art to determine which method would be most appropriate for measuring a specific marker. Thus, for example, a ELISA assay may be best suited for use in a physician's office while a measurement requiring more sophisticated instrumentation may be best suited for use in a clinical laboratory. Regardless of the method selected, it is important that the measurements be reproducible.

The markers of the invention can be measured by mass spectrometry, which allows direct measurements of analytes with high sensitivity and reproducibility. A number of mass spectrometric methods are available. Electrospray ionization (ESI), for example, allows quantification of differences in relative concentration of various species in one sample against another; absolute quantification is possible by normalization techniques (e.g., using an internal standard). Matrix-assisted laser desorption ionization (MALDI) or the related SELDI® technology (Ciphergen, Inc.) also could be used to make a determination of whether a marker was present, and the relative or absolute level of the marker. Mass spectrometers that allow time-of- flight (TOF) measurements have high accuracy and resolution and are able to measure low abundant species, even in complex matrices like serum or CSF.

For protein markers, quantification can be based on derivatization in combination with isotopic labeling, referred to as isotope coded affinity tags ("ICAT"). In this and other related methods, a specific amino acid in two samples is differentially and isotopically labeled and subsequently separated from peptide background by solid phase capture, wash and release. The intensities of the molecules from the two sources with different isotopic labels can then be accurately quantified with respect to one another. Quantification can also be based on the isotope dilution method by spiking in an isotopically labeled peptide or protein analogous to those being measured. Furthermore, quantification can also be determined without isotopic standards using the direct intensity of the analyte comparing with another measurement of a standard in a similar matrix.

In addition, one- and two-dimensional gels have been used to separate proteins and quantify gels spots by silver staining, fluorescence or radioactive labeling. These differently stained spots have been detected using mass spectrometry, and identified by tandem mass spectrometry techniques.

In one embodiment, the markers are measured using mass spectrometry in connection with a separation technology, such as liquid chromatography-mass

spectrometry or gas chromatography-mass spectrometry. In particular, coupling reverse- phase liquid chromatography to high resolution, high mass accuracy ESI time-of- flight (TOF) mass spectroscopy allows spectral intensity measurement of a large number of biomolecules from a relatively small amount of any complex biological material.

Analyzing a sample in this manner allows the marker (characterized by a specific RT and m/z) to be determined and quantified.

As will be appreciated by one of skill in the art, many other separation

technologies may be used in connection with mass spectrometry. For example, a wide selection of separation columns is commercially available. In addition, separations may be performed using custom chromatographic surfaces (e.g., a bead on which a marker specific reagent has been immobilized). Molecules retained on the media subsequently may be eluted for analysis by mass spectrometry.

In other preferred embodiments, the level of the markers may be determined using microarray analysis or standard immunoassay, such as sandwiched ELISA using matched antibody pairs and chemiluminescent detection. Commercially available or custom monoclonal or polyclonal antibodies are typically used. However, the assay can be adapted for use with other reagents that specifically bind to the marker. Standard protocols and data analysis are used to determine the marker concentrations from the assay data.

A number of the assays discussed above employ a reagent that specifically binds to the marker. Any molecule that is capable of specifically binding to a marker is included within the invention. In some embodiments, the binding molecules are antibodies or antibody fragments. In other embodiments, the binding molecules are non-antibody species, such as aptamers. Thus, for example, the binding molecule may be an enzyme for which the marker is a substrate. The binding molecules may recognize any epitope of the targeted markers.

As described above, the binding molecules may be identified and produced by any method accepted in the art. Methods for identifying and producing antibodies and antibody fragments specific for an analyte are well known. Examples of other methods used to identify the binding molecules include binding assays with random peptide libraries (e.g., phage display) and design methods based on an analysis of the structure of the marker.

The markers of the invention also may be detected or measured using a number of chemical derivatization or reaction techniques known in the art. Reagents for use in such techniques are known in the art, and are commercially available for certain classes of target molecules.

Finally, the chromatographic separation techniques described above also may be coupled to an analytical technique other than mass spectrometry such as fluorescence detection of tagged molecules, NMR, capillary UV, evaporative light scattering or electrochemical detection.

Measurement of the relative amount of an RNA or protein marker of the invention may be by any method known in the art (see, e.g., Sambrook, J., Fritsh, E. F., and Maniatis, T. Molecular Cloning: A Laboratory Manual. 2nd, ed., Cold Spring Harbor Laboratory, Cold Spring Harbor Laboratory Press, Cold Spring Harbor, NY, 1989; and Current Protocols in Molecular Biology, eds. Ausubel et al. John Wiley & Sons: 1992). Typical methodologies for RNA detection include RNA extraction from a cell or tissue sample, followed by hybridization of a labeled probe (e.g., a complementary

polynucleotide) specific for the target RNA to the extracted RNA, and detection of the probe (e.g., Northern blotting). Typical methodologies for protein detection include protein extraction from a cell or tissue sample, followed by hybridization of a labeled probe (e.g., an antibody) specific for the target protein to the protein sample, and detection of the probe. The label group can be a radioisotope, a fluorescent compound, an enzyme, or an enzyme co-factor. Detection of specific protein and polynucleotides may also be assessed by gel electrophoresis, column chromatography, direct sequencing, or quantitative PCR (in the case of polynucleotides) among many other techniques well known to those skilled in the art.

Detection of the presence or number of copies of all or a part of a marker gene of the invention may be performed using any method known in the art. Typically, it is convenient to assess the presence and/or quantity of a DNA or cDNA by Southern analysis, in which total DNA from a cell or tissue sample is extracted, is hybridized with a labeled probe (e.g., a complementary DNA molecule), and the probe is detected. The label group can be a radioisotope, a fluorescent compound, an enzyme, or an enzyme co- factor. Other useful methods of DNA detection and/or quantification include direct sequencing, gel electrophoresis, column chromatography, and quantitative PCR, as is known by one skilled in the art.

Polynucleotide similarity can be evaluated by hybridization between single stranded nucleic acids with complementary or partially complementary sequences. Such experiments are well known in the art. High stringency hybridization and washing conditions, as referred to herein, refer to conditions which permit isolation of nucleic acid molecules having at least about 80% nucleic acid sequence identity with the nucleic acid molecule being used to probe in the hybridization reaction (i.e., conditions permitting about 20% or less mismatch of nucleotides). Very high stringency hybridization and washing conditions, as referred to herein, refer to conditions which permit isolation of nucleic acid molecules having at least about 90%> nucleic acid sequence identity with the nucleic acid molecule being used to probe in the hybridization reaction (i.e., conditions permitting about 10%> or less mismatch of nucleotides). As discussed above, one of skill in the art can use the formulae in Meinkoth et al., ibid, to calculate the appropriate hybridization and wash conditions to achieve these particular levels of nucleotide mismatch. Such conditions will vary, depending on whether DNA:R A or DNA:DNA hybrids are being formed. Calculated melting temperatures for DNA:DNA hybrids are 10°C less than for DNA:R A hybrids. In particular embodiments, stringent hybridization conditions for DNA:DNA hybrids include hybridization at an ionic strength of 6X SSC (0.9 M Na⁺) at a temperature of between about 20°C and about 35°C (lower stringency), more preferably, between about 28°C and about 40°C (more stringent), and even more preferably, between about 35°C and about 45°C (even more stringent), with appropriate wash conditions. In particular embodiments, stringent hybridization conditions for DNA:R A hybrids include hybridization at an ionic strength of 6X SSC (0.9 M Na⁺) at a temperature of between about 30°C and about 45°C, more preferably, between about 38°C and about 50°C, and even more preferably, between about 45°C and about 55°C, with similarly stringent wash conditions. These values are based on calculations of a melting temperature for molecules larger than about 100 nucleotides, 0% formamide and a G + C content of about 40%. Alternatively, T_m can be calculated empirically as set forth in Sambrook et al, supra, pages 9.31 to 9.62. In general, the wash conditions should be as stringent as possible, and should be appropriate for the chosen hybridization conditions. For example, hybridization conditions can include a combination of salt and temperature conditions that are approximately 20-25°C below the calculated T_m of a particular hybrid, and wash conditions typically include a combination of salt and temperature conditions that are approximately 12-20°C below the calculated T_m of the particular hybrid. One example of hybridization conditions suitable for use with DNA:DNA hybrids includes a 2- 24 hour hybridization in 6X SSC (50% formamide) at about 42°C, followed by washing steps that include one or more washes at room temperature in about 2X SSC, followed by additional washes at higher temperatures and lower ionic strength (e.g., at least one wash as about 37°C in about 0.1X-0.5X SSC, followed by at least one wash at about 68°C in about 0.1X-0.5X SSC). Other hybridization conditions, and for example, those most useful with nucleic acid arrays, will be known to those of skill in the art.

The present invention also includes methods of diagnosing bladder cancer, or node positive bladder cancer and related methods. In general, it is expected that the biomarkers described herein will be measured in combination with other signs, symptoms and clinical tests of bladder cancer, bladder cancer disease state, cancer stage nodal involvement or presence of metastases such as MRI or ultrasound abnormalities, or bladde cancer biomarkers reported in the literature. Likewise, more than one of the biomarkers of the present invention may be measured in combination. Measurement of the biomarkers of the invention along with any other markers known in the art, including those not specifically listed herein, falls within the scope of the present invention. Markers appropriate for this embodiment include those that have been identified as increased or decreased in samples obtained from bladder cancer samples and node positive bladder cancer subjects compared with samples from non-bladder cancer samples or node negative bladder cancer subjects (e.g., markers described in Table A, as well as antibodies produced by a patient in response to an increased level of a polypeptide marker. Other markers appropriate for this embodiment include fragments, precursors, successors and modified versions of such markers, polypeptides having substantial sequence identity to such markers, components having an m/z value and RT value of about the values set forth for the markers described in Table A, and molecules comprise one of the foregoing. Other appropriate markers for this embodiment will be apparent to one of skill in the art in light of the disclosure herein.

In one embodiment, the present invention provides a method for determining whether a bladder cancer patient is likely to benefit from the therapeutic administration of neoadjuvant therapy. In another aspect, the invention provides methods for identifying a bladder cancer patient that is likely to suffer a recurrence of the cancer following cystectomy. In another aspect, the invention provides methods for identifying a bladder cancer patient that is likely to survive or be diagnosed disease-fee of the cancer following cystectomy and/or adjuvant therapy. These methods comprise obtaining a tumor sample from a subject suspected of having bladder cancer, detecting the level or activity of one or more biomarkers in the sample, and comparing the result to the level or activity of the marker(s) in a sample obtained from a bladder cancer subject known to respond or not respond to neoadjuvant chemotherapy, or to a reference range or value. As used herein, the term "biological sample" includes a sample from any body fluid or tissue (e.g., serum, plasma, blood, cerebrospinal fluid, urine, bladder tissue). Typically, the standard biomarker level or reference range is obtained by measuring the same marker or markers in a set of normal controls. Measurement of the standard biomarker level or reference range need not be made contemporaneously; it may be a historical measurement.

Preferably the normal control is matched to the patient with respect to some attribute(s) (e.g., disease stage at diagnosis). Depending upon the difference between the measured and standard level or reference range, the patient can be predicted to respond to or benefit from the therapeutic administration of neoadjuvant therapy. In some embodiments, the patient is predicted to benefit from the therapeutic administration of neoadjuvant therapy if the expression level of the biomarker or biomarkers in the patient sample is statistically more similar to the expression level of the biomarker or biomarkers that has been associated with bladder cancer than the expression level of the biomarker or biomarkers that has been associated with the normal controls.

The methods of the present invention may be used to identify patients likely to benefit from the therapeutic administration of neoadjuvant therapy independently from other information such as the patient's symptoms or the results of other clinical or paraclinical tests. However, the methods of the present invention may be used in conjunction with such other data points.

Because a diagnosis is rarely based exclusively on the results of a single test, the method may be used to determine whether a subject is more likely than not to benefit from the therapeutic administration of neoadjuvant therapy, based on the difference between the measured and standard level or reference range of the biomarker. Thus, for example, a patient with a putative diagnosis of bladder cancer may be diagnosed as being "more likely" or "less likely" to benefit from the therapeutic administration of neoadjuvant therapy in light of the information provided by a method of the present invention. If a plurality of biomarkers are measured, at least one and up to all of the measured biomarkers must differ, in the appropriate direction, for the subject to be identified as likely to benefit from the therapeutic administration of neoadjuvant therapy. In some embodiments, such difference is statistically significant.

The biological sample may be of a bladder tumor tissue or fluid, including a serum or tissue sample, but other biological fluids or tissue may be used. Possible biological fluids include, but are not limited to, plasma, and urine. In some embodiments, the level of a marker may be compared to the level of another marker or some other component in a different tissue, fluid or biological "compartment." Thus, a differential comparison may be made of a marker in tissue and serum. It is also within the scope of the invention to compare the level of a marker with the level of another marker or some other component within the same compartment.

As will be apparent to those of ordinary skill in the art, the above description is not limited to making an initial identification of patients that may benefit from the therapeutic administration of neoadjuvant therapy, but also is applicable to confirming a provisional diagnosis of bladder cancer or nodal involvement in bladder cancer or disease stage of bladder cancer or "ruling out" such a diagnosis. Furthermore, an increased or decreased level or activity of the marker(s) in a sample obtained from a subject suspected of having bladder cancer, or at risk for developing bladder cancer, is indicative that the subject has or is at risk for developing bladder cancer, nodal involvement and/or metastases.

The invention also provides a method for determining a subject's risk of developing bladder cancer, the method comprising obtaining a biological sample from a subject, detecting the level or activity of a marker in the sample, and comparing the result to the level or activity of the marker in a sample obtained from a non- bladder cancer subject, or to a reference range or value wherein an increase or decrease of the marker is correlated with the risk of developing bladder cancer.

The invention also provides methods for determining the stage or severity of bladder cancer, the method comprising obtaining a biological sample from a subject, detecting the level or activity of a marker in the sample, and comparing the result to the level or activity of the marker in a sample obtained from a non-bladder cancer subject or node negative bladder cancer, or to a reference range or value wherein an increase or decrease of the marker is correlated with the stage or severity of the disease.

In another aspect, the invention provides methods for monitoring the progression of the disease in a subject who has bladder cancer, the method comprising obtaining a first biological sample from a subject, detecting the level or activity of a marker in the sample, and comparing the result to the level or activity of the marker in a second sample obtained from the subject at a later time, or to a reference range or value wherein an increase or decrease of the marker is correlated with progression of the disease.

The marker expression measurement values for the markers listed in Table A are elevated in node positive bladder cancer samples. A significant difference in the elevation of the measured value of one or more of the markers indicates that the patient has (or is more likely to have, or is at risk of having, or is at risk of developing, and so forth) bladder cancer and/or node positive disease. If only one biomarker is measured, then that value must increase to indicate bladder cancer and/or node positive disease. If more than one biomarker is measured, then a diagnosis of bladder cancer can be indicated by a change in only one biomarker, all biomarkers, or any number in between. In some embodiments, multiple markers are measured, and a diagnosis of bladder cancer and/or node positive disease is indicated by changes in multiple markers. For example, a panel of markers may include markers that are increased in level or activity in node-positive bladder cancer subject samples as compared to node negative-bladder cancer subject samples, markers that are decreased in level or activity in bladder cancer subject samples as compared to non-bladder cancer subject samples, or a combination thereof.

The marker(s) may be detected in any biological sample obtained from the subject, by any suitable method known in the art (e.g., immunoassays, hybridization assay) see supra. Preferably, the marker(s) are detected in a tumor sample obtained from the patient by surgical procedure(s).

In an alternative embodiment of the invention, a method is provided for monitoring a bladder cancer patient over time to determine whether the disease is progressing or an administered therapy is effective. The specific techniques used in implementing this embodiment are similar to those used in the embodiments described above. The method is performed by obtaining a biological sample, such as serum or tissue, from the subject at a certain time (t;); measuring the level of at least one of the biomarkers in the biological sample; and comparing the measured level with the level measured with respect to a biological sample obtained from the subject at an earlier time (to). Depending upon the difference between the measured levels, it can be seen whether the marker level has increased, decreased, or remained constant over the interval (ti-to). A further deviation of a marker in the direction indicating bladder cancer, or the measurement of additional increased or decreased bladder cancer markers, would suggest a progression of the disease during the interval. Subsequent sample acquisitions and measurements can be performed as many times as desired over a range of times Ϊ2 to t„. The ability to monitor a patient by making serial marker level determinations would represent a valuable clinical tool. Rather than the limited "snapshot" provided by a single test, such monitoring would reveal trends in marker levels over time. In addition to indicating a progression of the disease, tracking the marker levels in a patient could be used to predict exacerbations or indicate the clinical course of the disease. For example, as will be apparent to one of skill in the art, the biomarkers of the present invention could be further investigated to distinguish between any or all of the known forms of bladder cancer or any later described types or subtypes of the disease. In addition, the sensitivity and specificity of any method of the present invention could be further investigated with respect to distinguishing bladder cancer from other diseases or to predict relapse or remission.

In an analogous manner, administration a chemotherapeutic drug or drug combination can be evaluated or re-evaluated in light of the assay results of the present invention. For example, the drug(s) can be administered differently to different subject populations, and measurements corresponding to administration analyzed to determine if the differences in the inventive biomarker signature before and after drug administration are significant. Results from the different drug regiments can also be compared with each other directly. Alternatively, the assay results may indicate the desirability of one drug regimen over another, or indicate that a specific drug regimen should or should not be administered to a bladder cancer patient. In one preferred embodiment, the finding of elevated levels of the marker genes of the present invention in a bladder cancer patient is indicative of a good prognosis for benefit from the therapeutic administration of neoadjuvant therapy. In another preferred embodiment, the absence of elevated levels of the marker genes of the present invention in a bladder cancer patient is indicative of a poor prognosis for benefit from the therapeutic administration of neoadjuvant therapy, and may further recommend not administering neoadjuvant therapy.

In an analogous manner, the markers of the present invention can be used to assess the efficacy of a therapeutic intervention in a subject. The same approach described above would be used, except a suitable treatment would be started, or an ongoing treatment would be changed, before the second measurement (i.e., after to and before ti). The treatment can be any therapeutic intervention, such as drug administration, dietary restriction or surgery, and can follow any suitable schedule over any time period as appropriate for the intervention. The measurements before and after could then be compared to determine whether or not the treatment had an effect effective. As will be appreciated by one of skill in the art, the determination may be confounded by other superimposed processes (e.g., an exacerbation of the disease during the same period).

In another aspect, the invention provides a kit for detecting polynucleotide or polypeptide marker(s) of the present invention. The kit may be prepared as an assay system including any one of assay reagents, assay controls, protocols, exemplary assay results, or combinations of these components designed to provide the user with means to evaluate the expression level of the marker(s) of the present invention.

In another aspect, the invention provides a kit for predicting the likelihood of benefit from the therapeutic administration of neoadjuvant therapy in a patient including reagents for detecting at least one polypeptide or polynucleotide marker in a biological sample from a subject.

The kits of the invention may comprise one or more of the following: an antibody, wherein the antibody specifically binds with a polypeptide marker, a labeled binding partner to the antibody, a solid phase upon which is immobilized the antibody or its binding partner, a polynucleotide probe that can hybridize to a polynucleotide marker, pairs of primers that under appropriate reaction conditions can prime amplification of at least a portion of a polynucleotide marker or a polynucleotide encoding a polypeptide marker (e.g., by PCR), instructions on how to use the kit, and a label or insert indicating regulatory approval for diagnostic or therapeutic use.

The invention further includes polynucleotide or polypeptide microarrays comprising polypeptides of the invention, polynucleotides of the invention, or molecules, such as antibodies, which specifically bind to the polypeptides or polynucleotides of the present invention. In this aspect of the invention, standard techniques of microarray technology are utilized to assess expression of the polypeptides biomarkers and/or identify biological constituents that bind such polypeptides. Protein microarray technology is well known to those of ordinary skill in the art and is based on, but not limited to, obtaining an array of identified peptides or proteins on a fixed substrate, binding target molecules or biological constituents to the peptides, and evaluating such binding. Polynucleotide arrays, particularly arrays that bind polypeptides of the invention, also can be used for diagnostic applications, such as for identifying subjects that have a condition characterized by expression of polypeptide biomarkers, e.g., bladder cancer.

The assay system preferably also includes one or more controls. The controls may include: (i) a control sample for evaluating the benefit from the therapeutic administration of neoadjuvant therapy use in a patient; (ii) information containing a predetermined control level of markers to be measured with regard to benefit from the therapeutic administration of neoadjuvant therapy in a patient.

In another embodiment, a means for detecting the expression level of the marker(s) of the invention can generally be any type of reagent that can include, but is not limited to, antibodies and antigen binding fragments thereof, peptides, binding partners, aptamers, enzymes, and small molecules. Additional reagents useful for performing an assay using such means for detection can also be included, such as reagents for performing

immunohistochemistry or another binding assay.

The means for detecting of the assay system of the present invention can be conjugated to a detectable tag or detectable label. Such a tag can be any suitable tag which allows for detection of the reagents used to detect the gene or protein of interest and includes, but is not limited to, any composition or label detectable by spectroscopic, photochemical, electrical, optical or chemical means. Useful labels in the present invention include: biotin for staining with labeled streptavidin conjugate, magnetic beads (e.g., Dynabeads™), fluorescent dyes (e.g., fluorescein, texas red, rhodamine, green

3 125 35 14 32

fluorescent protein, and the like), radio labels (e.g. , H, I, S, C, or P), enzymes (e.g., horse radish peroxidase, alkaline phosphatase and others commonly used in an ELISA), and colorimetric labels such as colloidal gold or colored glass or plastic (e.g., polystyrene, polypropylene, latex, etc.) beads.

In addition, the means for detecting of the assay system of the present invention can be immobilized on a substrate. Such a substrate can include any suitable substrate for immobilization of a detection reagent such as would be used in any of the previously described methods of detection. Briefly, a substrate suitable for immobilization of a means for detecting includes any solid support, such as any solid organic, biopolymer or inorganic support that can form a bond with the means for detecting without significantly affecting the activity and/or ability of the detection means to detect the desired target molecule. Exemplary organic solid supports include polymers such as polystyrene, nylon, phenol- formaldehyde resins, and acrylic copolymers (e.g., polyacrylamide). The kit can also include suitable reagents for the detection of the reagent and/or for the labeling of positive or negative controls, wash solutions, dilution buffers and the like. The assay system can also include a set of written instructions for using the system and interpreting the results.

The assay system can also include a means for detecting a control marker that is characteristic of the cell type being sampled can generally be any type of reagent that can be used in a method of detecting the presence of a known marker (at the nucleic acid or protein level) in a sample, such as by a method for detecting the presence of a biomarker described previously herein. Specifically, the means is characterized in that it identifies a specific marker of the cell type being analyzed that positively identifies the cell type. For example, in a bladder tumor assay, it is desirable to screen bladder cancer cells for the level of the biomarker expression and/or biological activity. Therefore, the means for detecting a control marker identifies a marker that is characteristic of a bladder cell, so that the cell is distinguished from other cell types, such as a connective tissue or inflammatory cell. Such a means increases the accuracy and specificity of the assay of the present invention. Such a means for detecting a control marker include, but are not limited to: a probe that hybridizes under stringent hybridization conditions to a nucleic acid molecule encoding a protein marker; PCR primers which amplify such a nucleic acid molecule; an aptamer that specifically binds to a conformationally-distinct site on the target molecule; and/or an antibody, antigen binding fragment thereof, or antigen binding peptide that selectively binds to the control marker in the sample. Nucleic acid and amino acid sequences for many cell markers are known in the art and can be used to produce such reagents for detection.

The assay systems and methods of the present invention can be used not only to identify patients that are predicted to be responsive to neoadjuvant theapy, but also to identify treatments that can improve the responsiveness of cancer cells which are resistant to known combinations of chemotherapeutic agents, used routinely for the treatment of bladder cancer, and to develop adjuvant treatments that enhance the response of these chemotherapeutic agents.

Each publication or patent cited herein is incorporated herein by reference in its entirety.

The invention now being generally described will be more readily understood by reference to the following examples, which are included merely for the purposes of illustration of certain aspects of the embodiments of the present invention. The examples are not intended to limit the invention, as one of skill in the art would recognize from the above teachings and the following examples that other techniques and methods can satisfy the claims and can be employed without departing from the scope of the claimed invention. EXAMPLES

Patients and tissue samples

Specimens from three independent cohorts of patients who underwent cystectomy and lymphadenectomy for bladder cancer were used to develop and validate the GEM for nodal prediction. All patients were clinically node negative before cystectomy and had complete pathological staging information. No patients in any cohort were treated by systemic chemotherapy (adjuvant, neoadjuvant, or otherwise) before tissue used in the study was harvested. Two studies used archival formalin- fixed paraffin embedded (FFPE) tissues of cystectomy specimens: 1) the Laval cohort (95 patients) from l'Hopital de l'Hotel-Dieu at Laval University, Quebec, Canada and, 2) the German

Arbeitsgemeinschaft Urologische Onkologie (AUO) cohort (188 patients), which was from a phase 3, multicentre randomised controlled trial that compared two adjuvant chemotherapy regimens (AUO-AB 05/95).

Archival tissues could only be retrieved for 200 of 327 patients reported in the AUO cohort, of which 185 produced nucleic acid extracts of sufficient quality for microarray analysis. Distributions of clinicopathological variables, including survival, did not differ significantly between the subset profiled (185 patients) and the non-profiled group (142 patients).

Background information on Patient Cohorts

Three patient cohorts of cystectomy specimens of bladder carcinoma, profiled using Affymetrix U133 gene expression arrays, were analyzed in this study. We profiled formalin- fixed paraffin embedded (FFPE) archival tissues from the Laval Cohort (N=90) and the AUO Cohort (N=185) using the Affymetrix HGU133 Plus 2.0 oligonucleotide microarray platform. Of note, the AUO Cohort was from a previously reported Phase III, multicenter randomized control trial by the of two different adjuvant chemotherapy regimens, AUO-AB 05/95(Lehmann J, Retz M, Wiemers C, Beck J, Thuroff J, Weining C, et al. Adjuvant cisplatin plus methotrexate versus methotrexate, vinblastine, epirubicin, and cisplatin in locally advanced bladder cancer: results of a randomized, multicenter, phase III trial (AUO-AB 05/95). J Clin Oncol. 2005 Augl; 23(22):4963-74). This study originally encompassed 327 patients, though due to the retrospective nature of our project and multicentrality of the original trial, only 200 tissue blocks (61%) could be obtained and reviewed (A. Hartmann) for inclusion in our profiling project. Of those 200, sufficient quantity and quality RNA could be isolated for 188 (94%), while three of the

hybridizations were excluded due to poor quality, resulting in the final 185. Importantly, given the fact that our cohort was significantly reduced from the original one, we tested whether any key clinicopathologic parameter was different between the final subset profiled and used for the study as compared to the remainder which were not able to be used. We found that distributions of clinicopathologic variables did not differ significantly between the subset profiled and the entire group (gender, Ρ=0·46; grade, Ρ=0·49; T stage, P=l 0; N stage Ρ=0·91, chi-squared test for independence). A small, statistically significant but clinically unimportant trend toward younger age was noted (median 60 versus 62, Ρ<0·01). Most importantly, we tested whether the subset profiled exhibited different disease specific survival characteristics as compared to those not able to be used. As plotted by Kaplan Meier curve with 95% confidence intervals, survival between the subset studied and the remainder was not significantly different (Logrank, P=0.18).

Affymetrix HG-U133A data for a third patient cohort was available for download from a prior publication by Sanchez-Carbayo et al. of the Memorial Sloan-Kettering Cancer Center, the MSKCC Cohort (N= 66)(2). The Laval and MSKCC Cohorts were used for training, while the AUO Cohort was used strictly for independent testing.

The three cohorts were normalized on a per gene basis in order to allow for direct comparability. To achieve this, we implemented a weighted z-score approach wherein node positive and node negative classes are equally weighted instead of being biased by their proportions in a given dataset. In this manner, a positive weighted z-score indicates increased expression in node positive samples and a negative weighted z-score indicates decreased expression in node negative samples. However, in subset normalization the weights are binary terms that effectively center the data relative to the reference class. In our formulation each class is given equal weight, thereby centering the data in-between the two classes. The third group of muscle-invasive bladder cancer (>pT2) specimens were preserved by fresh freezing, and were obtained from supplementary data for a previous publication by Sanchez-Carbayo and colleagues at the Memorial Sloan-Kettering Cancer Center, New York, NY, USA— the MSKCC cohort (66 patients). Additional gene expression datasets were used for adaptation of the GEM to transurethral resection (TUR) specimens, adaptation of the GEM to use on FFPE rather than fresh frozen material to permit widespread applicability of the test, and assessment of the ability of the GEM to predict overall survival.

The node prediction datasets tested are summarized as follows:

Training Cohorts Laval Cohort - N=90 (pNO 17, pNl-3 73) / Affymetrix HG-U133 Plus 2.0, cohort and microarray data, unpublished

MS CC Cohort - N=66 (pNO 41, pNl-3 25) / Affymetrix HG-U133A, Sanchez-Carbayo et al. JCO 2006, supplementary microarray data at www.jco.org

Aarhus Cohort - N=30 (all stage group IV; >T4b &/or Nl-3, &/or Ml) / Affymetrix HG- U133A, Als et al. Clin Cancer Res 2007, NCBI GEO Accession# GSE5287

Independent Testing Cohort

AUO Cohort - N=185 (pNO 82, pNl-3 103) / Affymetrix HG-U133 Plus 2.0, cohort published as Lehmann et al. JCO 2005, microarray data unpublished

Cohort for Comparison of Formalin Fixed Paraffin Embedded to Flash Frozen

UVA/Almac Cohort - N=32 matched FFPE/FF tissue pairs / Affymetrix HG-U133 Plus 2.0. Cohort and microarray data unpublished. (22 profiled at the University of Virginia Microarray Core Facility, Charlottesville, VA, US, and 10 profiled at Almac Diagnostics, Durham, NC, USA).

Statistical Analysis

Weighted Nearest Neighborhood (WNN) Classification Algorithm

This instance-based learner is similar to the commonly employed kNN classifier, employing the notion of a Parzen Window, and Bayesian decision theory. We utilized the non-parametric Spearman rank based correlation (r) as the core similarity metric. This was converted to what is referred to as a correlation distance, by taking 1- r, which results in a dissimilarity measure ranging from 0 (r = l) to 2 (r = -l). This is a necessary step in order to utilize common kernel functions in kernel density estimation (KDE). Additionally, since we desired to only base our assessments on positively correlated samples, an indicator function was incorporated into the kernel function to accomplish this. Using this KDE technique we can generate estimates of the conditional probability of x (the test sample of gene expression data) given a certain class (i.e. node positive tumors, A) within the training data, P(x | A). Furthermore, we can use the known baseline prevalence rates of nodal involvement in this tumor type we examined to generate prior probabilities, P(A). By taking into account the background prevalence rate for nodal involvement, we avoid being biased by the prevalence rate present in the given training data. If pre-test probabilities were not easily identifiable, then one would simply input equal priors, which in this case would have been 50/50 in a two case scenario. With this information in hand, we can then directly calculate the posterior probability of nodal involvement according to Bayes' Theorem, as is depicted and explained in Figures 1 and 2. Gene expression profiling

FFPE archival patient tissues from the Laval and AUO cohorts described above were reviewed by certified pathologists, who selected and harvested representative areas containing 80% or greater tumor cells with a biopsy instrument (33-31 A P/25; Miltex Inc, York, PA, USA) that retrieved a 1.5 mm by 3 mm tissue core from the FFPE block. These samples were used for nucleic acid extraction, verification, amplification, and

hybridization to HG-U133 plus 2.0 arrays (Affymetrix, Santa Clara, CA, USA), with analysis by robust multichip average. Of the samples, 90 of 95 Laval and 185 of 188 AUO provided high quality hybridizations for use in GEM development.

Development of a GEM

The Laval and MSKCC cohorts were used for training and the AUO cohort was used strictly for independent validation. Before model development, two filters were applied. The first removed candidates that did not meet FFPE to fresh frozen high-fidelity criteria to ensure applicability to both fresh frozen and FFPE samples. The identified candidates that had similar differential expression with respect to node status between TUR and the cystectomy were used, an important step to ensure applicability of the test to the precystectomy TUR specimen. This step involved the comparison of gene expression profiles between cystectomy and TUR acquired specimens.

To develop the model, with only gene expression probes that passed the above filters, we calculated a Wilcoxon two-sample test p value for each probe's association with pathological node status (pNl-3 vs pNO) in the Laval cohort (training). We then ranked the probes from lowest to highest p value. Our model selection started with the top five probes most significantly associated with node status in the Laval cohort; we used these probes to predict node status in the MSKCC cohort (also training, optimization) with a weighted nearest-neighbor classification algorithm rooted in Bayesian decision theory and assuming a baseline prevalence (prior probability) of 23% nodal-positive disease at cystectomy. We then calculated the area under the curve (AUC) for its prediction and recorded the value. We repeated these two steps for the top six most significant probes, the top seven most significant probes, the top eight most significant probes, and so on to the top 150 most significant probes (an arbitrary cutoff). We found that of models including the top five to top 150 probes, the top 21 did best on the MSKCC cohort (AUC 0-72; 95% CI 0-58-0-85). The model with the top 21 probes was thus selected as the final model. Hierarchical clustering was used for visualization, whereas significance was assessed at an a of 0 05 in non-parametric t tests or random permutation tests, as appropriate. For identification of transcripts detected with high fidelity by oligonucleotide microarray from analytes prepared from fresh frozen (FF) or archival (formalin fixed, paraffin embedded, FFPE), matched samples, we used 32, matched, diverse human normal and neoplastic tissues preserved by FF and FFPE, profiled at the University of Virginia Microarray Core Facility, Charlottesville, VA, USA (N=22), and at Almac Diagnostics, Durham, NC, USA (N=10). This approach employed a bootstrapping strategy using 1000- fold resampling, implemented in the statistics software, R (www.r-project.org). Such resampling strategies are used in statistics to estimate the stability (variance) of a parameter of interest (here, FF to FFPE correlation of each probe) by testing it across a large numbers sampled subsets (here, 1000). Groups of the pairs of tissues within the cohort were randomly sampled (with replacement); then, the correlation coefficient for each microarray probe from the remaining FFPE samples to the corresponding paired FF samples was calculated. This resulted in 1000 correlation values calculated for each probe's correlation between FFPE and FF, allowing assessment of stability of its detection from tissues preserved by either means. To measure and select probes based on the stability of its correlation in the resampling strategy, we arbitrarily set a threshold for "high fidelity" such that if the lower 2.5th percentile of the 1000 correlation coefficients was greater than > 0 it was designated "high fidelity." By our resampling-based methods, we identified 28684/54675 probes on the Affymetrix HG-U133 plus 2.0 platform as "high- fidelity." Of these, 12402 are shared between the U133 plus 2-0 and HG-U133A platforms.

To discover potential biomarkers of nodal status, we first, in a univariate fashion, calculated the Wilcoxon rank sum P-values for each microarray probe's association with node status in the Laval Cohort. For these analyses, we only employed probes which we had uncovered as high fidelity in terms of their ability to detect transcripts from FFPE versus FF tissues, based on our analysis of 32 matched tissues as described above.

To additionally ensure that any probe selected in the model would also be associated with nodal status in tumor tissues sampled by transurethral resection (TUR) specimens, we excluded any probes that were significantly different (Ρ<0·01) in the Aarhus Cohort (all advanced Stage IV, nodal positive or metastatic) than nodal positive cases of the MSKCC Cohort, which was profiled on the same platform. The Aarhus Cohort, was of transurethral resection (TUR) specimens of stage IV MIBCs, preserved by FF (N=30) from the Aarhus University Hospital, Aarhus, Denmark, from a prior publication by Als et al (3). Candidate node-associated probes from the Laval Cohort surviving these criteria were then sorted by P-value from most to least significant. For model development, we performed a linear, sequential search of the top 5 through 150 probes from this sorted list. We assessed performance of each potential GEM (i.e., the top 5, the top 6, the top 7, etc.) in the setting of the weighted nearest neighborhood (WNN) classifier described above to predict nodal status in the MSKCC Cohort, using the Laval Cohort as the training data. Performance of each potential model was assessed based on the area under the curve from the receiver operating characteristic curve (AUC analysis). From all the GEMs examined (the top 5 through top 150 node associated probes), the best performing GEM exhibited an AUC of 0.72 and was comprised of 21 Affymetrix microarray probes (representing 20 unique genes, two probes representing TOX3, all probes/genes listed in Table 1). We initially evaluated whether selecting one or either of the probes of TOX3 would suffice, however we observed that either way, removal of a probe (or averaging the probes) resulted in significantly detraining the model, with reduction of the AUC in training cohorts to approximately 0.67. Thus, both probes were retained in the GEM.

Table 1. Genes and probes representing 20 unique genes of best performing GEM.

20241 l_at IFI27

222206_s_at NCLN

210514_x_at HLA-G

213495_s_at RRBP1

215485_s_at ICAM1

Note that the Laval Cohort is from FFPE samples while the MSKCC Cohort is from FF samples. Thus, during this phase of development we also examined whether limiting our search to the FFPE/FF high fidelity probes had resulted in superior performance in prediction of the MSKCC Cohort using all of the node-associated probes from the Laval Cohort. Limiting our search to the FFPE/FF high-fidelity probes resulted in significantly increased performance.

Interpretation of GEM results

Because the weighted nearest-neighbour prediction algorithm outputs a prediction score between 0 and 1 , which is difficult to interpret alone, we used another resampling- based cross-validation strategy in the training cohorts (Laval and MSKCC cohorts), to prospectively develop thresholds based on relative risk (RR) of nodal-positive disease to identify patient groups at statistically significantly higher or lower risk. These thresholds were defined on the basis of a priori criteria that with 95% confidence the RR in predicted high-risk and low-risk groups must be increased or decreased, less than 15% of calls in a hypothetical population could be deemed intermediate, and with 95% confidence the RR of nodal positive disease in the intermediate prediction group must include unity. Of the thresholds that met our criteria, optimum high-risk and low-risk cutoffs were selected, equally weighting the RR for the positive-node and negative-node prediction strata.

The classifier developed above produces a continuous prediction score for a given sample to be node positive, naturally ranging from 0 to 1. A rather simple corollary of this is that a call of 0-51 is not the same as a 0 90 for nodal involvement. Intuitively, thresholds needed to be defined so that three discrete possible outputs were generated from the classifier: high risk of nodal positive disease, low risk of nodal positive disease, and intermediate risk of nodal positive disease. The clinical question addressed is the identification of groups of patients at risk of harboring cancers with nodal involvement so that they can be triaged for appropriate therapy. In order to properly address this question the known, baseline probability of nodal involvement in these tumors must be considered. Once known, we can compare any stratification we produce with this baseline rate to assess whether our predictions alter the relative risk of any patient stratum. In simple statistical terms, the relative risk is calculated by taking the ratio of the probability of nodal involvement in two populations. As the baseline probability of nodal involvement, we used the prevalence of 23% reported by earlier (Stein JP, Lieskovsky G, Cote R, Groshen S, Feng AC, Boyd S, et al. Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients. J Clin Oncol. 2001 Feb 1;19(3):666- 75).

To calculate the probability of nodal involvement in a given stratum from our WNN classifier score, we constructed the post-test probability defined below in Figure 3. This can be understood in terms of Bayesian theory or likelihood ratio adjustments. Figure 3 details the construction of post-test probabilities from a Bayesian standpoint. An important intermediate term that is calculated in this approach is the marginal probability, marked with an asterisk in Figure 3. These values can be interpreted as percentages of a hypothetical population. Of particular interest is the percent of the population that will be given inconclusive results, in order to avoid developing a test that does not provide information for an unacceptably large proportion of the population. Additionally, the post- test probability is equivalent to the notion of positive and negative predictive values (PPV/NPV).

With the post-test probability for each of the three possible outputs from the classifier and the known baseline probability of nodal involvement from the literature, we can calculate relative risk. Additionally, we can generate 95% confidence intervals for these relative risk values. We believe that taking into account the baseline prevalence rate in a relative risk term is critical in creating a tool that will guide a clinical decision. To highlight this sentiment, we draw your attention to the 54% post-test probability/PPV in Table 2.

Table 2. Relative Risk Assessments of Nodal Involvement in Training and Independent Cohorts

Results in Training data (MSKCS and Laval Cohorts)

On its own, this number would seem unimpressive. However, this is based on an assumption of a baseline prevalence of disease rate of 50%. If this were the case, then the node positive predicted stratum would have no difference relative to the assumed baseline (50%)). However, when we incorporate the true baseline node positive prevalence of 23% into a relative risk based assessment, it is clear that we have identified a stratum with significantly (P < 0-05) increased risk from the general population; thereby, creating a reason to initiate a clinical intervention.

These thresholds need to be generated in the training data and tested in an independent fashion as well if they are to generalize well. We accomplished this by producing WNN prediction scores for all training data (MSKCC and Laval Cohorts) using the GEM developed above by implementing cross-validation within these two training cohorts. Specifically, over 1000 iterations, 30% of the training data was examined using the remaining 70% as reference in a random selection scheme with replacement. Note that these splits of the training data were always balanced with respect to percent node positive and original dataset as to avoid bias. From these data we applied the following criteria to identify possible threshold values: (1) less than 15% of calls in a hypothetical population could be deemed inconclusive (2) with 95% confidence, the relative risk in the

inconclusive prediction stratum must include unity (3) with 95% confidence the relative risk in positive and negative prediction groupings must differ appropriately from unity. Of the thresholds that met our criteria, the following metric was used to select the final ones: \n{RRpos) - \n{RRneg), which equally weights the relative risk for the positive and negative node prediction strata. Table 2 presents the data from the final threshold settings in the training data (MSKCC and Laval Cohorts).

Independent testing

After development of the GEM and R -based cutoffs, both were tested in the independent AUO cohort. Additionally, multivariate logistic regression was used to test the GEM prediction scores for independent association with nodal status, adjusting for traditional clinical and pathological tumour factors. To test for the association between the GEM genes and survival, a Cox proportional hazards regression model was trained with survival data for both the Laval and MSKCC training cohorts. This model was then tested for its ability to stratify the disease specific outcomes of the AUO cohort and another recently published cohort of muscle-invasive bladder cancers (62 patients) from Chungbuk National University Hospital, South Korea (WJ Kim, EJ Kim and SK Kim et al.,

Predictive value of progression-related gene classifier in primary non-muscle invasive bladder cancer, Mol Cancer 9 (2010), p. 3.), which was profiled on a separate microarray platform.

To test the performance of this GEM in a large, independent validation cohort, we used the AUO Cohort, where the patients were recruited prospectively as part of a clinical trial. Using the 20 gene GEM and WNN classifier as before with both the Laval Cohort and the MSKCC Cohort as reference data, we calculated WNN predication scores for node positive status in the independent AUO Cohort. The classifier performed well in the independent dataset, resulting in an AUC of 0.68.

The significance of this finding was rigorously examined using random

permutation testing by repeating the entire development of the GEM from the beginning through validation on the AUO Cohort 500 times, randomly shuffling the node status labels in the Laval Cohort. We restricted this analysis to the FFPE/FF high-fidelity probes, as before. This procedure was performed 500 times and the distribution of random permutation AUO Cohort AUC values was compared to that of the 20 gene GEM. We found that the probability of obtaining results equal to or greater than the observed due to random chance was less than Ρ<0·002 (i.e., equivalent performance was not observed in a single random permutation, an estimate for the exact P-value). To independently validate the RR system described earlier, we applied the thresholds identified from the training data to the WNN prediction scores of the AUO Cohort. As shown in Table 2, this RR system of stratifying risk into high, equivocal, and low generalized quite well. Both high and low risk strata, based on WNN prediction scores, significantly exhibited relative risk intervals that appropriately deviated from unity, P < 0 05.

Statistical analysis

Microarray data extraction with robust multichip average and bootstrapping high- fidelity analyses were done in the R statistical software, version 2.8.1. Wilcoxon two- sample tests, model development, weighted nearest-neighbour prediction algorithm, receiver operator characteristic testing, cutoff generation, cutoff testing, and logistic regression analyses were done in Mat lab R2009B. AUC curves were plotted and 95% CIs were calculated in Prism 5.0c. Cox proportional hazards regression models were also built and tested in R.

To have clinical value, molecular prediction strategies should show significantly improved prediction compared to standard clinicopathologic parameters, measurable as a difference in the AUC of prediction. In the case of bladder cancer, age, gender, pathologic tumor stage, and, of course, nodal status have been shown to be significantly related to disease specific survival post-cystectomy. Our overall strategy to evaluate the relative contribution of the GEM to the AUC was to train a logistic regression model to predict node status in the two training cohorts (Laval and MSKCC), then use the model to predict pathologic node status in the AUO cohort, with and without including the GEM in the model, to gauge its contribution. For these analyses, we used the GLMFIT function in Matlab (The Mathworks, Natick, MA).

Starting from the Laval and MSKCC datasets, we first found that sex (P=0.27) and age (P=0.70) were not significantly associated with node status. Of the variables available, only pathologic stage, (pT) (P=0.025), was significantly associated with nodal status at cystectomy. Training univariate (pT only) and bivariate (pT and GEM score) logistic regression models to predict node stage on the combined Laval and MSKCC training cohorts, we evaluated the predictive performance of the models on the independent AUO cohort. (For these analyses, we used only the pT3 and pT4 samples of the AUO Cohort, as the smaller number of pTl and pT2 samples were required to be node positive, by definition, under the inclusion and exclusion criteria of the AUO adjuvant clinical trial). We observed that pathologic tumor stage exhibited minimal prediction

performance in the AUO cohort (AUC= 0.52, 95%CI 0.42-0.61), whereas the combination of pathologic tumor stage plus GEM prediction score improved the combined prediction accuracy substantially (AUC=0.65, 95%CI 0.56-0.74). Strikingly, this incremental increase in the AUC persisted despite the use of pathologic staging parameters from cystectomy specimens, much more accurate and precise staging than the clinical stage determined from transurethral resection specimens, which may understage tumors in excess of 50% of cases. Of course, TURBT/clinical stage is the only information which would be available at the time the GEM would be implemented in practice, highlighting the stringency of this comparison. Perhaps a more important test is for independence of association between the GEM prediction scores and all other clinicopathologic variables. Using multivariate logistic regression including age, gender, pathologic stage, lymphoascular space invasion (LVSI) at cystectomy, and the GEM score, we found that in the AUO test cohort, the GEM score was significantly independently associated with nodal status P=0.019. Again, independence of the GEM beyond these variables is a stringent test, as not only would pathologic stage not be available at the time the GEM would be applied in clinical practice (as above), nor would definitive LVSI, which also may differ substantially between transurethral resection and cysectomy specimens.

Survival Analysis as a function of pathological node status or GEM prediction

Additional Dataset

In addition to the MSKCC, Laval, and AUO cohorts, we also used the Chungbuk National University Hospital (CNUH) cohort that contains 62 primary bladder cancer samples from patients with muscle invasive bladder cancer that were treated with cystectomy as in the MSKCC, Laval, and AUO cohorts. These were profiled on the Illumina human-6 v2.0 expression beadchip. These data are available at NCBI with GEO Accession# GSE13507.

Data Processing, Merging, and Normalization of Data Sets

We used the processed data that is publicly available for all data sets, then performed Z-score normalization so that each sample has mean 0 and variance 1. The 20 gene node signature corresponds to 21 probes whose expression profiles are available in the MSKCC, Laval, and AUO Cohorts, which all use Affymetrix microarrays. For the CNUH Cohort, which was profiled on an Illumina platform, probes were identified by either matching their gene symbols or GenBank accession numbers to those of the 20 gene signature. Out of the 20 genes, we identified 19 probes/genes on the Illumina platform (all but TOX3, which has two Affymetrix probes).

Survival Analysis Methods

The MSKCC Cohort contains only data for overall survival, whereas the remaining cohorts contain data for disease specific survival data. Because we were interested in predicting disease specific survival and needed a sufficiently large dataset for training, we censured the MSKCC survival data at 60 months since up to this point the disease specific and overall survival rates are similar. This allowed us to combine it with the Laval Cohort for the training. The risk prediction model is based on a Cox proportional hazards model. An initial Cox proportional hazards model is used to determine a patient's risk score and to assign each individual to a low or high risk group. Risk scores were calculated by using a linear combination of products of the Cox coefficients (β) and corresponding probe expression values. If a corresponding probe expression value is not available, as is the case with the two TOX3 probes in the CNUH Cohort, the coefficient is not included in the model. The median risk score was selected as a high risk threshold; individuals with scores above the threshold are classified as high risk and individuals with scores below the threshold are classified as low risk. After identifying low and high risks groups, a second Cox proportional hazards model is fit to obtain a hazard ratio and a p-value.

When building our model, we used training data sets of MSKCC and Laval to find the Cox coefficients, which are then used to calculate risk scores in the validation data sets of AUO and CNUH. Importantly, none of the validation data are used in the training, so that the validation is completely independent of the training process. We also assessed the accuracy of the model to predict survival in the training data set by using a leave one out classification approach as follows: We leave out the first sample, fit the initial Cox model to the remaining samples, find the high risk threshold, calculate the first sample's risk score using the coefficients of the fitted model, and assign the individual to the low or high risk group. We repeat this process for each additional sample, so that all individuals are assigned to low or high risk groups. Note that for each individual, that individual's survival information and gene signature is not used in the training of the classifier that will assign the patient to low/high risk group. Then we fit the secondary Cox model to all risk groups to obtain a hazard ratio and p-value. Results

Table 3 shows the characteristics of the three independent cohorts of patients who underwent cystectomy and lymphadenectomy for bladder cancer and were used to develop and validate the GEM for nodal prediction. Tissue preparation remains a substantial challenge in the development of molecular diagnostics, particularly if the former are to be based on RNA species. To broaden the use of an eventual clinical test based on our work, we identified RNA transcripts detected with high fidelity independently of sample preservation (fresh freezing or FFPE), before development of our GEM. To that end, we used 32 paired fresh frozen and FFPE tissues profiled by microarrays to identify probes which were further refined to ensure they are expressed in TUR-acquired tumours as well as cystectomies. These probes were used for GEM development on the Laval and MSKCC cohorts. Figure 4 shows the development of the GEM predictor of pathological nodal status at cystectomy. Figure 4 (A) shows high-fidelity transcript discovery: for ease of implementation, we first used Affymetrix HG-U133 plus 2-0 microrarray data for a cohort of 32 paired tissues that had been preserved by FFPE and FF to develop a set of probesets detected with high fidelity by either means of tissue preservation. After 1000-times bootstrapping the correlation of probes across the paired tissues, we selected probes that maintained positive correlation at the 2 -5th percentile. 12 402 of the high-fidelity probes were common to the U133 plus 2-0 platform and the U133A platforms, and after ensuring that these genes were not expressed differently in TUR specimens they were used for GEM development. Figure 4 (B) shows the model development with training cohorts: the GEM was developed by linear forward searching of probes most significantly correlated with nodal status in the Laval cohort and predicting nodal status on the MSKCC cohort, with a WNN predictor based on the correlation of samples. The maximum AUC for discrimination was attained at 21 probes, representing 20 transcripts, and this model was selected as the final model on the basis of training data alone. Figure 4 (C) shows the 20- gene model: univariate differences for each probe's expression across nodal status are shown. The normalised expression in pNl-3 samples in the Laval cohort is plotted in terms of SDs of difference compared with the pNO samples of the Laval cohort (error bars represent 95% CIs). In Figure 4, FF=fresh freezing. FFPE=formalin fixation, paraffin embedding. pN0=node negative. pNl-3=node positive. WNN=weighted nearest neighbour. ROC=receiver operating characteristic. AUC=area under the curve.

TUR=transurethral resection. GEM=gene expression model. Table 3. Clinical and pathological characteristics of the patient cohorts

Data are median (range) or n (%).

An unpublished cohort of 90 archival specimens that we profiled on the HG-U133 plus 2-0 microarray platform.

^† A subset of 66 muscle-invasive cancer samples from a recent expression profiling study by Sanchez-Carbayo and colleagues where frozen tumour samples were profiled on the HG-U133A microarray platform.

* 185 samples from a previous randomised controlled trial (AUO-AB-05/95) reported by Lehmann and colleagues archival samples of which we profiled on the HG-U133 plus 2-0 microarray platform. About 20-40% of patients with clinically organ-confined muscle-invasive bladder cancer have node-positive disease at cystectomy (JP Stein, G Lieskovsky and R Cote et al. , Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1,054 patients, J Clin Oncol 19 (2001), pp. 666-675). We used microarray data from two (Laval and MSKCC) cohorts of cystectomy patients to develop a GEM predictive of nodal status (see Figure 4). We identified an optimum GEM consisting of probes for the top 20 genes (21 microarray probes) most significantly associated with node status in the Laval cohort, based on their perfor mance predicting the nodal status of the MSKCC cohort (AUC 0.72; 95% CI 0.58-0.85). Figure 1 shows the 21 probes and their univariate association with pathological nodal status.

To make our findings clinically useful, we developed high-risk and low-risk cutoffs of these continuous scores as has been done for other molecular tests (S Paik, S Shak and G Tang et al., A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer, N EnglJ Med 351 (2004), pp. 2817-2826.). These cutoffs were based on the RR of nodal-positive disease at cystectomy, and developed with a cross- validation approach in both training cohorts. Cutoffs were designed such that patients would be identified by the GEM as high risk, low risk, or intermediate risk. Optimum cutoffs were chosen on the basis of training data only, and yielded a low-risk group with an RR of 0.47 (95% CI 0.31-0.70) and a high-risk group with an RR of 2.25 (1.18-4.28; see Figure 5). Figure 5 (A) shows a cross-validation approach applied to both Laval and MSKCC cohorts, cutoffs were developed from the training cohorts to identify patient risk strata with significantly high or low relative risk of nodal-positive disease. Comparison of high fidelity versus all transcript use in GEM-based prediction of node-positive disease. Figure 5 (B) shows the box-whisker plot (boxes, median and IQR; whiskers 5th and 95th percentiles) of distributions of area under the ROC curve (AUC) performance of models based on the top five to 150 FFPE to FF high-fidelity node-associated genes compared with models based on the top five to 150 of all genes derived from the Laval cohort (FFPE). Predictions were made from the Laval cohort (FFPE) on the MSKCC cohort (FF). The distributions show a highly significant trend of superiority of the models based on high-fidelity genes, supporting the usefulness of this method. The top performing models are also plotted (solid triangles and AUCs for these shown). In Figure 5, RR=relative risk. WNN=weighted nearest neighbor. pNl-3=node positive. ROC=receiver operating characteristic. GEM=gene expression model. FFPE=formalin fixation, paraffin

embedding. FF=fresh freezing. An advantage of using both specimens preserved with FFPE (Laval) and fresh freezing (MSKCC) for training is that this approach allows the examination of the extent to which the use of the FFPE to fresh frozen high-fidelity transcripts facilitated the generalization of the model across the two differently preserved groups of specimens. We assessed the contribution of this technique to the efficacy of intercohort FFPE to fresh frozen prediction by examining the performance of the top five to 150 probes most associated with nodal status in the Laval cohort (FFPE) with and without preselection of only high-fidelity probesets. We noted that high-fidelity probesets were better at predicting nodal status in the MSKCC cohort than the Laval cohort, showing that this preselection method contributes to better intercohort prediction (ρ<0·0001; Figure 5).

Ideally, to establish efficacy and avoid bias, predictive biomarkers should be assessed on specimens collected prospectively. We did this by testing the 20-gene model on tissues from patients enrolled in the AUO-AB 05/95 trial (AUO cohort). We noted a significant ability of the predictor to discriminate between pathologically nodal-positive and nodal-negative tumors (AUC 0.67, 95% CI 0.60-0.75; ρ<0·0001; Figure 6). We also found the risk stratification scheme we developed generalized to this independent cohort; wherein high-risk and low-risk groups with increased (RR 1.74, 95% CI 1.03-2.93) or decreased (0.70, 0.51-0.96) RR of nodal-positive disease were identified. As shown in Figure 6, receiver operating characteristic analysis of the predictions outputted by the 20- gene model showed highly significant ability to discriminate between nodal-negative and nodal-positive tumors in the AUO cohort (AUC=0.67, 95% CI 0.60-0.75; Figure 6A). Testing of the performance of the relative-risk cutoffs on the independent test cohort found that the cutoffs developed from training data again identified groups within the AUO cohort with significantly increased or decreased risk of nodal-positive disease (Figure 6B).

To have clinical value, molecular prediction strategies should show significantly improved prediction compared with standard clinicopathological variables. Starting from the Laval and MSKCC training cohorts, we found that of sex (p=0.27), age (p=0.70), and pathological stage (p=0.025), only pathological tumor stage was significantly associated with nodal status at cystectomy. We compared the node status prediction performance of pathological tumor stage alone versus pathological tumor stage plus GEM prediction in the independent AUO cohort with univariate and bivariate logistic-regression models built only with the Laval and MSKCC training cohorts. We noted that pathological tumor stage provided minimal prediction performance in relevant cases of the AUO cohort (AUC 0.52, 95% CI 0.42-0.61), whereas the combination of pathological tumor stage plus GEM prediction score improved the combined prediction accuracy substantially (0.65, 0.56- 0.74). This incremental increase in the AUC persisted despite the use of pathological staging parameters from cystectomy specimens— much more accurate and precise staging than the clinical stage established from TUR specimens, which can understage tumors in more than 50% of cases.

We also tested the GEM prediction for independence of association from other clinicopathological variables. With multivariate logistic regression including age, sex, pathological stage, lymphovascular space invasion (LVSI) at cystectomy, and the GEM score, we found that in the AUO test cohort the GEM score was significantly

independently associated with nodal status (See Table 4). Independence of the GEM beyond these variables is a very stringent test, because pathological stage would not be available at the time GEM assessments would be applied in clinical practice, nor would definitive LVSI, which also can differ substantially between TUR and cystectomy specimens.

Table 4. Multivariate logistic regression of clinicopathological parameters of GEM in the test cohort.

GEM=gene expression model.

Coefficient and p value for independent association of indicated parameter with node status in included AUO test cohort cases.

^ Transurethral resection of bladder tumour clinical staging data were not available for this cohort.

Given the strong relation of pathological node status with disease-free survival noted in several large series (JP Stein, G Lieskovsky and R Cote et al., Radical cystectomy in the treatment of invasive bladder cancer: long-term results in 1 ,054 patients, J Clin Oncol 19 (2001), pp. 666-675; BH Bochner, MW Kattan and KC Vora, Postoperative nomogram predicting risk of recurrence after radical cystectomy for bladder cancer, J Clin Oncol 24 (2006), pp. 3967-3972), the corollary question arises as to whether the 20 node- related GEM genes could also model patient survival after cystectomy. Such an observation would further support the relevance of these genes by relating them to another key prognostic variable. We used multivariate Cox proportional hazards regression on the 20 GEM genes to model survival, again trained on the Laval and MSKCC training cohorts. We found that the model genes could stratify survival in the AUO cohort (p=0.048), as well as another cohort recently profiled with a separate gene expression profiling technology (p=0.027), supportive of the generality of their association with poor prognosis.

This research example demonstrates that the present inventors have developed a test that can predict a powerful determinant of prognosis after cystectomy: node-positive disease, and that such molecular intelligence, for which no other molecular marker exists, provides a technique that allows more effective and frequent use of neoadjuvant chemotherapy by selection of high-risk patients and avoidance of overtreatment of low- risk patients. Our use of multiple cohorts from different institutions on different continents, including a prospective, multicentre cohort from a randomized controlled trial supports the generalizability of our approach across very different populations. Our study, to our knowledge, is the largest genome-wide profiling study of bladder cancer so far (275 patients), and the first report profiling or generating a biomarker from FFPE bladder tissues. Lastly, we hope that identification of high-fidelity transcripts preserved from tissues processed either by FFPE or fresh freezing will be of use to other research groups in the development of gene expression-based molecular bio markers on FFPE material given its widespread use.

Interpretation

Despite the preponderance of evidence in its favor, neoadjuvant chemotherapy, which results in a small, significant increase in survival in bladder cancer, is used infrequently because of the inability to risk stratify patients before definitive surgical staging. Our study proves the principle that molecular staging before surgery can change the way we view urothelial cancer management and practice by assessing, a priori, staging parameters before surgery, after which only adjuvant chemotherapy (unsupported by any level 1 evidence) remains an option. Most importantly, our model showed the ability to significantly predict node- positive and node-negative patients when tested on prospectively collected tissues from the AUO cohort, with a level of performance that is similar to that of gene expression- based molecular predictors in clinical use for other tumor types. Notably, our assessment was independent of classic clinical and pathological predictive factors, the latter being the so-called gold standard variables in establishing prognosis after cystectomy. To allow the use of our technique in clinical practice, we developed, on the basis of training data alone, a system of two cutoffs in the predictor's scores, designed to identify groups with significantly increased or decreased risk of nodal involvement, relative to a baseline rate of about 25% node-positive disease, taken from a published series of more than 1000 patients. These cutoffs again performed as intended when applied to the independent AUO cohort. The reported AUC of 0.67 allowed us to develop a system of cutoffs, that identified clinically actionable populations, with potential clinical use beyond the raw overall performance curve.

However, our findings come with caveats. Our GEM was assessed on samples obtained at cystectomy rather than TUR, the latter being the technique used to obtain diagnostic tissue before neoadjuvant chemotherapy and upon which the GEM would be applied in clinical practice. Although studies have successfully validated GEMs developed from cohorts containing both cystectomies and TUR specimens or TUR specimens alone, we went further to ensure that our GEM was suitable for use on TUR specimens. First, we ensured that the genes from which the model was built had similar expression patterns in both TUR and cystectomy specimens. Second, the tissue we used for gene expression sampling was harvested with a biopsy instrument that retrieves a 1 - 5 mm tissue core from an FFPE block containing part of the bladder tumor. Arguably, such minute sampling is equally prone to intratumoral variability of gene expression whether it is done from a cystectomy specimen or the diagnostic TUR, since in both cases the core sample that would be used for profiling is very small compared with the overall tumor bulk.

Further, the optimum extent of intraoperative lymphadenectomy remains controversial for bladder cancer. Given the large numbers of specimens in our study and the international distribution of the patients they were derived from, it is likely that differences in the node dissection were present in the training sets compared with the validation set. However, despite this, it is notable that when the model was assessed against AUO cohort patients enrolled in a phase 3 study explicitly mandating a complete node dissection for enrolment, the GEM exhibits significant performance, independent of classic clinical and pathological variables, and similar in performance to molecular predictors used for other tumor types. Additionally, patients found to have pathologically negative nodes despite a strong GEM prediction to the contrary, might indeed harbor metastasis not detected by conventional pathological node assessment. We are considering testing this issue directly in future studies, in cases where archival nodal tissues remain by use of sensitive molecular methods for cancer cell detection in pathologically negative nodes of tumors suggested to be high risk by our GEM assay.

As newer imaging modalities find a role in clinical practice, future studies should compare their performance with that of our model. Indeed, future multivariate models could be built and validated with GEMs, and new imaging technologies could be coupled with pathological findings on TUR specimens (such as LVSI) to provide superior performance than any individual technique. We also see the opportunity to use data obtained from microarray profiling of tumor tissue not only for prediction of node status but also for prediction of response to chemotherapy and selection of the most appropriate drugs.

In summary, our nodal prediction model is an important clinical advance in the personalization of bladder cancer treatment by allowing selection of those patients at high risk for recurrence and thus likely the best candidates for neoadjuvant chemotherapy. At present such information is lacking, leading to a gap between evidence and practice. It is our hope that assays like ours could reduce this gap.

The foregoing examples of the present invention have been presented for purposes of illustration and description. Furthermore, these examples are not intended to limit the invention to the form disclosed herein. Consequently, variations and modifications commensurate with the teachings of the description of the invention, and the skill or knowledge of the relevant art, are within the scope of the present invention. The specific embodiments described in the examples provided herein are intended to further explain the best mode known for practicing the invention and to enable others skilled in the art to utilize the invention in such, or other, embodiments and with various modifications required by the particular applications or uses of the present invention. It is intended that the appended claims be construed to include alternative embodiments to the extent permitted by the prior art.

Claims

What is claimed is:

1. A method for identifying a bladder cancer patient predicted to respond to the

administration of neoadjuvant chemotherapy comprising:

detecting in a sample of bladder tumor cells from a patient, a level of gene expression of a marker gene or plurality of marker genes selected from the group consisting of:

i) a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCH1, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGA10, MT1E, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBPl, ICAMl, or homologs or variants thereof;

ii) polypeptides encoded by the marker genes of i);

iii) fragments of polypeptides of ii); and

iv) a polynucleotide which is fully complementary to at least a portion of a marker gene of i) - iii);

wherein the expression of the plurality of markers is indicative of whether the bladder cancer patient will benefit from the administration of neoadjuvant chemotherapy.

2. The method of claim 1, wherein the chemotherapy comprises at least one platinum- based chemotherapeutic.

3. The method of claim 1, wherein the neoadjuvant chemotherapy further comprises the administration of an additional anti-cancer agent.

4. The method of claim 3, wherein the additional anti-cancer agent is radiation.

5. The method of claim 1, wherein the bladder tumor cells are fixed, paraffin- embedded, fresh, or frozen.

6. The method of claim 1, wherein the bladder tumor cells are formalin fixed, paraffin-embedded (FFPE).

7. The method of claim 1, wherein the bladder tumor cells are collected from cystectomy specimen.

8. The method of claim 1 , wherein the bladder tumor cells are obtained by

transurethral resection (TUR).

9. The method of claim 1, wherein the detecting comprises microarray analysis.

10. The method of claim 1, wherein the detecting is done by immunohistochemistry.

11. The method of claim 1 , wherein the detecting comprises determining the level of an RNA transcript of a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCH1, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGA10, MT1E, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBP1, ICAM1, or homologs or variants thereof.

12. The method of claim 1 , further comprising creating a report indicating the likelihood of benefit to the bladder cancer patient of neoadjuvant chemotherapy.

13. The method of claim 1 , further comprising calculating a quantitative score indicating the likelihood of lymph node metastasis of the bladder cancer.

14. The method of claim 13, wherein the quantitative score is calculated using the relative risk from a Weighted Nearest Neighborhood (WNN) classification algorithm.

15. The method of claim 1, wherein the genes detected share 100% sequence identity with the corresponding marker genes in i).

16. The method of claim 1, wherein a level of at least one of the plurality of markers is determined and compared to a standard level or reference range.

17. The method of claim 1, wherein the standard level or reference range is determined according to a statistical procedure for risk prediction.

18. The method of claim 1 , wherein the presence of the marker is determined by detecting the presence of a polypeptide.

19. The method of claim 6, wherein the method further comprises detecting the presence of the polypeptide using a reagent that specifically binds to the polypeptide or a fragment thereof.

20. The method of claim 7, wherein the reagent is selected from the group consisting of an antibody, an antibody derivative, and an antibody fragment.

21. The method of claim 1 , wherein the presence of the marker is determined by obtaining RNA from the bladder tumor cells; generating cDNA from the RNA; amplifying the cDNA with probes or primers for marker genes; obtaining from the amplified cDNA the expression levels of the genes or gene expression products in the sample.

22. The method of Claim 1, further comprising:

a) comparing the expression level of the marker gene or plurality of marker genes in the tumor cell sample to a control level of the marker gene(s) selected from the group consisting of:

i) a control level of the marker gene that has been correlated with beneficial response to the administration of neoadjuvant chemotherapy; and

ii) a control level of the marker that has been correlated with lack of beneficial response to neoadjuvant chemotherapy; and b) selecting the patient as being predicted to respond to neoadjuvant chemotherapy, if the expression level of the marker gene in the patient's bladder tumor cells is statistically similar to, or greater than, the control level of expression of the marker gene that has been correlated with beneficial response to the administration of neoadjuvant chemotherapy, or

c) selecting the patient as being predicted to not respond to neoadjuvant chemotherapy, if the level of the marker gene in the patient's bladder tumor cells is statistically less than the control level of the marker gene that has been correlated with beneficial response to the administration of a to neoadjuvant chemotherapy.

23. The method of Claim 1, further comprising:

a) comparing the expression level of the marker gene, or plurality of marker

genes, in the tumor cell sample to a level of the marker gene(s) in a second patient known to be node-negative (organ confined) bladder cancer, and, b) selecting the patient as being predicted to benefit from the administration of neoadjuvant chemotherapy, if the expression level of the marker gene in the patient's bladder tumor cells is greater than the level of expression of the marker gene(s) in the second patient, or,

c) selecting the patient as being predicted to not benefit from the administration of neoadjuvant chemotherapy, if the level of the marker gene in the patient's bladder tumor cells is less than or equal to the level of expression of the marker gene(s) in the second patient.

24. A method for identifying a bladder cancer patient predicted to suffer recurrence of the cancer following cystectomy comprising:

detecting in a sample of bladder tumor cells from a patient, a level of gene expression of a marker gene or plurality of marker genes selected from the group consisting of:

i) a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLCl 1A2, FAM36A, LIMCHl, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBPl, ICAMl, or homologs or variants thereof;

ii) polypeptides encoded by the marker genes of i);

iii) fragments of polypeptides of ii); and

iv) a polynucleotide which is fully complementary to at least a portion of a marker gene of i) - iii); wherein the expression of the plurality of markers is indicative of whether the cancer is likely to recur in the patient following cystectomy of the patient's bladder.

25. The method of claim 24, wherein the bladder tumor cells are fixed, paraffin- embedded, fresh, or frozen.

26. The method of claim 24, wherein the bladder tumor cells are formalin fixed, paraffin-embedded (FFPE).

27. The method of claim 24, wherein the bladder tumor cells are collected from cystectomy specimen.

28. The method of claim 24, wherein the bladder tumor cells are obtained by transurethral resection (TUR).

29. The method of claim 24, wherein the detecting comprises microarray analysis.

30. The method of claim 24, wherein the detecting is done by immunohistochemistry.

31. The method of claim 24, wherein the detecting comprises determining the level of an RNA transcript of a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCHl, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBP1, ICAM1, or homologs or variants thereof.

32. The method of claim 24, further comprising creating a report indicating the likelihood of the cancer to recur in the patient following cystectomy of the patient's bladder.

33. The method of claim 24, further comprising calculating a quantitative score indicating the likelihood of lymph node metastasis of the bladder cancer.

34. The method of claim 33, wherein the quantitative score is calculated using the relative risk from a Weighted Nearest Neighborhood (WNN) classification algorithm.

35. The method of claim 24, wherein the genes detected share 100% sequence identity with the corresponding marker genes in i).

36. The method of claim 24, wherein a level of at least one of the plurality of markers is determined and compared to a standard level or reference range.

37. The method of claim 36, wherein the standard level or reference range is determined according to a statistical procedure for risk prediction.

38. The method of claim 24, wherein the presence of the marker is determined by detecting the presence of a polypeptide.

39. The method of claim 38, wherein the method further comprises detecting the presence of the polypeptide using a reagent that specifically binds to the polypeptide or a fragment thereof.

40. The method of claim 39, wherein the reagent is selected from the group consisting of an antibody, an antibody derivative, and an antibody fragment.

41. The method of claim 24, wherein the presence of the marker is determined by obtaining RNA from the bladder tumor cells; generating cDNA from the RNA; amplifying the cDNA with probes or primers for marker genes; obtaining from the amplified cDNA the expression levels of the genes or gene expression products in the sample.

42. The method of Claim 24, further comprising:

a) comparing the expression level of the marker gene or plurality of marker genes in the tumor cell sample to a control level of the marker gene(s) selected from the group consisting of:

iii) a control level of the marker gene that has been correlated with recurrence of the cancer following cystectomy; and

iv) a control level of the marker that has been correlated with lack of recurrence of the cancer following cystectomy; and

b) selecting the patient as being predicted to be diagnosed with a recurrence of the cancer following cystectomy, if the expression level of the marker gene in the patient's bladder tumor cells is statistically similar to, or greater than, the control level of expression of the marker gene that has been correlated with recurrence of the cancer following cystectomy, or

c) selecting the patient as being predicted to be diagnosed with a recurrence of the cancer following cystectomy, if the level of the marker gene in the patient's bladder tumor cells is statistically less than the control level of the marker gene that has been correlated with recurrence of the cancer following cystectomy.

43. The method of Claim 24, further comprising:

a) comparing the expression level of the marker gene, or plurality of marker

genes, in the tumor cell sample to a level of the marker gene(s) in a second patient known to have been diagnosed with a recurrence of the cancer following cystectomy, and,

b) selecting the patient as being predicted to be diagnosed with a recurrence of the cancer following cystectomy, if the expression level of the marker gene in the patient's bladder tumor cells is greater than the level of expression of the marker gene(s) in the second patient, or,

c) selecting the patient as being predicted to not be diagnosed with a recurrence of the cancer following cystectomy, if the level of the marker gene in the patient's bladder tumor cells is less than or equal to the level of expression of the marker gene(s) in the second patient.

44. A method for monitoring the progression of bladder cancer in a subject, the

method comprising:

a) measuring the expression level of a plurality of marker in a first biological sample obtained from the subject, wherein the plurality of markers comprise a plurality of markers selected from the group consisting of:

i) a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCH1, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGA10, MT1E, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBPl, ICAMl, or homo logs or variants thereof;

ii) polypeptides encoded by the marker genes of i);

iii) fragments of polypeptides of ii); and

iv) a polynucleotide which is fully complementary to at least a portion of a marker gene of i) - iii);

b) measuring the expression level of the plurality of markers in a second biological sample obtained from the subject; and

c) comparing the expression level of the marker measured in the first sample with the level of the marker measured in the second sample.

45. The method of claim 44, wherein the genes detected share 100% sequence identity with the corresponding marker gene in i)-iv).

46. The method of claim 44, wherein the first biological sample from the subject is obtained at a time to, and the second biological sample from the subject is obtained at a later time ti.

47. The method of claim 46, wherein the first biological sample and the second

biological sample are obtained from the subject are obtained more than once over a range of times.

48. A method for selecting a therapeutic treatment for a bladder cancer patient

comprising: measuring the level of a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2,

FAM36A, LIMCH1, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, RRBP1, ICAMl, or homo logs or variants thereof from a tumor, tissue, or cell sample from a patient;

correlating the level of the biomarker(s) to a biomarker profile for node positive bladder cancer and/or node negative bladder cancer; and

selecting a therapeutic treatment selected from the group consisting of neoadjuvant therapy and cystectomy, based on the comparison of the biomarker profile from said tumor or tissue sample and the biomarker profile for benefit from neoadjuvant therapy.

49. The method of Claim 48, wherein the neoadjuvant therapy is chemotherapy.

50. The method of claim 48, further comprising creating a report indicating the level of the biomarker(s) relative to a biomarker profile for node positive bladder cancer and/or node negative bladder cancer.

51. The method of claim 48, further comprising creating a report identifying the

relative risk of the patient for recurrence of bladder cancer following cystectomy.

52. The method of claim 48, further comprising creating a report identifying the

patient as high risk or low risk.

53 A method for predicting a predisposition to recurrence of cancer in a bladder

cancer patient following cystectomy, comprising:

detecting the level of expression of a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCH1, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA- G, RPvBPl, ICAMl, or homo logs or variants thereof from a tumor, tissue, or cell sample from a patient;

wherein the elevate expression level of the marker is indicative of a recurrence of cancer in the bladder cancer patient following cystectomy.

54. An assay system for predicting bladder cancer patient response or outcome to neoadjuvant chemotherapy comprising a means to detect:

the expression of a marker gene or plurality of marker genes selected from the group consisting of: i) a marker gene having at least 95% sequence identity with a sequence selected from the group consisting of TOX3, SLC11A2, FAM36A, LIMCHl, RAB15, AVL9, PCMTD2, PTHLH, DPP4, PCDHGAIO, MTIE, MAP4K4, SLC16A1, BST2, MMP14, IFI27, NCLN, HLA-G, R BP1, ICAM1, or homologs or variants thereof;

ii) polypeptides encoded by the marker genes of i);

iii) fragments of polypeptides of ii); and

iv) a polynucleotide which is fully complementary to at least a portion of a marker gene of i) - iii).

55. The assay system of claim 54, wherein the means to detect comprises nucleic acid probes comprising at least 10 to 50 contiguous nucleic acids of the marker gene(s), or complementary nucleic acid sequences thereof.

56. The assay system of claim 54, wherein the means to detect comprises binding ligands that specifically detect polypeptides encoded by the marker genes.

57. The assay system of claim 54, wherein the genes detected share 100% sequence identity with the corresponding marker gene in i).

58. The assay system of claim 54, wherein the means to detect comprises at least one of nucleic acid probes and binding ligands disposed on an assay surface.

59. The assay system of claim 58, wherein the assay surface comprises a chip, array, or fluidity card.

60. The assay system of claim 58, wherein the probes comprise complementary nucleic acid sequences to at least 10 to 50 nucleic acid sequences of the marker genes.

61. The assay system of claim 58, wherein the binding ligands comprise antibodies or binding fragments thereof.

62. The assay system of claim 54, further comprising: a control selected from the group consisting of:

a) information containing a predetermined control level of the marker gene that has been correlated with response or outcome to neoadjuvant

chemotherapy in a bladder cancer patient; and

b) information containing a predetermined control level of the marker gene that has been correlated with a lack of response or outcome to neoadjuvant chemotherapy in a bladder cancer patient.