Bladder cancer biomarkers and methods of use

Abstract

Compositions, kits, and methods for the prognosis of bladder cancer in a subject are provided by detecting in tumor tissue a combination of biomarkers consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

Description

[0001] Cross-reference to related applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 346,468, filed on May 27, 2022, and U.S. Provisional Application No. 63 / 483,679, filed on February 7, 2023, the contents of each of which are hereby incorporated by reference in their entirety.

[0003] Reference to a sequence listing submitted as a compliant XML 1.0 formatted file (.xml)

[0004] In accordance with the EFS-Web legal framework and 37 CFR §§ 1.821-825 (see MPEP § 2442.03 (a)), Rule 30 EPC and § 11 Pat V, an electronic sequence listing in accordance with WIPO Standard ST.26 is submitted as an XML 1.0 format file (titled "300047-005977_Sequence_Listing.xml", created on May 26, 2023, and 26,185 bytes in size) simultaneously with this application, and the entire contents of the sequence listing are incorporated herein by reference. For the avoidance of doubt, if there is a difference between the sequence mentioned in the specification and the electronic sequence listing, the sequence in the specification shall be deemed to be the correct sequence. Technical Field

[0005] The present invention relates to compositions, kits and methods for cancer detection, and in particular to such compositions, kits and methods in bladder cancer prognosis. In addition, such compositions, kits and methods can be used as aids in pathological assessment. Background Art

[0006] Bladder cancer is one of the top five most common malignancies worldwide. In the United States alone, it is estimated that there will be 83,730 new cases of bladder cancer and 17,200 deaths in 2021. Siegel et al. (2021) CA Cancer Journal (CA Cancer J Clin)》 71(1):7-33. Since 2000, the absolute number of bladder cancer cases and deaths has increased by 57% and 41%, respectively. Siegel et al. (2021) CA Journal of Cancer for Clinicians 71(1):7-33; Greenlee et al. (2000) CA Journal of Cancer for Clinicians 50(1):7-33. When detected early (ie, NMIBC or stage 1), the 5-year survival rate is approximately 94%, compared to up to 50% when the disease is noted as MIBC (stage 2) and less than 20% when the disease is metastatic (stages 3 and 4). Brausi et al. Journal of Urology (J Urol.)(2011)186(6):2158-67;Stenzl et al. European Urology (Eur Urol.) (2011) 59(6):1009-18; Calabro et al. Curr Opin Support Palliat Care. (2012) 6(3):304-9; Sternberg et al. Journal of Clinical Oncology (J Clin Oncol.) (2001)19(10):2638-46.

[0007] Oncologists have several treatment options available to them, including surgery, radiation, chemotherapy drugs, and immuno-oncology agents. The best chance of a good treatment outcome requires that patients be assigned to the best available cancer treatment, and that this assignment occur as soon as possible after diagnosis.

[0008] There is a need in the art for early, rapid detection of cancer to improve clinical outcomes. Summary of the invention

[0009] In one embodiment, a method for predicting the likelihood of long-term survival in a bladder cancer patient may comprise: (a) obtaining a biological sample from the patient; (b) isolating mRNA from the biological sample; (c) determining the level of mRNA of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA in the biological sample; (d) normalizing the mRNA levels relative to the level of at least one reference mRNA transcript in the sample to provide normalized mRNA levels of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. mRNA levels; (e) comparing the normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA levels with the normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA levels in a reference bladder tumor sample; and (f) predicting the likelihood of long-term survival without recurrence of bladder cancer, wherein increased levels of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA indicate a decreased likelihood of long-term survival without recurrence of bladder cancer. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0010] In one embodiment, a method for detecting bladder cancer biomarkers may comprise: (a) obtaining a biological sample from a patient; (b) isolating RNA from the biological sample; and (c) determining the level of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA mRNA in the biological sample.

[0011] In one embodiment, a method for classifying test data, the test data comprising RNA expression data, the method may include: (a) using at least one processor to access an electronically stored set of training data vectors, each training data vector representing an individual cancer patient and comprising RNA expression data for the corresponding cancer patient, each training data vector further comprising a classification of expression levels of a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; (b) using the electronically stored set of training data vectors to train an electronic representation of a classification system; (c) receiving test data comprising RNA expression data at the at least one processor; (d) using the at least one processor to evaluate the test data using the electronic representation of the classification system; and (e) outputting a classification of the test data regarding the likelihood of long-term survival without recurrence of bladder cancer based on the evaluating step. The biomarkers may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0012] In one embodiment, the classification system can be AdaBoost, artificial neural network (ANN) learning algorithm, Bayesian belief network, Bayesian classifier, Bayesian neural network, boosted tree, case-based reasoning, classification tree, convolutional neural network, decision tree, deep learning, elastic net, fully convolutional network (FCN), genetic algorithm, gradient boosted tree, k-nearest neighbor classifier, LASSO, linear classifier, naive Bayes, neural network, penalized logistic regression, random forest, ridge regression, support vector machine or a set thereof. The classification system can be a set of classification systems.

[0013] In one embodiment, the mRNA level can be determined by microarray analysis, RNAseq, RT-PCR, RT-qPCR, quantitative PCR (qPCR), Northern blot analysis, dot blot, Southern blot analysis, RNA sequencing, fluorescence in situ hybridization (FISH), or a combination thereof. The mRNA can be determined by quantitative PCR (qPCR). The microarray can contain cDNAs of biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof. The microarray can contain cDNAs that can be fixed to a substrate. The biomarkers can consist of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0014] In one embodiment, the determining step may use a primer selected from the group consisting of: SEQ ID NO: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 or a combination thereof. The determining step may use a primer pair selected from the group consisting of: SEQ ID NO: 1 and 2; 3 and 4; 5 and 6; 7 and 8; 9 and 10; 11 and 12; 13 and 14; 15 and 16; 17 and 18; 19 and 20; or a combination thereof.

[0015] In one embodiment, the determining step may use a labeled nucleic acid probe, which may be a radioactive label, a fluorescent label, an enzyme, a chemiluminescent label, a colorimetric label, or a combination thereof.

[0016] In one embodiment, the RNA can be sequenced.

[0017] In one embodiment, the biological sample can be blood, serum, whole blood, circulating tumor cells, tumor cells, plasma, urine, tissue, tumor or a combination thereof. The biological sample can be tissue, optionally tumor tissue. The tissue can be a fixed wax-embedded tissue sample.

[0018] In one embodiment, the levels of amplicons of RNA transcripts of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA can be expressed as threshold cycle (Ct) values, and the normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA amplicon levels are expressed as normalized Ct values.

[0019] In one embodiment, the reference bladder cancer samples may comprise at least 30 bladder cancer samples.

[0020] In one embodiment, the method may further comprise detecting and quantifying at least one additional biomarker of a urogenital-related cancer type in the biological sample or in a different biological sample.

[0021] In one embodiment, the method may further comprise detecting at least one additional biomarker of a different cancer type in the biological sample or in a different biological sample and quantifying the at least one additional biomarker.

[0022] In one embodiment, the method may be performed at several time points or time intervals as part of monitoring the subject at least one of before, during, and after treatment of the cancer.

[0023] In one embodiment, the method may further comprise the step of preparing a report indicating that the patient has an increased or decreased likelihood of long-term survival without bladder cancer.

[0024] In one embodiment, a non-transitory computer readable medium stores an executable program that may include instructions for performing the methods described herein.

[0025] In one embodiment, a system includes: a server, the server including at least one processor and a memory, the memory may include computer-readable instructions, and the computer-readable instructions, when executed by the processor, cause the processor to perform the following steps: receiving mRNA expression data from a computer terminal located away from the server; processing the mRNA expression data using a classification system.

[0026] In one embodiment, a method for detecting an upper tract urothelial carcinoma (UTUC) biomarker may comprise: (a) obtaining a biological sample from a subject; (b) contacting the biological sample obtained from the subject with a set of binding agents, wherein the set comprises binding agents that bind to and form a complex with a protein selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; and (c) detecting the presence and amount of protein-binding agent complexes formed in the biological sample. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0027] In one embodiment, a method for classifying test data, wherein the test data includes protein expression data, may include: (a) using at least one processor to access an electronically stored set of training data vectors, each training data vector representing an individual cancer patient and including protein expression data for the corresponding cancer patient, each training data vector further including a classification regarding the expression level of a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; (b) using the electronically stored set of training data vectors to train an electronic representation of a classification system; (c) receiving test data including protein expression data at the at least one processor; (d) using the at least one processor to evaluate the test data using the electronic representation of the classification system; and (e) outputting a classification of the test data regarding the likelihood of upper tract urothelial carcinoma (UTUC) based on the evaluation step. The classification system can be AdaBoost, artificial neural network (ANN) learning algorithm, Bayesian belief network, Bayesian classifier, Bayesian neural network, boosted tree, case-based reasoning, classification tree, convolutional neural network, decision tree, deep learning, elastic net, fully convolutional network (FCN), genetic algorithm, gradient boosted tree, k nearest neighbor classifier, LASSO, linear classifier, naive Bayes, neural network, penalized logistic regression, random forest, ridge regression, support vector machine or a set thereof. The biomarker can be composed of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA. The classification system can be a set of classification systems.

[0028] In one embodiment, a subject may be diagnosed with UTUC.

[0029] In one embodiment, the sample may be obtained from a subject having at least one symptom of UTUC.

[0030] In one embodiment, the biological sample can be blood, serum, whole blood, circulating tumor cells, tumor cells, plasma, urine, tissue, tumor, or a combination thereof. The biological sample can be blood, urine, plasma, or a combination thereof. The biological sample can be urine.

[0031] In one embodiment, the binding agent may be an antibody or an antibody fragment. The binding agent may be an antibody. The binding agent may be a monoclonal antibody. The binding agent may be a polyclonal antibody.

[0032] In one embodiment, the array may comprise biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof fixed to a substrate. The biomarkers may consist of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. The biomarkers may be mRNA transcripts. The biomarkers may be cDNAs of the mRNA transcripts. The biomarkers may be peptides.

[0033] In one embodiment, the kit may comprise a specifically binding nucleic acid primer comprising a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0034] In one embodiment, the kit may comprise an antibody that specifically binds to a biomarker selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination. The biomarker may consist of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] This patent or application file contains at least one drawing drawn in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0036] Advantages and features of the present invention will be better understood with reference to the following more detailed description in conjunction with the accompanying drawings, in which:

[0037] Figure 1 An exemplary method used by the inventors to identify a diagnostic bladder cancer signature is depicted.

[0038] Figure 2 Single-cell RNA sequencing of 25 human bladder cancers is profiled.

[0039] Figure 3 AB depicts (A) a heat map showing the relationship between Oncuria TMUtility of each of the 10 associated biomarkers in luminal versus basal tumors within the stratified TCGA cohort. Trends in gene expression from low to high are shown in blue to brown. (B) Gene expression results for individual biomarkers associated with luminal versus basal subtypes from the bladder cancer signature.

[0040] Figure 4 Association of individual 10 analytes with bladder cancer outcomes is depicted - TCGA.

[0041] Figure 5 Association of the combined 10 analytes with bladder cancer outcomes is depicted - TCGA.

[0042] Figure 6 Associations of individual 10 analytes with bladder cancer outcomes are depicted - Black race cohort.

[0043] Figure 7 Associations of the combined 10 analytes with bladder cancer outcomes are depicted - Black ethnic group.

[0044] Figure 8 Associations of individual 10 analytes with bladder cancer outcomes are depicted - GSE 32894.

[0045] Fig. 9 Association of the combined 10 analytes with bladder cancer outcomes is depicted - GSE32894.

[0046] Fig.10 Association of individual 10 analytes with bladder cancer outcomes is depicted - GSE48075.

[0047] Fig.11 Association of the combined 10 analytes with bladder cancer outcomes is depicted - GSE48075.

[0048] Fig.12 Kaplan-Meier survival curves for high versus low expression of biomarker signatures in the TCGA cohort are depicted; the inset depicts TCGA analyzed by the consensus model.

[0049] Fig.13 AC depict Kaplan-Meier survival curves for high versus low expression of the combined biomarker signature described herein: (A) GSE87304 cohort; inset depicts GSE87304 analyzed by the consensus subtyping system, (B) GSE48075 cohort; inset depicts GSE48075 analyzed by the MDA subtyping system; and (C) GSE32894 cohort; inset depicts GSE32894 analyzed by the related GSE32894 manuscript (Damrauer et al. National Academy of Sciences Journal of the Academy (Proc Natl Acad Sci USA) (2014)111:3110-3115;Choi et al. Cancer Cells Cell (2014) 25:152-165; Seiler et al. European Urology (2017) 72:544-554) for the model analysis reported in GSE32894.

[0050] Fig.14 A heat map is depicted showing the application of ONCURIA's 10 biomarkers associated with 6 consensus molecular subtypes in the TCGA cohort. Blue to brown show the trend of gene expression from low to high.

[0051] Fig.15 Comparison of urine concentrations of 10 proteinuria biomarkers in UTUC and controls is depicted. The median levels are depicted by horizontal lines. DETAILED DESCRIPTION

[0052] Before further describing the subject disclosure, it should be understood that the disclosure is not limited to the specific embodiments of the disclosure described below, as changes can be made to the specific embodiments and still fall within the scope of the appended claims. It should be understood that the terminology used is for the purpose of describing specific embodiments and is not intended to be limiting. Instead, the scope of the disclosure will be determined by the appended claims.

[0053] Bladder cancer biomarkers with prognostic value

[0054] The present disclosure relates to selected gene sets whose expression has prognostic value, particularly with respect to disease-free survival in, for example, bladder cancer.

[0055] Diagnostic tests used in clinical practice are based on single analytes and therefore fail to capture the potential value of known relationships between multiple biomarkers. Given the redundancy of signaling pathways, crosstalk between molecular networks, and the oligoclonality of tumors, single biomarker assays lack sufficient power to make critical diagnostic decisions. The inventors have discovered an RNA biomarker panel that shows unexpectedly improved prognosis for bladder cancer detection in tumor tissue.

[0056] Bladder cancer is a biologically heterogeneous disease with variable clinical manifestations, outcomes, and responses to therapy. Therefore, the clinical utility of a single biomarker for detecting and predicting the biological behavior of bladder cancer is limited. The inventors identified and validated a bladder cancer diagnostic signature that includes 10 biomarkers (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) and can be incorporated into a multiple immunoassay bladder cancer test. The inventors confirmed that these 10 biomarkers can help predict bladder cancer clinical outcomes. Tumor gene expression and patient survival data from bladder cancer cases from the Cancer Genome Atlas (TCGA) were analyzed. The comparison between the mRNA expression of 10 biomarkers and the TCGA2017 subtype classification was evaluated. Kaplan-Meier analysis of multiple gene expression data sets showed that high expression of the combined 10 biomarkers was associated with a significant decrease in overall survival. Analysis of three independent, publicly available gene expression data sets confirmed that multiple prognostic models are superior to single biomarkers. Eight of the 10 biomarkers (APOE, A1AT, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) were significantly associated with either the luminal or basal molecular subtypes, and thus the test has the potential to help predict clinical outcomes.

[0057] Bladder cancer is a biologically heterogeneous disease with variable clinical presentation, response to therapy, and clinical outcome. The molecular complexity of bladder cancer limits the clinical utility of tests that rely on single features or biomarkers for detecting and predicting bladder cancer behavior. The advent of high-throughput molecular profiling technologies has enabled the development of multiplexed molecular signatures that may be used for diagnosis, staging, prognosis, and treatment decisions. There are currently two FDA-approved multiplexed molecular tests for bladder cancer, the UroVysion and Immunocyt / Ucyt+ tests, but their clinical utility is hampered by limited sensitivity and specificity.

[0058] Multiplex immunoassays for quantitative monitoring of bladder cancer-related diagnostic signatures can include 10 protein biomarkers (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA). In a series of studies, molecular signatures were developed and tested to detect bladder cancer non-invasively by urinalysis. In addition, immunostaining studies in resected bladder tumor tissues showed that the expression of these 10 biomarkers was increased in neoplasms compared to benign urothelium, and high levels were associated with decreased overall patient survival.

[0059] Molecular subtyping of a range of solid tumors has become a valuable tool for classifying patients into genetically homologous groups to guide clinical management. Many subtyping schemes have been proposed for bladder cancers with varying degrees of complexity. The inventors analyzed a series of gene expression data sets from TCGA and Gene Expression Omnibus (GEO) to evaluate the potential utility of 10 biomarkers (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1 and VEGFA) in the molecular subtyping of bladder cancer and the prediction of clinical outcomes.

[0060] RNA-based tests have the disadvantage of RNA degradation, and it is difficult to obtain fresh tissue samples from patients for analysis. Fixed paraffin-embedded tissue is more readily available, and methods can be used to detect higher quantities and qualities of RNA and extract higher quantities and qualities of RNA from fixed tissue. The microarray may comprise cDNAs of biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof. The microarray may comprise cDNAs that can be fixed to a substrate.

[0061] Classification of bladder cancer by RNA gene expression analysis has focused on improving and refining the classification typically seen in bladder cancer and has not yet provided any new insights into bladder cancer biology or the relationships of the differentially expressed genes, nor have these studies been successful in linking findings to clinical outcomes for improved cancer therapies.

[0062] The challenge in cancer treatment remains to target specific treatment regimens to tumor types that differ in pathogenicity and ultimately to personalize tumor therapy in order to maximize outcomes. The methods described herein provide tests that simultaneously provide prognostic information about a patient's clinical outcome, for example, in bladder cancer, the biology of which is poorly understood.

[0063] The classification of biomarkers selected by the inventors was trained on archival paraffin-embedded biopsy material to test all markers in the panel and is therefore compatible with the most widely available types of biopsy material. The methods described herein are also compatible with several different tumor tissue collection methods, such as circulating tumor cells. In addition, for each member of the gene set, the methods described herein specify oligonucleotide sequences that can be used in the test.

[0064] Cancer biomarkers (also called tumor biomarkers) are molecules present in the body, such as DNA, RNA, metabolites, hormones, enzymes, and immunoglobulins, that are associated with cancer and whose measurement or identification can be used for clinical management of patients. They can be cancer cells themselves or products of the body in response to cancer or other conditions. Most cancer biomarkers are RNA. Like other cancer biomarkers, the biomarkers described herein can be used for a variety of purposes, such as: screening healthy populations or high-risk populations for the presence of bladder cancer; diagnosing bladder cancer or a specific type of bladder cancer; determining a subject's prognosis; and predicting / monitoring the course of a subject's disease when in remission or receiving surgery, radiation, chemotherapy, or other cancer treatments.

[0065] The methods described herein can be used for prognosis, prediction and / or monitoring of cancer, optionally bladder cancer, and can be performed at several time points or intervals as part of monitoring a subject before, during or after cancer treatment.

[0066] A method for evaluating the prognosis of a subject having or suspected of having cancer (optionally bladder cancer) may comprise: (a) determining the level of one or more cancer biomarkers listed in Table 1 in a biological sample obtained from the subject; (b) comparing the level determined in step (a) with the level or range of one or more cancer biomarkers known to be present in a biological sample obtained from a normal subject without cancer; and (c) determining the prognosis of the subject based on the comparison of step (b), wherein a high level of one or more cancer biomarkers in step (a) indicates a more aggressive form of cancer and, therefore, a poor prognosis. The biomarker may comprise one or more nucleotides or polypeptides encoded by a nucleic acid listed in Table 1.

[0067] A method for predicting the likelihood of long-term survival of a bladder cancer patient, the method may comprise determining the expression level of one or more prognostic RNA transcripts or their expression products in a bladder cancer tissue sample obtained from the patient, normalized to the expression level of all RNA transcripts or their products in the bladder cancer tissue sample, or a reference set of RNA transcripts or their expression products, wherein the prognostic RNA transcripts are transcripts of one or more genes selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA, an increase of which together indicates a decreased likelihood of long-term survival without recurrence of bladder cancer.

[0068] In the methods described herein, the expression levels of at least two or at least 5 or 10 prognostic RNA transcripts or their expression products may be determined.

[0069] In the methods described herein, the method may comprise determining the expression levels of all prognostic RNA transcripts or their expression products. A preferred subset of RNA transcripts may comprise ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA, which together increase the likelihood of long-term survival without recurrence of bladder cancer. The bladder cancer may be invasive bladder cancer. RNA may be isolated from a fixed wax-embedded bladder cancer tissue specimen of the patient. It may be isolated by any technique known in the art, such as from a biopsy or transurethral resection of a bladder tumor or fine needle aspiration of cells or cystectomy tissue. RNA may be isolated from circulating tumor cells of the patient. Separation may be performed by any technique known in the art. See, for example, Gjerde et al. "RNA Purification and Analysis: Sample Preparation, Extraction, Chromatography" (p. 1 Edition) (2009) Wiley-VCH Publisher (Wiley-VCH).

[0070] A method of predicting the likelihood of long-term survival in a patient diagnosed with invasive bladder cancer may comprise:

[0071] (a) determining the expression level of an RNA transcript or an expression product of a gene or a set of genes selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA (Table 1);

[0072] (b) performing statistical analysis on the data obtained in step (1); and

[0073] (c) determining whether the likelihood of long-term survival has increased or decreased.

[0074]

[0075] The gene sequences listed in Table 2 and the PCR primer-probe sets listed in Table 3 can be used to detect and / or quantify the biomarkers in the methods described herein.

[0076]

[0077]

[0078] Table 3 Primer probes for the 10 analytes comprising the bladder cancer diagnostic signature

[0079]

[0080] Prognostic methods for bladder cancer may include:

[0081] (a) subjecting a sample comprising bladder cancer cells obtained from a patient to quantitative analysis of the expression level of RNA transcript of at least one gene selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA, or an expression product thereof, and (b) identifying the patient as having a decreased likelihood of long-term survival without recurrence of bladder cancer if the normalized expression level of the gene or its product is elevated above a defined expression threshold.

[0082] The kit may comprise one or more of the following: (1) extraction buffers / reagents and protocols; (2) reverse transcription buffers / reagents and protocols; and (3) qPCR buffers / reagents and protocols suitable for performing any of the methods described herein. The kit may comprise an array, optionally a microarray, comprising cDNA transcripts consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0083] The array may comprise biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA and combinations thereof fixed to a substrate. The biomarkers may consist of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA. The biomarkers may be mRNA transcripts. The biomarkers may be cDNAs of the mRNA transcripts. The biomarkers may be peptides.

[0084] The kit may comprise a specifically binding nucleic acid primer comprising a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0085] The kit may comprise an antibody that specifically binds to a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA.

[0086] The methods described herein may also be used in combination with other diagnostic techniques currently known in the art, either invasive or non-invasive.

[0087] Here, specificity is defined as the probability that a patient without bladder cancer is assigned to the normal group, and sensitivity is the probability that a patient with bladder cancer is assigned to the disease group.

[0088] Upper tract urothelial carcinoma (UTUC)

[0089] Despite advances in technology, the diagnosis of upper urinary tract tumors remains challenging, and often cytology and / or biopsy are inconclusive or not performed due to difficulty in reaching the lesion of interest. Therefore, the development of an accurate diagnostic assay that can be applied to non-invasively obtained urine samples would benefit both patients and the healthcare system.

[0090] Urine-based assays currently play a limited role in the management of patients with upper tract urothelial carcinoma (UTUC) due to insufficient accuracy. The application of robust urine-based multiplexed assays to aid in the diagnosis of UTUC addresses this shortcoming and facilitates the possibility of accurate, non-invasive diagnosis.

[0091] The multiplex immunoassay described herein consisting of A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA showed an AUC of 0.897 (95% CI: 0.817-0.977), with an overall sensitivity of 93.5%, a specificity of 75.6%, an NPV of 93.9%, and a PPV of 74.4%. The sensitivity values ​​of the diagnostic panel for high-grade UTUC, low-grade UTUC, non-invasive UTUC, and invasive UTUC were 88.9%, 92.3%, 86.7%, and 100%, respectively. Urine cytology or selective ureteral washing / cytology was associated with an overall sensitivity of 58.3%, a specificity of 100%, an NPV of 79.2%, and a PPV of 100%. The sensitivity values ​​of cytology for high-grade UTUC, low-grade UTUC, non-invasive UTUC, and invasive UTUC were 50%, 100%, 80%, and 42.9%, respectively.

[0092] Urinary levels of a biomarker panel consisting of A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, VEGFA, and combinations thereof are provided for accurate differentiation of UTUC from control non-tumor bearing individuals. The multiplex immunoassay test described herein can achieve effective and accurate detection of UTUC in a non-invasive patient setting. The multiplex immunoassay can use an array comprising a biomarker panel consisting of: A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, VEGFA, and combinations thereof. The multiplex immunoassay can use an array comprising a biomarker panel consisting of: A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA.

[0093] The protein biomarkers described herein (A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, VEGFA, and combinations thereof) can be found in biological fluids within biomarker-positive cancer cells that are shed or released in the fluid or biological sample being studied (e.g., urine). Optionally, the sample can be blood, serum, plasma, urine, or a combination thereof. The sample can be urine. The biomarkers (A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, VEGFA, and combinations thereof) can also be found directly in a fluid or biological sample (i.e., without cells).

[0094] Methods for detecting upper tract urothelial carcinoma (UTUC) biomarkers may comprise: (a) obtaining a biological sample from a subject; (b) contacting the biological sample obtained from the subject with a set of binding agents, wherein the set comprises binding agents that bind to and form a complex with a protein selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; and (c) detecting the presence and amount of protein-binding agent complexes formed in the biological sample.

[0095] The presence and quantity of a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA and combinations thereof can be determined by immunoassay. The protein biomarker can be composed of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA. As used herein, an "assay" or diagnostic assay can be of any type used in the field of diagnosis. Preferred detection methods include various forms of immunoassays, such as radioimmunoassays, chemiluminescent immunoassays and fluorescent immunoassays, enzyme-linked immunosorbent assays (ELISAs), Luminex-based bead arrays, protein microarray assays, assays suitable for point-of-care testing, and rapid test formats, such as immunochromatographic strip tests. Such assays can be based on the binding of an analyte to be detected to one or more capture probes having a certain affinity. As used herein, an immunoassay is a biochemical test that measures the presence or concentration of a macromolecule / polypeptide in a solution by using antibodies or immunoglobulins. According to the present invention, the antibodies can be monoclonal antibodies as well as polyclonal antibodies. Thus, at least one antibody is a monoclonal or polyclonal antibody.

[0096] The immunoassay may be selected from the group consisting of: luminescent immunoassay (LIA), radioimmunoassay (RIA), chemiluminescent immunoassay and fluorescent immunoassay, enzyme immunoassay (EIA), enzyme-linked immunosorbent assay (ELISA), sandwich immunoassay, luminescence-based bead array or a combination thereof. Immunoassay techniques are described in the art, for example, Darwish et al. "biology Journal of Medical Science (Int J Biomed Sci) (2006)2(3):217-235.

[0097] A protein selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof may be fixed to a substrate. The substrate may be a microplate or an array. The substrate may be an array. The biomarker may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. The array may comprise an antibody that specifically binds to the biomarker fixed to the substrate, the biomarker fixed to the substrate being selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof.

[0098] Machine Learning

[0099] Classification system

[0100] The present invention is particularly concerned with characterizing biomarkers based on quantitative data on the expression levels of RNA transcripts, preferably on the expression levels of RNA transcripts from tissue samples. The quantitative data on the expression levels of RNA transcript datasets may be proprietary or accessible from publicly available databases. This data may be used to train a machine learning system to generate a classification for the diagnosis of cancer (optionally bladder cancer) and / or a classification for the prognosis of survival of a subject with cancer (optionally bladder cancer).

[0101] The classification system used herein may include computer executable software, firmware, hardware, or a combination thereof. For example, the classification system may include a reference to a processor and supporting data storage. In addition, the classification system may be implemented on multiple devices or other components that are local or remote to each other. The classification system may be implemented in a centralized system, or as a distributed system to achieve additional scalability. In addition, any reference to software may include a non-transitory computer-readable medium that causes the computer to perform a series of steps when executed on the computer.

[0102] The classification system described herein may include data storage devices, such as network accessible storage devices, local storage devices, remote storage devices (e.g., "cloud"), or combinations thereof. The data storage device may utilize redundant arrays of inexpensive disks ("RAID"), tapes, disks, storage area networks ("SAN"), Internet Small Computer System Interface ("iSCSI") SAN, Fibre Channel SAN, Common Internet File System ("CIFS"), Network Attached Storage ("NAS"), Network File System ("NFS"), or other computer accessible storage devices. The data storage device may be a database, such as an Oracle database, a Microsoft SQL Server database, a DB2 database, a MySQL database, a Sybase database, an object-oriented database, a hierarchical database, a cloud-based database, a public database, or other databases. The data storage device may utilize a flat file structure to store data.

[0103] In the first step, a classifier is used to describe a predetermined data set. This is the "learning step" and is performed on the "training" data.

[0104] The training database is a computer-implemented data storage reflecting multiple RNA expression level (multiple) data of multiple peptides related to the classification of diagnosis and / or prognosis characterization about biomarker levels. RNA expression level data may include experimental RNA expression level data, predicted RNA expression level data or a combination thereof. The format of the stored data may be a flat file, a database, a table or any other retrievable data storage format known in the art. Test data may be stored as multiple vectors, each vector corresponding to a single peptide, each vector including multiple RNA expression level data metrics of multiple experimental RNA expression level data and classification of antigenic characterization about peptides. The vector may further include retention time data metrics of multiple experimental peptide retention data and classification of diagnosis and / or prognosis characterization about biomarker levels. Typically, each vector contains entries of each RNA expression level data metric in multiple RNA expression level data metrics. Entry may further include retention time data. The training database may be linked to a network, such as the Internet, so that its content may be remotely retrieved by an authorized entity (e.g., a human user or a computer program). Alternatively, the training database may be located in a network-isolated computer. Additionally, the training database can be cloud-based, including proprietary and public databases containing RNA expression level data (e.g., experimental, predicted, and combinations thereof) for biomarkers used in immunological methods.

[0105] In the optional second step, the classifier is applied to the "verification" database, and various accuracy metrics are observed, including sensitivity and specificity. In an exemplary embodiment, only a portion of the training database is used for the learning step, and the remainder of the training database is used as the verification database. In the third step, the RNA expression level data metrics from the experimenter are submitted to a classification system, and the classification system outputs the classification (for example, diagnosis and / or prognosis characterization of biomarker levels) of the experimenter's calculation. In addition, other diagnostic data can also be used.

[0106] Many possible classifiers can be used on the data.Machine and deep learning classifiers include but are not limited to AdaBoost, artificial neural network (ANN) learning algorithm, Bayesian belief network, Bayesian classifier, Bayesian neural network, boosting tree, case-based reasoning, classification tree, convolutional neural network, decision tree, deep learning, elastic net, full convolutional network (FCN), genetic algorithm, gradient boosting tree, k nearest neighbor classifier, LASSO, linear classifier, naive Bayesian classifier, neural network, penalty logistic regression, random forest, ridge regression, support vector machine or its set, can be used to classify data.See, for example, Han and Kamber (2006) Chapter 6, "Data Mining: Concepts and Technology (Data Mining, Conceptsand Techniques)", 2nd Edition. Elsevier Press (Elsevier: Amsterdam) in Amsterdam.As described herein, any classifier or combination of classifiers (e.g., set) can be used in a classification system.As discussed herein, data can be used to train a classifier.

[0107] In addition, feature selection algorithms can be used in machine learning applications. For example, feature selection algorithms can be used, including but not limited to packaging methods (forward, backward and stepwise selection), filtering methods (ANOVA, Pearson correlation, variance selection method) and embedding methods (Lasso, Ridge, decision tree).

[0108] Classification Tree

[0109] Classification trees are easily interpretable classifiers with built-in feature selection. Classification trees recursively partition the data space in such a way that the proportion of observations that are from one class in each subspace is maximized.

[0110] The process of recursively partitioning the data space creates a binary tree with a condition tested at each vertex. New observations are classified by following the branches of the tree until a leaf is reached. At each leaf, a probability is assigned to the observation that it belongs to a given class. The class with the highest probability is the class to which the new observation is classified.

[0111] Classification trees are essentially decision trees whose properties are structured in statistical language. They are very flexible but very noisy (the variance of the error is large compared to other methods).

[0112] As a non-limiting example, tools for implementing classification trees are available in the statistical software computing language and environment R. For example, the R package "trees" version 1.0-28 includes tools for creating, processing, and utilizing classification trees. Examples of classification trees include, but are not limited to, random forests. See also Kamiński et al. (2017) "A framework for sensitivity analysis of decision trees". Central European Journal of Operational Research Journal of Operations Research)》. 26(1):135-159; Karimi and Hamilton (2011) “Generation and Interpretation of Temporal Decision Rules”, International Journal of Computer Information Systems and Industrial Management Applications, Vol. 3.

[0113] Random Forest

[0114] Classification trees are usually noisy. Random forests attempt to reduce this noise by taking the average of many trees. The result is a classifier whose error has reduced variance compared to classification trees. Methods (including software) for constructing random forest classifiers are known in the art. Prinzie and Poel (2007) "Random Multiclass Classification: Generalizing Random Forests to Random MNL and Random NB". Database and Expert Systems Applications. Lecture Notes in Computer Science. 4653; Denisko and Hoffman (2018) "Classification and interaction in random forests". PNAS 115(8):1690-1692.

[0115] To classify a new observation using random forests, each classification tree in the random forest is used to classify the new observation. The category that the new observation is most often classified into in the classification tree is the category that the random forest classifies the new observation into. Random forests reduce many of the problems found in classification trees, but come with a tradeoff in interpretability.

[0116] As a non-limiting example, tools for implementing random forests as discussed herein are available in the statistical software computing language and environment R. For example, the R package "random forests" version 4.6-2 includes tools for creating, processing, and utilizing random forests.

[0117] AdaBoost (Adaptive Boosting)

[0118] AdaBoost provides a method for classifying each of n subjects into two or more categories based on a k-dimensional vector for each subject, called a k-tuple. AdaBoost takes a series of "weak" classifiers that have poor predictive performance (although better than random) and combines them to produce a superior classifier. The weak classifier used by AdaBoost is the Classification and Regression Tree (CART). CART recursively divides the data space into regions where all new observations that fall within the region are assigned a class label. AdaBoost constructs a series of CARTs based on a weighted version of the data set, with the weights depending on the performance of the classifier in the previous iteration. Han and Kamber (2006) Data Mining: Concepts and Techniques, vol. 2 Edition. Elsevier Publishers, Amsterdam, the contents of which are incorporated by reference. AdaBoost technically works only when an observation can belong to two categories. For g>2 categories, a (g / 2) model must be created to classify the observation as not belonging to a group. The results from these models can then be combined to predict the group membership of a particular observation. The predictive performance in this case is defined as the proportion of observations that are misclassified.

[0119] Convolutional Neural Networks

[0120] Convolutional Neural Networks (CNN or ConvNet) are a class of deep feed-forward artificial neural networks most commonly applied to analyzing visual images. CNNs use variations of multi-layer perceptrons that are designed to require minimal preprocessing. They are also called shift-invariant or spatially invariant artificial neural networks (SIANN), based on their shared weight architecture and translation-invariant properties. Convolutional networks are inspired by biological processes because the pattern of connections between neurons resembles the organization of the visual cortex of animals. Individual cortical neurons respond to stimuli only in a restricted area of ​​the visual field called the receptive field. The receptive fields of different neurons partially overlap, making them cover the entire visual field. Compared to other image classification algorithms, CNNs use relatively little preprocessing. This means that the network learns filters that are manually designed in traditional algorithms. This independence from prior knowledge and human effort in feature design is a major advantage. LeCun and Bengio (1995) "Convolutional networks for images, speech, and time-series", Arbib (ed.), The Handbook of Brain Theory and Neural Networks, MIT Press. Fully convolutional indicates that the neural network is composed of convolutional layers without any fully connected layers or MLPs typically found at the ends of the network. Convolutional neural networks are an example of deep learning.

[0121] Support Vector Machine

[0122] Support vector machines (SVMs) are well-recognized in the art. Generally speaking, SVMs provide a model for classifying each of n subjects into two or more disease categories based on a k-dimensional vector of biomarker measurements for each subject, called a k-tuple. The SVM first transforms the k-tuple into a space of equal or higher dimensionality using a kernel function. The kernel function projects the data into a space where a hyperplane can be used to separate the categories better than was possible in the original data space. To determine the hyperplane that distinguishes between the categories, a set of support vectors that are closest to the boundaries between the disease categories can be selected. The hyperplane is then selected by known SVM techniques so that the distance between the support vectors and the hyperplane is maximized within a cost function that penalizes incorrect predictions. This hyperplane is the hyperplane that best separates the data in terms of predictions. Vapnik (1998) Statistical Learning Theory; Vapnik "An overview of statistical learning theory" IEEE Transactions on Neural Networks 10(5):988-999 (1999). Any new observation is then classified as belonging to any one of the classes of interest based on its position relative to the hyperplane. When more than two classes are considered, this process is performed for all classes in pairs, and the results are combined to create rules to distinguish between all classes.

[0123] A kernel function called Gaussian Radial Basis Function (RBF) can be used. Vapnik, 1998. RBF is often used when no prior knowledge is available that allows selection from a number of other defined kernel functions such as polynomial or sigmoid kernels. Han et al. Data Mining: Concepts and Techniques MorganKaufman 3 (2012). RBF projects the original space into a new space of infinite dimensions. A discussion of this topic and its implementation in the R statistical language can be found in Karatzoglou et al. "Support Vector Machines in R" Journal of Statistical Software 15(9) (2006), the contents of which are incorporated by reference in their entirety. All SVM statistical calculations described herein were performed using the statistical software programming language and environment R 2.10.0. SVMs were fitted using the ksvm() function in the kernlab package. Cristianini, N and Shawe-Taylor, J. (2000). An Introduction to Support Vector Machines and Other Kernel-based Learning Methods. Cambridge: Cambridge University Press. Some notes on support vector machines are provided, as well as an overview of the way they distinguish observations from multiple groups.

[0124] Other suitable kernel functions include, but are not limited to, linear kernels, radial basis kernels, polynomial kernels, uniform kernels, triangular kernels, Epanechnikov kernels, quaternary (dual weight) kernels, ternary (triplet) kernels, and cosine kernels.

[0125] A support vector machine is one of many possible classifiers that can be applied to the data.

[0126] As a non-limiting example, and as discussed below, other methods such as naive Bayes classifiers, classification trees, k-nearest neighbor classifiers, etc. can be used on the same data used to train and validate the support vector machine.

[0127] Naive Bayes Classifier

[0128] The group of Bayesian classifiers is a group of classifiers based on Bayes' theorem. See, for example, Joyce (2003), Zalta, Edward N, ed., "Bayes' Theorem", The Stanford Encyclopedia of Philosophy (Spring 2019 edited edition), Metaphysics Research Lab, Stanford University.

[0129] All classifiers of this type try to find the probability that an observation belongs to a certain class, given the observed data. The class with the highest probability is the class to which each new observation is assigned. In theory, a Bayesian classifier has the lowest error rate among this set of classifiers. In practice, this does not always happen due to violations of assumptions on the data when applying the Bayesian classifier.

[0130] The Naive Bayes classifier is an example of a Bayesian classifier. It simplifies the calculation of probabilities used in classification by assuming that each class is independent of other classes given the data.

[0131] Naive Bayes classifiers are used in many prominent anti-spam filters due to their ease of implementation and speed of classification, but have the disadvantage that the required assumptions are rarely met in practice.

[0132] Tools for implementing a naive Bayes classifier as discussed herein are available in the statistical software computing language and environment R. For example, the R package "e1071" version 1.5-25 includes tools for creating, processing, and utilizing naive Bayes classifiers.

[0133] Neural Networks

[0134] One way to think of a neural network is as a weighted directed graph, where the edges and their weights represent the influence of each vertex on the other vertices it is connected to. A neural network has two parts: an input layer (formed by data) and an output layer (in this case, the value to be predicted). Between the input and output layers is a network of hidden vertices. Depending on how the neural network is designed, there may be several vertices between the input and output layers.

[0135] Neural networks are widely used in artificial intelligence and data mining, but there is a danger that the model produced by the neural network will overfit the data (i.e., the model will fit the current data very well, but will not fit future data very well). Tools for implementing neural networks as discussed herein are available in the statistical software computing language and environment R. For example, the R package "e1071" version 1.5-25 includes tools for creating, processing, and utilizing neural networks.

[0136] k-Nearest Neighbor Classifier (KNN)

[0137] Nearest neighbor classifiers are a subset of memory-based classifiers. These classifiers must "remember" what was in the training set in order to classify new observations. Nearest neighbor classifiers do not require model fitting.

[0138] To create a k-nearest neighbor (knn) classifier, the following steps are taken:

[0139] 1. Calculate the distance from the observation to be classified to each observation in the training set. Any valid metric can be used to calculate the distance, but Euclidean and Mahalanobis distances are commonly used.

[0140] 2. Count the number of observations among the k closest observations belonging to each group.

[0141] 3. The group with the highest count is the group to which the new observation is assigned.

[0142] The Mahalanobis distance is a measure that takes into account the covariance between the variables in the observations.

[0143] Nearest neighbor algorithms have problems with categorical data because they need to calculate the distance between two points, but this can be overcome by arbitrarily defining the distance between any two groups. Such algorithms are also sensitive to changes in scale and metric. With these issues in mind, nearest neighbor algorithms can be very powerful, especially on large data sets.

[0144] Tools for implementing a k-nearest neighbor classifier as discussed herein are available in the statistical software computing language and environment R. For example, the R package "e1071" version 1.5-25 includes tools for creating, processing, and utilizing a k-nearest neighbor classifier.

[0145] Training Data

[0146] On the other hand, the methods described herein include training about 75%, about 80%, about 85%, about 90% or about 95% of the data in a library or database, and testing the remaining percentage of a total of 100% of the data. On the one hand, about 70% to about 90% of the data are trained, and about 10% to about 30% of the remaining portion of the data are tested, about 80% to about 95% of the data are trained, and about 5% to about 20% of the remaining portion of the data are tested, or about 90% of the data are trained, and about 10% of the remaining portion of the data are tested.

[0147] On the one hand, the database or library contains data from more than about 500, about 1000, more than about 1500, more than about 2000, more than about 2500 or more than about 3000 tissue samples, preferably tumor tissue samples. On the one hand, tumor tissue and healthy tissue from the same individual are analyzed.

[0148] Methods for categorizing data using classification systems

[0149] The present invention provides methods for classifying data (test data, e.g., quantitative RNA expression data) obtained from an individual. These methods relate to preparing or obtaining training data, and evaluating the test data (compared to the training data) obtained from an individual using one of the classification systems comprising at least one classifier as described above. Preferred classification systems use classifiers, such as but not limited to support vector machines (SVMs), AdaBoost, penalized logistic regression, naive Bayesian classifiers, classification trees, k nearest neighbor classifiers, deep learning classifiers, neural networks, random forests, fully convolutional networks (FCNs), convolutional neural networks (CNNs) and / or their collections. Deep learning classifiers are more preferred classification systems. The classification system outputs the classification of peptides based on test data (e.g., quantitative RNA expression data).

[0150] Particularly preferred for the present invention is an integrated method used on a classification system that combines multiple classifiers. For example, the integrated method can include SVM, AdaBoost, penalized logistic regression, naive Bayes classifier, classification tree, k nearest neighbor classifier, neural network, fully convolutional network (FCN), convolutional neural network (CNN), random forest, deep learning or any set thereof, so that peptide antigenicity (e.g., HLA peptides, antigenic peptides) is predicted. An integrated method has been developed to utilize the benefits provided by each classifier in the classifier, and to repeat the measurement of each RNA expression level data.

[0151] A method for classifying test data comprising quantitative RNA expression data of a subset of biomarkers, the method comprising: (a) accessing a set of electronically stored training data vectors, each training data vector or k-tuple representing a single biomarker and comprising RNA expression level data of the corresponding biomarker for each replicate, the training data vector further comprising a classification of diagnostic and / or prognostic characterizations for each corresponding biomarker; (b) using the electronically stored set of training data vectors to train an electronic representation of a classifier or classifier set as described herein; (c) receiving test data comprising a plurality of RNA expression level data of the biomarkers; (d) evaluating the test data using an electronic representation of a classifier and / or classifier set as described herein; and (e) outputting a classification of peptides based on the evaluation step. The test data may further comprise other data from the subject, including but not limited to histology, metabolic data, signs, symptoms, or a combination thereof.

[0152] In another embodiment, the present invention provides a method for classifying test data, wherein the test data comprises quantitative RNA expression data, the method comprising: (a) accessing an electronically stored set of training data vectors, each training data vector or k-tuple representing a single person and comprising quantitative RNA expression data of the corresponding person for each repeated test, wherein the training data further comprises a classification of the diagnostic and / or prognostic value of each corresponding biomarker; (b) using the electronically stored set of training data vectors to construct a classifier and / or a classifier set; (c) receiving test data comprising a plurality of quantitative RNA expression data of a human test subject; (d) evaluating the test data using the classifier; and (e) outputting the classification of the human test subject based on the evaluation step. Alternatively, all repeated tests (or any combination of repeated tests) can be averaged to produce a single value for each biomarker for each subject. Outputting according to the present invention includes displaying information about the classification of the human test subject in an electronic display in a human-readable form.

[0153] The set of training vectors may include at least 20, 25, 30, 35, 50, 75, 100, 125, 150 or more vectors.

[0154] The test data may be any sign, symptom, or other data measure, such as possibly histology data, metabolite data, patient demographics, tumor (cancer) characteristics, treatments, outcomes, or combinations thereof.

[0155] The data used to train a machine learning system (e.g., deep learning) can include data from a tumor including at least 5, 10, 15, 20, or 25 different indications, data from normal tissue including at least about 5, 10, 15, 20, 25, 30, 35, 40, or 45 normal (non-tumor) tissues, or a combination thereof. In addition, the data used to train a machine learning system (e.g., deep learning) can include CID (collision-induced dissociation) data, HCD (high-energy collision dissociation) data, or a combination thereof.

[0156] It should be understood that the method of classifying data can be used in any of the methods described herein. Specifically, the method of classifying data described herein can be used in a method of characterizing biomarkers, such as ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA, for use in immunological methods.

[0157] Particularly preferred for use in the present invention is an ensemble method used on a classification system that combines multiple classifiers. For example, the ensemble method may include a support vector machine (SVM), AdaBoost, penalized logistic regression, a naive Bayes classifier, a classification tree, a k-nearest neighbor classifier, a neural network, a deep learning system, a random forest, or any combination thereof, to predict diagnostic and / or prognostic characterizations of biomarkers, including a subset of biomarkers, such as ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. In addition, the set can be used to predict the association of a subset of biomarkers (ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA) with cancer types and patient outcomes. The ensemble approach exploits the benefits provided by each of the classifiers and replicates the measurement of each biomarker (ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA).

[0158] Computer-implemented method

[0159] As used herein, the term "computer" is understood to include at least one hardware processor using at least one memory. The at least one memory can store a set of instructions. The instructions can be permanently or temporarily stored in one or more memories of the computer. The processor executes the instructions stored in the one or more memories to process data. The set of instructions can include various instructions to perform a specific task or multiple specific tasks, such as those tasks described herein. Such a set of instructions for performing a specific task can be characterized as a program, a software program, or simply software.

[0160] As described above, a computer executes instructions stored in a memory or memories to process data. For example, the processing of such data may be in response to a command of a user or users of the computer, in response to previous processing, in response to a request from another computer, and / or any other input.

[0161] The computer used to implement the embodiments at least in part may be a general purpose computer. However, the computer may also utilize any of a variety of other technologies, including a special purpose computer, a computer system including a microcomputer, a microcomputer or a mainframe, such as a programmed microprocessor, a microcontroller, a peripheral integrated circuit component, a CSIC (customer specific integrated circuit) or an ASIC (application specific integrated circuit), or other integrated circuits, logic circuits, digital signal processors, programmable logic devices, such as FPGAs, PLDs, PLAs or PALs, or any other device or arrangement of devices capable of implementing at least some steps of the process of the present invention.

[0162] It should be understood that in order to practice the method of the present invention, the processor and / or memory of the computer need not be physically located in the same geographical location. That is, each of the processor and memory used by the computer can be located in geographically different locations and connected to communicate in any suitable manner. In addition, it should be understood that each of the processor and / or memory can be composed of different physical pieces of equipment components. Therefore, the processor does not have to be a single piece of equipment in one location, and the memory is another single piece of equipment in another location. That is, it is conceivable that, for example, the processor can be two or more pieces of equipment in two different physical locations. Two or more different pieces of equipment can be connected in any suitable manner, such as a network. In addition, the memory can include two or more parts of the memory in two or more physical locations.

[0163] For example, various techniques may be used to provide communications between various computers, processors, and / or memories, as well as to allow the processor and / or memory of the present invention to communicate with any other entity; for example, to obtain additional instructions or to access and use remote memory storage devices. For example, such techniques for providing such communications may include a network, the Internet, an intranet, an extranet, a LAN, an Ethernet, or any client server system that provides communications. For example, such communication techniques may use any suitable protocol, such as TCP / IP, UDP, or OSI.

[0164] Furthermore, it should be understood that the computer instructions or a set of instructions used in the implementation and operation of the present invention are in a suitable form so that the computer can read the instructions.

[0165] In some embodiments, various user interfaces can be used to allow a human user to be connected to a computer or machine for at least partially implementing an embodiment. The user interface can be in the form of a dialogue screen. The user interface can also include any of a mouse, a touch screen, a keyboard, a voice reader, a voice recognizer, a dialogue screen, a menu box, a list, a check box, a toggle switch, a button, or any other device that allows a user to receive information about the operation of the computer when the computer processes a set of instructions and / or provides information to the computer. Therefore, a user interface is any device that provides communication between a user and a computer. For example, the information provided to the computer by the user through the user interface can be in the form of a command, data selection, or some other input.

[0166] It is also contemplated that the user interface of the present invention may interact with another computer rather than a human user, e.g., to transmit and receive information. Thus, the other computer may be characterized as a user. Additionally, it is contemplated that the user interface used in the systems and methods of the present invention may partially interact with another computer or computers or multiple computers, while also partially interacting with a human user.

[0167] Nucleic acid assay

[0168] Nucleic acids, including naturally occurring nucleic acids, oligonucleotides, antisense oligonucleotides and synthetic oligonucleotides that hybridize to nucleic acids encoding biomarker polypeptides of the invention, can be used as agents for detecting the presence of the biomarkers of the invention in biological samples of cancer patients or humans at risk of developing cancer (preferably in the urine of bladder cancer patients or humans at risk of developing bladder cancer). The invention contemplates the use of nucleic acid sequences corresponding to the coding sequences of the biomarkers of the invention and their complementary sequences, as well as sequences complementary to the biomarker transcript sequences, which further occur upstream or downstream of the coding sequences (e.g., sequences contained in or extending to the 5' and 3' untranslated regions) as agents for detecting the expression of the biomarkers of the invention in biological samples of cancer patients or humans at risk of developing cancer (preferably in the urine of bladder cancer patients or humans at risk of developing bladder cancer).

[0169] Preferred oligonucleotides for detecting the presence of the biomarkers of the present invention in a biological sample are oligonucleotides complementary to at least a portion of the cDNA sequence encoding the biomarker. These complementary sequences are also referred to in the art as "antisense" sequences. These oligonucleotides can be oligoribonucleotides or oligodeoxyribonucleotides. In addition, the oligonucleotides can be natural oligomers composed of biologically significant nucleotides, i.e., A (adenine), dA (deoxyadenine), G (guanine), dG (deoxyguanine), C (cytosine), dC (deoxycytosine), T (thymine) and U (uracil) or modified oligonucleotide species that replace, for example, a methyl or sulfur atom of a phosphate oxygen in an internucleotide phosphodiester linkage. In addition, the nucleotides themselves and / or the ribose moiety can be modified.

[0170] Oligonucleotides can be chemically synthesized using any known chemical oligonucleotide synthesis method known in the art. Ausubel et al.[edit] Short Protocols in Molecular Biology Biology)》 (No. 5 Edition) (2002). For example, the oligonucleotides can be prepared using any commercially available automated nucleic acid synthesizer. Alternatively, the oligonucleotides can be produced by standard recombinant DNA techniques, such as inducing transcription of the non-coding strand. The DNA sequence encoding the biomarker can be reversed in a recombinant DNA system, for example, inserted downstream of a suitable promoter in a reverse orientation so that the non-coding strand is now transcribed.

[0171] Although oligonucleotides of any length may be utilized for hybridization to nucleic acids encoding biomarker polypeptides, oligonucleotides in the range of 8-100 nucleotides are generally preferred. The most preferred oligonucleotides for detecting biomarkers in urine samples are those in the range of 15-50 nucleotides.

[0172] The oligonucleotides selected for hybridization to the biomarker nucleic acid molecules are then isolated and purified using standard techniques, whether chemically synthesized or synthesized by recombinant DNA technology, and then preferably labeled using standard labeling protocols (e.g., with 35 S or 32 P).

[0173] The oligonucleotide pair can be used in a polymerase chain reaction (PCR) to detect the expression of a biomarker in a biological sample, optionally a quantitative PCR method. The oligonucleotide pair includes a forward primer and a reverse primer.

[0174] The presence of biomarkers in a sample from a patient can be determined by nucleic acid hybridization, such as, but not limited to, Northern blot analysis, dot blot, Southern blot analysis, fluorescence in situ hybridization (FISH), PCR, and RNA sequencing. Chromatography, preferably HPLC, and other known assays can also be used to determine the messenger RNA level of the biomarkers in a sample.

[0175] Nucleic acid molecules encoding biomarkers described herein can be found in biological fluids within biomarker-positive cancer cells, which are shed or released in the fluid or biological sample (e.g., urine) being studied. Optionally, the sample can be blood, serum, plasma, urine, or a combination thereof. The sample can be urine. Nucleic acids encoding biomarkers can also be found directly (i.e., without cells) in fluids or biological samples.

[0176] Nucleic acids used as agents for detecting biomarkers described herein in biological samples of patients can be labeled. The nucleic acids can be labeled with radioactive labels, fluorescent labels, enzymes, chemiluminescent labels, colorimetric labels, or combinations thereof. mRNA transcripts of biomarkers consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA fixed to substrates in a microarray. The microarray can contain cDNA transcripts of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA fixed to substrates in a microarray. The array can contain biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof fixed to substrates. The biomarkers may consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA. The biomarkers on the array may be mRNA transcripts. The biomarkers on the array may be cDNAs of the mRNA transcripts. The biomarkers on the array may be peptides.

[0177] The detection methods described herein can produce an output (e.g., a readout or signal) having information about the outcome of a subject with bladder cancer. For example, the output can be qualitative (e.g., "responder" or "non-responder") or quantitative (e.g., such as a concentration in nanograms / ml).

[0178] Representative terms

[0179] Unless otherwise defined, all terms used herein have the same meaning as would be understood by one skilled in the art.

[0180] In this specification and the appended claims, the singular forms "a / an" and "the" include plural referents unless the context clearly dictates otherwise. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present disclosure belongs.

[0181] As used herein, "AdaBoost" broadly refers to a bagging method that iteratively fits CART reweighting observations by the error made at the previous iteration.

[0182] As used herein, "cancer" and "cancerous" refer broadly to the physiological condition in mammals that is typically characterized by unregulated cell growth. Examples of cancer include, but are not limited to, bladder cancer, colon cancer, lung cancer, prostate cancer, hepatocellular carcinoma, stomach cancer, pancreatic cancer, cervical cancer, ovarian cancer, liver cancer, breast cancer, urinary tract cancer, thyroid cancer, kidney cancer, melanoma, and brain cancer.

[0183] As used herein, "classifier" refers broadly to machine learning algorithms such as support vector machines, AdaBoost classifiers, penalized logistic regression, elastic nets, regression tree systems, gradient tree boosting systems, naive Bayes classifiers, neural networks, Bayesian neural networks, k-nearest neighbor classifiers, deep learning systems, and random forests. The present invention contemplates methods using any of the listed classifiers, as well as methods using more than one classifier in combination.

[0184] As used herein, "classification and regression trees (CART)" broadly refers to a method based on recursively partitioning the data space to create decision trees in order to optimize some metric (usually model performance).

[0185] As used herein, a "classification system" broadly refers to a machine learning system that executes at least one classifier.

[0186] As used herein, "differentially expressed genes", "differential gene expression" refers broadly to genes whose expression is activated to a level higher or lower than that in normal subjects or control subjects in subjects with a disease (particularly cancer, such as bladder cancer). These terms also include genes whose expression is activated to a higher or lower level at different stages of the same disease. It should also be understood that differentially expressed genes can be activated or inhibited at the nucleic acid level or protein level, or can undergo alternative splicing to produce different polypeptide products. Such differences can be demonstrated by changes in, for example, the mRNA level, surface expression, secretion, or other partitions of the polypeptide. Differential gene expression can include expression comparisons between two or more genes or their gene products, or comparisons of expression ratios between two or more genes or their gene products, or even comparisons of two different processed products of the same gene, which are different between normal subjects and subjects with a disease, particularly cancer, or between different stages of the same disease. Differential expression includes both quantitative and qualitative differences in the temporal or cellular expression patterns in genes or their expression products, for example between normal cells and diseased cells, or in cells undergoing different disease events or disease stages. For purposes of the present invention, "differential gene expression" is considered to exist when there is at least about a two-fold, preferably at least about a four-fold, more preferably at least about a six-fold, and most preferably at least about a ten-fold difference in expression of a given gene between normal subjects and diseased subjects, or at various stages of disease development in diseased subjects.

[0187] As used herein, "elastic net" broadly refers to a method for performing linear regression in which the constraints are linear combinations of the L1 norm and the L2 norm of the vector containing the regression coefficients.

[0188] "Expression threshold" and "defined expression threshold" can be used interchangeably, and broadly refer to the level of the gene or gene product in question, above which the gene or gene product serves as a predictive marker for patient survival without cancer recurrence. The threshold is experimentally defined by clinical studies such as those described in the following examples. The expression threshold can be selected for maximum sensitivity, or for maximum selectivity, or for minimum error. It is entirely within the knowledge of those skilled in the art to determine the expression threshold for any situation.

[0189] As used herein, "false positive (FP)" and "false positive identification" refer broadly to an error in which an algorithmic test result indicates the presence of a disease when the disease is not actually present.

[0190] As used herein, "false negative (FN)" broadly refers to an error in which an algorithmic test result indicates that a disease is not present when the disease is actually present.

[0191] As used herein, "gene amplification" broadly refers to the process of forming multiple copies of a gene or gene fragment in a particular cell or cell line. The duplicated region (stretch of amplified DNA) is often referred to as an "amplicon." Typically, the amount of messenger RNA (mRNA) produced, i.e., the level of gene expression, also increases in proportion to the number of copies produced by the particular gene expressed.

[0192] As used herein, "HLA peptide" broadly refers to an antigenic peptide that is bound in a peptide-MHC complex and presented to T cells. HLA peptides are antigenic peptides.

[0193] As used herein, "LASSO" broadly refers to a method for performing linear regression where there is a constraint on the L1 norm of the vector of regression coefficients.

[0194] As used herein, the "L1 norm" is the sum of the absolute values ​​of the elements of a vector.

[0195] As used herein, the "L2 norm" is the square root of the sum of the squares of the elements of a vector.

[0196] As used herein, "long-term survival" refers broadly to survival of at least 3 years, more preferably at least 8 years, and most preferably at least 10 years after surgery or other treatment.

[0197] As used herein, "mammal" refers broadly to any and all warm-blooded vertebrates of the class Mammalia, characterized in that the skin is covered with hair, and in females, the milk-producing mammary glands are covered for nourishing young. Mammals include, but are not limited to, humans, livestock and farm animals, and zoo animals, sports animals, or pet animals. Examples of mammals include, but are not limited to, alpacas, llamas, capybaras, cats, camels, chimpanzees, chinchillas, cattle, dogs, gerbils, goats, gorillas, hamsters, horses, humans, lemurs, llamas, mice, non-human primates, pigs, rats, sheep, shrews, squirrels, and tapirs. Mammals include, but are not limited to, cattle, dogs, horses, cats, mice, sheep, pigs, primates, and rodent species. Mammals also include any and all mammals listed in the Mammal Species of the World preserved by the Smithsonian National Museum of Natural History, Smithsonian Institution in Washington DC. Similarly, the term "subject" or "patient" includes both human and veterinary subjects and / or patients.

[0198] As used herein, "negative predictive value (NPV)" is the number of true negatives (TN) divided by the number of true negatives (TN) plus the number of false negatives (FP), TP / (TN+FN).

[0199] As used herein, "neural network" broadly refers to a classification method that links perceptron-like objects together to produce a classifier.

[0200] As used herein, "performance score" broadly refers to the distance between the predicted values ​​and the actual values ​​in the training data. This is expressed as a number between 0-100%, where higher values ​​indicate that the predicted values ​​are closer to the actual values. Generally, a higher score means that the model performs better.

[0201] As used herein, "polynucleotide" refers broadly to any polyribonucleotide or polydeoxyribonucleotide, which may be unmodified RNA or DNA or modified RNA or DNA. Thus, for example, as defined herein, polynucleotides include but are not limited to single-stranded and double-stranded DNA, DNA including single-stranded and double-stranded regions; single-stranded and double-stranded RNA, and RNA including single-stranded and double-stranded regions; hybrid molecules, which include DNA and RNA, which may be single-stranded, or more generally double-stranded, or include single-stranded and double-stranded regions. In addition, as used herein, the term "polynucleotide" refers to a triple-stranded region comprising RNA or DNA or both RNA and DNA. The chains in such regions may be from the same molecule or different molecules. These regions may include all of one or more molecules in the molecule, but more generally only involve regions of some molecules in the molecule. One of the molecules in the triple helical region is typically an oligonucleotide. The term "polynucleotide" specifically includes cDNA. The term includes DNA (including cDNA) and RNA containing one or more modified bases. Thus, for stability or other reasons, DNAs or RNAs with modified backbones are "polynucleotides," as that term is meant herein. In addition, DNAs or RNAs containing abnormal bases (such as inosine) or modified bases (such as tritiated bases) are included within the term "polynucleotide" as defined herein. In general, the term "polynucleotide" encompasses all chemically, enzymatically and / or metabolically modified forms of unmodified polynucleotides, as well as chemical forms of DNA and RNA characteristic of viruses and cells (including simple and complex cells).

[0202] The "positive predictive value (PPV)" is the true positives (TP) divided by the true positives (TP) plus the false positives (FP), TP / (TP+FP).

[0203] As used herein, "prediction" refers broadly to the likelihood that a patient will respond favorably or unfavorably to a drug or a group of drugs, and the extent of these responses, or the likelihood that a patient will survive a period of time without cancer recurrence after surgical removal or primary tumor and / or chemotherapy. The prediction methods of the present invention can be used clinically to make treatment decisions by selecting the most appropriate treatment modality for any particular patient. The prediction methods of the present invention are valuable tools for predicting whether a patient is likely to respond favorably to a treatment regimen such as surgical intervention, chemotherapy, immunotherapy, radiotherapy, or any combination of these therapies, or whether a patient is likely to survive long term after surgery and / or termination of chemotherapy or other treatment modalities.

[0204] As used herein, the term "prognosis" refers broadly to predicting the likelihood of cancer-attributable mortality or progression, including recurrence, metastatic spread, and drug resistance, of a neoplastic disease, such as bladder cancer.

[0205] As used herein, "random forest" broadly refers to a bagging method that fits CART based on samples from a dataset used to train the model.

[0206] As used herein, "ridge regression" broadly refers to a method for performing linear regression in which there is a constraint on the L2 norm of the vector of regression coefficients.

[0207] "Sample", "biological sample" refers broadly to a type of material known or suspected to express or contain biomarkers of cancer (such as tumors). Test samples can be used directly as obtained from a source or after pretreatment to modify the characteristics of the sample. Samples can be derived from any biological source, such as tissues or extracts, including cells (e.g., tumor cells) and physiological fluids, such as whole blood, plasma, serum, peritoneal fluid, ascites, etc. The sample can be obtained from an animal, preferably a mammal, most preferably a human. The sample can be pretreated by any method and / or can be prepared in any convenient culture medium that does not interfere with the assay. The sample can be processed before use, such as preparing plasma from blood, diluting viscous fluids, applying one or more protease inhibitors to the sample (such as urine), etc. Sample processing can involve filtration, distillation, extraction, concentration, inactivation of interfering components, and addition of reagents.

[0208] As used herein, "standard of deviation (SD)" is the spread among a single data point (ie, among a set of replicates) to reflect the uncertainty of a single measurement.

[0209] "Subject" and "patient" are used interchangeably and refer broadly to a mammal that may have cancer, such as bladder cancer. The subject can be male or female.

[0210] As used herein, "subset" refers broadly to a proper subset, and a "subset" is a proper superset.

[0211] As used herein, a "training set" is a group of samples used to train and develop a machine learning system, such as an algorithm used in the methods and systems described herein.

[0212] As used herein, a "true negative (TN)" is an algorithmic test result indicating that a peptide is not antigenic when the peptide is actually antigenic.

[0213] As used herein, a "true positive (TP)" is when the algorithm test result indicates that a peptide is antigenic when the peptide is actually antigenic.

[0214] As used herein, "tumor" refers broadly to all neoplastic cell growth and proliferation, whether malignant or benign, and all precancerous and cancerous cells and tissues.

[0215] As used herein, a "validation set" broadly refers to a set of samples that are blinded and used to confirm the functionality of the algorithms used in the methods and systems described herein. This is also referred to as a blind set.

[0216] Examples

[0217] Example 1

[0218] Diagnostic Bladder Cancer Signature

[0219] Figure 1 The method used by the inventors to discover and validate a diagnostic bladder cancer signature is described in . The inventors developed this method to test many possible options until one is likely to achieve a successful result, and the prior art gives neither an indication of which parameters are critical nor an indication of which of the many possible options are likely to succeed. Briefly, two complementary techniques were applied to profile: urine samples from patients with and without bladder cancer; gene expression (mRNA) of the exfoliated urothelium (Rosser et al. Cancer Epidemiology, Biomarkers and Prevention (Cancer Epidemiol Biomarkers Prev.)》 (2009) 18(2):444–53; Urquidi et al. "cancer Epidemiology, Biomarkers, and Prevention (2012)21(12):2149–58), Glycoproteomic analysis of urine supernatant Analysis (Kreunin et al. Journal of Proteome Res .(2007)6(7):2631-9; Yang et al. Clin Cancer Res .(2011)17(10):3349–59).

[0220] To derive the diagnostically optimal multigene diagnostic signature from microarray data, the LoGo feature selection algorithm was applied. Sun et al. IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE Trans Pattern Anal Mach Intell) .(2010)32(9):1610-1626;Sun et al. Prostate(2009) 69(10):1119-1127. To avoid possible overfitting of the computational model to the training data, the leave-one-out cross-validation (LOOCV) method was used to estimate classifier parameters and predictive performance. Goodison et al. Bioanalysis (2010) 2(5):855-862. Receiver operating characteristic (ROC) curves obtained by varying the decision threshold (van Vliet et al. BMC Genomics Genomics (2008) 9:375) was used to provide a direct view of how the prediction model performed at different levels of sensitivity and specificity. Scatter plots were created to illustrate the relative prediction scores, and t-tests were used to assess significance between groups.

[0221] Then, using advanced bioinformatics, the two datasets were combined and a bladder cancer-associated signature containing 14 candidate biomarkers was identified. [ANG, A1AT, APOE, CA9, CCL18, CD44, IL8, MMP9, MMP10, OPN, PAI-1, PTX, SDC1, and VEFGA] This "signature" consists of both mRNA and glycoprotein. The inventors validated these 14 biomarkers at the protein level with 14 separate commercial ELISA kits. The potential clinical utility of the candidate 14 protein biomarkers was monitored in urine samples from an independent group of 127 patients (64 with bladder cancer). Of the 14 biomarkers, 10 were confirmed to be significantly different between cancer and controls. Goodison et al. PLoS One)》 (2012)7(10:e47469).

[0222] The 10 protein biomarkers included angiopoietin, ANG; apolipoprotein E, APOE; alpha-1 antitrypsin, A1AT; carbonic anhydrase 9, CA9; interleukin 8, IL8; matrix metallopeptidase 9, MMP9; matrix metallopeptidase 10, MMP10; plasminogen activator inhibitor 1, PAI1; syndecan 1, SDC1; and vascular endothelial growth factor A, VEGFA, resulting in a diagnostic sensitivity of 92% at a specificity of 97% when combined using logistic regression. Recognizing that benign conditions may adversely affect the performance of urine biomarkers, a bladder cancer-associated signature was validated in an independent cohort of 102 bladder cancer patients and 206 controls, with a sensitivity of 74% at a specificity of 90%. Controls included patients with a variety of benign conditions such as urinary tract infections, hematuria without cancer, kidney stones, moderate to severe urination symptoms, and erectile dysfunction. Rosser et al. Journal of Urology .(2013)190(6):2257-62.

[0223] Subsequently, the bladder cancer-associated signature was validated by an independent laboratory in a cohort of 183 bladder cancer patients and 137 controls with a sensitivity of 79% at a specificity of 79%. Chen et al. Cancer Epidemiology, Biomarkers and prevention" .(2014)23(9):1804-12. Next, the "signature" was also demonstrated to perform equally well in the detection of recurrent bladder cancer in a cohort of 125 patients (53 recurrent cancer and 72 non-tumor recurrences) under disease surveillance, outperforming both the UroVysion Bladder Cancer Kit (Abbott) and VUC in this case with 79% and 88%, 42% and 94%, and 33% and 90% sensitivity and specificity, respectively. Rosser et al. Cancer Epidemiology, Biomarkers and Prevention . (2014) 23(7):1340-5. Analysis of cumulative data from more than 1,100 patients confirmed the diagnostic power of a multifactorial protein “signature” relative to individual biomarkers, regardless of tumor histologic grade or disease stage (Masuda et al. Oncotarget (2018) 9:7101-11), prompted us to develop a multiplex immunoassay. The prototype of the multiplex immunoassay was tested in two large independent cohorts. In a US cohort of 200 patients (100 with bladder cancer), the immunoassay achieved 80% diagnostic sensitivity at 81% specificity (Shimizu et al. Journal of Translational Medicine (J Transl Med) (2016) 14:31), and in a Japanese cohort of 278 patients (211 with bladder cancer), an optimized iteration of a multiplex immunoassay achieved a diagnostic sensitivity of 85% at a specificity of 81% (Goodison et al. Journal of Translational Medicine . (2016) 14(1):287). In addition, in 2019, the performance of multiplexed immunoassays has been tested and compared on two different technology platforms (Furuya et al. Diagnostics (2019)9(4)).

[0224] However, the inventors performed gene expression arrays using more modern methods such as single-cell RNA sequencing and noted that 9 of the 10 analytes were overexpressed in human bladder tumors (n=25; Gouin et al. Nature Communications Commun) .(2021)12(1):4906)( Figure 2 Several immunohistochemical staining studies have demonstrated that these 10 analytes are present in human tumors at the protein level and may be associated with tumor grade, tumor stage, and clinical outcome.

[0225] Analytical validation of the test has been performed using excreted urine as the test matrix to evaluate selectivity, sensitivity, specificity, accuracy, linearity, dynamic range, and detection threshold (Huang et al. Cancer epidemiology, biomarkers, and prevention .(2016)25(9):1361-6. The lower and upper limits of quantitation (LLOQ and ULOQ), the effects of antigen cross-reactivity and potential interference of matrix substances on the assay have been defined.

[0226] A small clinical validation study consisting of a cohort of 362 patients (46 with bladder cancer) was conducted. The median age of the bladder cancer subjects was 69 years (range 38-87 years), 76.1% were male, and 67.4% were Caucasian. Of the 46 bladder cancer cases, 61.4% were classified as NMIBC; stage Ta, Tis, T1), and 38.6% were classified as MIBC; stage > T2, 19.6% of cases were reported as low-grade cancer, and 80.4% of cases were reported as high-grade (Hirasawa et al. Journal of Translational Medicine .(2021)19(1):141).

[0227] Transcriptional and survival data of bladder cancer patients from The Cancer Genome Atlas (TCGA) were analyzed for 10 analytes: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA as it relates to bladder cancer molecular subtypes. Survival analysis was performed to explore the prognostic significance of the individual as well as the 10 analytes. Eight of the 10 analytes were significantly associated with the reported bladder cancer molecular subtypes ( Figure 3 ).

[0228] Individually, only low expression of PAI-1 (HR = 1.34 (1-1.8), p = 0.0493) and MMP9 (HR = 1.36 (1.02-1.83), p = 0.0395) and high expression of VEGFA (HR = 0.66 (0.492-0.885), p = 00574) were noted to be associated with improved overall survival probability ( Figure 4 ).

[0229] Overall, it was noted that tumors with low overall expression of the 10 analytes were associated with improved overall survival probability compared with high expression (HR = 1.65 (1.23-2.22), p = 0.000819) ( Figure 5 ).

[0230] To further validate the 10 analytes in bladder cancer, the analytes from the black cohort (Seiler et al. Clinical Cancer Research.(2019)25(16):5082-5093), GSE32894(Damrauer et al. "beautiful Proceedings of the National Academy of Sciences .(2014)111(8):3110-3115) and GSE48075 (Choi et al. "cancer cell". (2014) 25(2):152-165) transcript database for analysis.

[0231] In the black cohort, only low expression of MMP10 (HR = 0.607 (0.35-1.05), p = 0.0683) approached significance for improved overall survival ( Figure 6 Overall, it was noted that tumors with low overall expression of the 10 analytes were associated with improved overall survival probability compared with high expression (HR = 2.05 (1.19-3.52), p = 0.0118) ( Figure 7 In the GSE32894 cohort, only low expression of APOE (HR=2.67 (1.22-5.85), p=0.0213), IL8 (HR=4.21 (1.92-9.23), p=0.00172), MMP9 (HR=3.3 (1.51-7.23), p=0.00669), A1AT (HR=4.32 (1.97-9.47), p=0.00134) and high expression of SDC1 (HR=0.158 (0.0719-0.347), p=0.000099) and VEGFA (HR=0.432 (0.197-0.946), p=0.0432) were associated with improved overall survival ( Figure 8 Overall, it was noted that tumors with low overall expression of the 10 analytes were associated with improved overall survival probability compared with high expression (HR = 4.38 (2-9.6), p = 0.0012) ( Fig. 9 In the GSE48075 cohort, only low expression of A1AT (HR=1.64 (0.912-2.95), p=0.095) approached significance for improved overall survival ( Fig.10 Overall, it was noted that tumors with low overall expression of the 10 analytes were associated with improved overall survival probability compared with high expression (HR = 1.98 (1.06-3.6), p = 0.0192) ( Fig.11 ).

[0232] The inventors validated a diagnostic molecular signature containing 10 analytes using an independent validation sample set of naturally excreted urine samples, comprising 37 non-cancer controls and 44 cancer cases (Urquidi et al. "cancer Epidemiology, Biomarkers, and Prevention.(2012)21(12):2149-58). Target transcripts were measured in urothelial cell RNA samples using quantitative real-time RT-PCR. The low density array (TLDA) includes 44 candidate biomarker targets plus 4 selected endogenous controls (described above and below) selected by screening the levels of 15 commonly used endogenous controls in the entire sample group. The biomarker targets are mainly selected from the p-value ranking and molecular signature model described above, but also include several putative biomarkers (TERT, KRT20, CLU, PLAU, CALR, CA9, ANG). When other selection criteria are equal, genes encoding integral membrane proteins or secreted proteins are selected because these categories have the potential to be developed as biomarkers for urine analysis. For quantitative PCR analysis, RNA extraction was performed as described (Urquidi et al. Cancer Flow Epidemiology, Biomarkers, and Prevention . (2012) 21(12):2149-58). Purified RNA samples were quantitatively and qualitatively evaluated using an Agilent Bioanalyzer 2000 before storage at -80°C. Complementary DNA was synthesized from 20 ng to 500 ng of total RNA using a high capacity cDNA reverse transcriptase kit (Applied Biosystems, Foster City, CA) according to the manufacturer's instructions in a total reaction volume of 20 μl with random primers.

[0233] The selection of endogenous reference controls was accomplished by using an aliquot of each sample cDNA in a multiplex PCR preamplification reaction of the following 15 endogenous reference targets: GAPDH; ACTB; B2M; GUSB; HMBS; HPRT1; IPO8; PGK1; POLR2A; PPIA; RPLP0; TBP; TFRC; UBC; YWHAZ. Subsequently, 12.5 μl of the pooled assay mix (0.2X) was mixed with 4 μl of each cDNA sample and 25 μl PreAmp Master Mix (2X) was combined in a final volume of 50 μl. Thermocycling conditions will be as follows: initial hold at 95°C during 10 minutes, and ten preamplification cycles of 15 seconds at 95°C and 4 minutes at 60°C. Endogenous control arrays (Applied Biosystems) were used to dilute the preamplified products 1:5 in TE buffer prior to amplification in singleplex reactions. Reactions were performed on a 7900HT Fast Real-Time PCR System (AB). Genes with minimal variable expression across previous samples (UBC; PPIA; PGK1; GAPDH) were identified using GeNorm software (Integromics, Granada, Spain) and deployed as endogenous controls.

[0234] A pre-designed assay constructed by Applied Biosystems (AB) using probes spanning exon junctions Custom array preamplification and amplification reactions were performed using a low-density array (TLDA). The targets included were: UBC; PPIA; PGK1; GAPDH (4 endogenous controls); ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA. A pooled 48 Gene expression assays were performed with multiplex PCR preamplification reactions. Assay reagents at 0.2X final concentration were combined with 7.5 μl of each cDNA sample and 15 μl of TaqMan PreAmp Master Mix (2X) in a final volume of 30 μl. Thermocycling conditions were as follows: initial hold at 95°C during 10 minutes; fourteen preamplification cycles of 15 seconds at 95°C and 4 minutes at 60°C and a final hold at 99.9°C for 10 minutes. Ten microliters of undiluted preamplification product were used in subsequent singleplex amplification reactions, mixed with Universal PCR master mix (AB) was combined in a final volume of 100 μl. One human universal reference total cDNA sample (Clontech) was included as a calibrator in each microfluidic card. Reactions were run in a 7900HT Fast Real-Time PCR System (AB).

[0235] The real-time PCR amplification results were processed with RQ Manager (AB) and StatMiner (Integromics) software packages. The baseline correction of each target was manually checked, and the Ct threshold was set to 0.2 for each target across all plates. The geometric mean of four endogenous reference targets (UBC, PPIA, PGK1 and GAPDH) was used as a normalization factor, and the δ-δ CT value was calculated using the universal reference total cDNA of people (Clontech) as a calibrator. Genes that were considered to be differentially expressed were determined by t-test comparison (p < 0.01).

[0236] The derivation of the best diagnostic molecular signature was achieved by using L1 regularized logistic regression to build a predictive model (i.e., to predict the actual status of a given sample as cancer or control). The strategy, optimal solution parameters, and application details of the fast implementation of the L1 regularized learning algorithm are presented below. Due to the small sample size, a leave-one-out cross-validation (LOOCV) method was used to estimate the predictive performance. Lodewyk et al. Bioinformatics (2005) 21(19):3755-3762. In each iteration, one sample is reserved for testing and the remaining samples are used for training. The regularization parameter λ is first estimated by ten-fold cross validation using the training data, and then the prediction model is trained using the estimated parameter and blindly applied to the held-out samples.

[0237] The experiment was repeated until all samples were tested. ROC curves were then drawn to visualize how to perform prediction models under different sensitivity and specificity levels, and the receiver operating characteristic area under the curve (AUC) was reported. In order to verify the data, a permutation test was also performed to estimate the p value of prediction performance. The permutation test was repeated 1000 times. In each iteration, the category marks were randomly disrupted, the experimental scheme described above was performed, and the area under the ROC curve was recorded. The p value was calculated as the frequency of occurrence of iteration, wherein the AUC obtained was better than the AUC obtained using the original category mark. P value < 0.01 was considered to be statistically significant. Statistical analysis was performed by SPSS13.0 and MedCalc version 8.0 (MedCalc Software, Mariakerke, Belgium, Mariakerke, Belgium).

[0238] Differential expression values ​​were calculated by normalization using reference targets (UBC, PPIA, PGK1, and GAPDH) and human universal reference total cDNA (Clontech) as calibrators on each plate. KM curves were plotted to visualize how each prediction model performed with low versus high expression of the 10 analytes.

[0239] Example 2

[0240] A diagnostic gene expression signature for bladder cancer can stratify cases into defined molecular subtypes and predict outcome

[0241] Materials and methods

[0242] Data collection

[0243] A discovery cohort of 430 samples from TCGA with gene transcriptome data, of which 404 patients had valid survival data (19 normal and 411 cancer). The dataset only included one non-muscle invasive bladder cancer (NMIBC) patient, of which the rest were muscle invasive bladder cancer (MIBC) patients. The following three additional datasets were accessed for validation analysis: GSE87304; including 303 MIBC patients with the primary outcome of recurrence-free survival (Seiler et al. European Urology .(2017)72:544-554), GSE48075; including 142 NMIBC patients with the primary outcome of overall survival (Choi et al. "cancer cell" (2014) 25:152-165) and Table 4 Demographic and clinicopathological characteristics of the study cohort

[0244]

[0245] GSE32894; including 215 NMIBC and 93 MIBC patients (Damrauer et al. Proceedings of the National Academy of Sciences Journal (2014) 111: 3110-3115), the patients each have the primary outcome of disease-specific survival. These data sets are an open resource with no obvious ethical issues. The study populations within these four groups are presented in Table 4. In brief, TCGA mainly has MIBC treated by cystectomy, GSE87304 has MIBC treated with neoadjuvant chemotherapy (NAC) before cystectomy, GSE48075 has a mixture of NMIBC and MIBC treated with or without NAC, and GSE32894 has transurethral resection of bladder tumors (TURBT).

[0246] Data processing and analysis

[0247] Illumina Hi-Seq counts of bladder urothelial carcinoma from TCGA were downloaded from the Genomic Data Commons (GDC) data portal, and corresponding clinical annotations including survival information were accessed via the TCGA clinical data resource. Consensus MIBC classification of TCGA cases was obtained from the consensus MIBC study. Comprehensive analysis was performed using the edgeR package to obtain gene expression values ​​(Robinson et al. Bioinformatics (2010)26:139-140.

[0248] Survival analysis

[0249] Plan-Meier curves were used to determine the association between individual biomarkers (low expression versus high expression) and prognosis. High expression was defined as > median and low expression was defined as < median.

[0250] Univariate and multivariate analyses

[0251] Biomarkers associated with each multiple test were evaluated by univariate Cox regression, and then correlated biomarkers were evaluated using a multivariate Cox regression model to select the biomarker most closely associated with survival. All statistical analyses were performed using SPSS 19.0.

[0252] KEGG pathway analysis

[0253] The Database for Annotation, Visualization, and Integrated Discovery (DAVID, david.ncifcrf.gov / ) was used to perform Gene Ontology (GO) functional analysis (Dennis et al. Genome Biol .(2003)4:p.3), thereby reporting top-level biological processes and cellular components.

[0254] result

[0255] Using a series of gene expression datasets from TCGA and GEO, the inventors evaluated the association of 10 biomarkers (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1 and VEGFA) comprising the diagnostic signature with the defined bladder cancer molecular subtypes and clinical outcomes described herein. The TCGA cohort was parsed into luminal (n=242, 59%) and basal (n=166, 41%) subtypes based on gene expression profiles ( Figure 3 A)(Choi et al. "cancer cell" (2014) 25: 152-165). Analysis showed that 79% of the samples in the luminal subtype showed high expression of VEGFA (p = 4.36e-8) and SDC1 (1.62-13) ( Figure 3 B). Notably, tumors with papillary morphology were significantly enriched in the luminal subtype (luminal 80% vs. basal 20%; p = 1.29e-4). In contrast, 83% of samples in the basal subtype had high expression of MMP9 (p = 1.49e-29), MMP10 (p = 1.14e-2), IL8 (p = 1.52e-8), SERPINE1 (p = 2.7e-9), APOE (p = 1.05e-10), and SERPINA1 (p = 2.04e-23) ( Figure 3 B). The basal subtype was enriched in higher stage tumors (T2-4, 93% vs. Ta and T1, 7.3%; p=5.4e-34).

[0256] The inventors also tested whether the subset of biomarkers described herein compares to a more modern consensus set of the following six molecular classes of bladder cancer (Kamoun et al. European Urology (2020) 77:420-433) differentially expressed: luminal papillary, luminal non-specified, luminal unstable, matrix-rich, basal / squamous, and neuroendocrine-like. Despite the limited number of subjects in some molecular categories (e.g., neuroendocrine-like and luminal non-specified), the analysis showed that the biomarker subsets described herein could separate samples into six consensus subtypes ( Fig.14 ). Taken together, these findings show that the expression patterns of the biomarker subsets described herein are associated with reported molecular subtypes of bladder cancer.

[0257] The reported molecular subtypes have been reported to be associated with overall survival. Here, Plan-Meier analysis of TCGA subjects showed that high expression of the subset biomarker signature described herein (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) was associated with a significant decrease in overall survival ( Fig.12 Analysis of each of the 10 individual biomarkers described herein (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) showed that high expression of MMP9 (HR=1.36; p=0.0395) and SERPINE1 (HR=1.34; p=0.0493) was associated with a significant decrease in overall survival, while high expression levels of VEGFA (HR=0.66; p=0.00574) were associated with a significant improvement in overall survival ( Figure 4 Table 5 reports the Cox univariate and multivariate analyses, in which VEGFA, MMP9, SERPINA1, and SERPINE1 were associated with the probability of survival.

[0258] Table 5 Based on 10 Oncuria TM Coefficients of Cox regression analysis of biomarkers

[0259]

[0260] Three independent publicly available datasets (GSE87304, GSE48075, GSE32894) were used for validation studies. Nine of the 10 biomarkers described herein (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) were present in each dataset. Of note, IL-8 was not present in GSE87304, and ANG was not present in GSE48075 and GSE32894. Similar to the analysis of TCGA data, tumors with relatively low expression of the combined biomarker signature described herein (ANG, APOE, A1AT, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) were found to be associated with improved survival probabilities in all three datasets (GSE87304, recurrence-free survival probability, HR=2.05 (1.19-3.52), p=0.0118; GSE48075, overall survival probability, HR=1.98 (1.06-3.6), p=0.0192; GSE32894, disease-specific survival probability, HR=4.38 (2-9.6), p=0.0012) (respectively Fig.13 AC). In the GSE87304 cohort, only low expression of MMP10 (HR = 0.607 (0.35-1.05), p = 0.0683) approached significance for improved recurrence-free survival (Supplementary Figure 2 A). In the GSE48075 cohort, only low expression of SERPINA1 (HR=1.64 (0.912-2.95), p=0.095) approached significance for improved overall survival (Supplementary Figure 2 B). In the GSE32894 cohort, low expression of APOE (HR=2.67 (1.22-5.85), p=0.0213), IL8 (HR=4.21 (1.92-9.23), p=0.00172), MMP9 (HR=3.3 (1.51-7.23), p=0.00669), SERPINA1 (HR=4.32 (1.97-9.47), p=0.00134) and high expression of SDC1 (HR=0.158 (0.0719-0.347), p=0.000099) and VEGFA (HR=0.432 (0.197-0.946), p=0.0432) were associated with improved disease-specific survival (Supplementary Figure 2). Figure 2C). In conclusion, these findings validate the concept that multiple signatures provide better prognostic models than individual biomarkers. Prospective validation in larger cohorts may lead to the derivation of weighted algorithms that will maximize the utility of molecular signatures for subtyping and prognosis. Subsequently, GO enrichment analysis showed that biomarkers associated with the bladder cancer diagnostic signature were significantly enriched in the regulation of positive chemotaxis and induction of vascular permeability affecting the extracellular matrix, which are key processes in tumor growth.

[0261] Table 6 Top biological processes and cellular components reported in the gene ontology pathway analysis of differentially expressed genes associated with bladder cancer-related signatures

[0262]

[0263] discuss

[0264] Several attempts have been made to identify biomarker panels for potential cancer detection, but these studies have had relatively small sample sizes, limited populations (i.e., including few benign, mixed conditions), and have not been extensively validated. This demonstrates the inherent difficulty in identifying, characterizing, and validating a biomarker subset for bladder cancer that has high sensitivity and accuracy.

[0265] Urothelial carcinoma is pathologically classified as either non-muscle-invasive bladder cancer (NMIBC) or muscle-invasive bladder cancer (MIBC). Standard treatment for NMIBC is transurethral resection of bladder tumor (TURBT) for low-risk cases, or TURBT followed by intravesical therapy (such as BCG) for high-risk NMIBC, and the universal treatment for MIBC is radical cystectomy. A significant number of patients with NMIBC (50% to 80%) have tumor recurrence (van der Heijden and Witjes et al., 2011). European Urology Supplement (European Urology Supplements)》 (2009) 8:556-562), and up to 45% progress to MIBC after 5 years, resulting in poor survival rates associated with more advanced disease. Pathological staging is a key factor in current clinical decision-making and prognosis of bladder cancer; however, clinical outcomes of patients with the same stage are often different, indicating that current staging systems are inadequate to reflect biological heterogeneity and that accurately determining a patient's prognosis is challenging. Prognostic assessment models based on molecular signatures or subtypes may be able to better guide individualized treatment and improve outcome prediction.

[0266] The inventors' analysis showed that biomarker levels within a diagnostic test using a subset of biomarkers described herein could stratify patients into luminal (VEGFA (p = 4.36e-8) and SDC1 (p = 1.62e-13) or basal (MMP9 (p = 1.49e-29), MMP10 (p = 1.14e-2), IL8 (p = 1.52e-8), SERPINE1 (p = 2.7e-9), APOE (p = 1.05e-10), and SERPINA1 (p = 2.04e-23) subtypes. In addition, survival curve analysis showed that a multivariate model including specific biomarkers described herein (VEGFA, MMP9, SERPINA1, and SERPINE1) was associated with outcome. Validation analysis using three publicly available datasets (GSE87304, GSE48075, GSE32894) confirmed that the multiple signatures provided a better prognostic model than individual biomarkers ( Figure 4 , 12 , 13). Finally, GO enrichment analysis indicated enriched biomarker activity within the extracellular space (Table 6). The immunostaining patterns of the biomarker subsets described herein were enriched in human bladder interstitial tissue in malignant disease and were associated with decreased overall survival.

[0267] In summary, the inventors have demonstrated that the biomarkers comprise an established diagnostic signature that is valuable for molecular subtyping and prediction of clinical outcomes in patients with bladder cancer. Specifically, patients with high expression of the biomarker signature described herein are associated with a significant decrease in overall survival.

[0268] Example 3

[0269] Diagnostic Accuracy of Biomarker Panels for Detecting Upper Tract Urothelial Tumors

[0270] Overview

[0271] To evaluate the performance of multiple immunoassays that can query 10 protein biomarkers A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1 and VEGFA in urine samples discharged. Its application to UTUC was evaluated in a multi-institutional group, with 31 prospectively collected subjects undergoing upper urinary tract mass evaluation and 41 prospectively collected matched controls (i.e., non-tumor-bearing). The ability of the test to identify patients with UTUC was evaluated. UTCU status was confirmed by endoscopy and tissue biopsy or definitive surgery. ROC curves were used to evaluate diagnostic performance.

[0272] The multiplex immunoassay described herein consisting of A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA showed an AUC of 0.897 (95% CI: 0.817-0.977), with an overall sensitivity of 93.5%, a specificity of 75.6%, an NPV of 93.9%, and a PPV of 74.4%. The sensitivity values ​​of the diagnostic panel for high-grade UTUC, low-grade UTUC, non-invasive UTUC, and invasive UTUC were 88.9%, 92.3%, 86.7%, and 100%, respectively. Urine cytology or selective ureteral washing / cytology was associated with an overall sensitivity of 58.3%, a specificity of 100%, an NPV of 79.2%, and a PPV of 100%. The sensitivity values ​​of cytology for high-grade UTUC, low-grade UTUC, non-invasive UTUC, and invasive UTUC were 50%, 100%, 80%, and 42.9%, respectively.

[0273] Urinary levels of a biomarker panel consisting of A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA are provided to accurately distinguish UTUC from control non-tumor bearing individuals. The multiplexed immunoassay test described herein enables efficient and accurate detection of UTUC in a non-invasive patient setting.

[0274] Despite technical advances, the diagnosis of upper urinary tract tumors remains challenging, and cytology and / or biopsy are often inconclusive or not performed due to the difficulty in reaching the lesions of interest. Therefore, the development of accurate diagnostic assays that can be applied to non-invasively obtained urine samples will benefit both patients and the healthcare system. In this study, the potential clinical utility of the multiplex immunoassay described herein for detecting UTUC bladder in a prospectively recruited patient cohort undergoing urological evaluation at three institutions was tested. The multiplex immunoassay described herein achieved a strong overall diagnostic performance, thereby achieving an AUC of 0.897 (95% CI: 0.817-0.977), with overall sensitivity and specificity values ​​of 93.5% and 75.6%, respectively, and negative predictive values ​​(NPV) and positive predictive values ​​(PPV) of 93.9% and 74.4%, respectively. The multiplex immunoassay described herein shows promise for clinical application in the non-invasive evaluation of patients suspected of carrying UTUC.

[0275] method

[0276] Patient characteristics.

[0277] According to the approval of the Institutional Review Board and informed consent, urine samples and related coded clinical information were collected in the tissue bank of Cedars-Sinai Medical Center (Los Angeles, CA) and Kindai University (Osaka, Japan). The tissue bank of subjects with biopsy-proven UTUC (n=31) and matched controls (n=41) was queried. The control group consisted of 41 subjects with no history of UTUC (microscopic hematuria, n=7; gross hematuria, n=4; urinary tract infection, n=1; urination symptoms, n=6, bladder cancer history, n=4, prostate cancer, n=1, kidney stones, n=5; control, n=13), and the subjects were matched for age, sex, and race. In cancer subjects and control subjects with hematuria, in addition to cystoscopy, axial imaging of the abdomen and pelvis was performed with and without intravenous contrast. In subjects with abnormalities found during upper urinary tract imaging or abnormalities found during cystoscopy, a formal evaluation was performed in the operating room under anesthesia. This evaluation consisted of cystoscopy for bladder lesions only or cystoscopy and ureteroscopy for upper urinary tract lesions. Urothelial cell carcinoma confirmed by histological examination of resected tumor tissue (biopsy and / or nephroureterectomy) was recorded in all cancer subjects. Relevant information about clinical presentation, stage, histological grade, and outcome was recorded (Table 7).

[0278] Table 7 Patient characteristics

[0279]

[0280]

[0281] Sample collection and processing.

[0282] Before any type of therapeutic intervention, 50-100mL of discharged urine is obtained from each subject. According to standard procedures, 50 milliliters of urine are used for clinical laboratory analysis (e.g., urine cytology and urinalysis). Before immediate laboratory processing, a unique identification number is assigned to the remaining urine aliquot. Each urine sample is centrifuged at 4 ° C for 5 minutes with 600 × g. The supernatant is decanted and aliquoted, and the urine precipitate is quickly frozen at the same time. Before analysis, both supernatant and precipitate are stored at -80 ° C. The aliquots of urine supernatant are thawed, and the protein content is analyzed using Pierce 660-nm protein assay kit (Thermo Fisher Scientific Inc., Waltham, MA, USA, Waltham, Massachusetts, USA).

[0283] Multiple testing.

[0284] The assays described herein were validated using the multiplex bead-based immunoassay for Luminex 200 from R&D Systems Inc. (Minneapolis, MN). ) to monitor the concentration of 10 proteins (A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA). Urine samples were passively thawed and centrifuged 10 minutes x 1,000g. Urine samples were passively thawed and processed on ice before being diluted 2-fold with R&D assay diluent. Samples, standards, and controls (50 μl) were added in duplicate to a 96-well plate. The multiplex immunoassay was performed according to the manufacturer's instructions. A seven-point standard curve spanning the 4-log dynamic range of the assay was included in the current assay design. Plates were read on a 100 / 200 Luminex Corp, Austin, TX. Calibration curves were generated with the best fit combining Akaike's information criteria (AIC) values.

[0285] Statistical analysis.

[0286] Fisher's exact test determined the association between key demographic characteristics (age, sex, race, cytology) and cancer status. The previous 10 biomarker molecular signatures and previous molecular signatures with three demographic variables (age, sex, and race) were applied to these new data and the fit was assessed. Statistical significance in this study was set at p<0.05, and all reported p values ​​were bilateral. All analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, Cary, NC).

[0287] result

[0288] According to the demographics of the participating institutions, many of the study subjects were elderly Asian males. Cytology was available in 12 of the 31 cancer groups and 19 of the 41 controls, with a reported sensitivity of 58.3% and a specificity of 100%. No subject had abnormal axial imaging in the control group. In addition, in follow-up, no control subjects were noted to develop urothelial carcinoma or gross hematuria events. In the cancer group, 48.4% of the subjects suffered from non-muscle invasive disease, and 41.9% of the subjects suffered from low-grade disease. The demographic, clinical and pathological characteristics of the group are presented in Table 7.

[0289] The concentrations of ten proteins measured in urine are presented in Table 8.

[0290]

[0291] Median urine levels of MMP9 (50,945 vs. 5,705), IL8 (173.3 vs. 110.4), VEGF-A (484.1 vs. 218.8), CA9 (188.8 vs. 14.5), SDC1 (12,187 vs. 8810), PAI1 (1,428 vs. 113.8), ApoE (17,692 vs. 1,939), A1AT (659,816 vs. 99,455), ANG (1,765 vs. 1,047), and MMP10 (1,025 vs. 39.3) were higher in subjects with UTUC compared to controls, with CA9, ApoE, and A1AT reaching significance. Box plots showing the levels of the 10 biomarker levels in cases and controls are presented in FIG. 16 .

[0292] The ability of 10 biomarkers (A1AT, APOE, ANG, CA9, IL8, MMP9, MMP10, PAI1, SDC1, and VEGFA) to predict the presence of UTUC was analyzed using Youden index cutoffs and nonparametric ROC analysis. The 10-biomarker signature achieved an AUC of 0.897 (95% CI: 0.817-0.977) with an overall sensitivity of 93.5%, a specificity of 75.6%, an NPV of 93.9%, a PPV of 74.4%, and an accuracy of 83.3%.

[0293] Table 9 shows the use of low-grade and high-grade as well as non-muscle-invasive bladder cancer and muscle-invasive bladder cancer Overall sensitivity and specificity achieved by the hybrid signature.

[0294]

[0295] discuss

[0296] Patients suspected of harboring UTUC are typically evaluated with computed tomography (CT), retrograde pyelography (RGP), and upper tract cytology. Imaging studies often demonstrate filling defects or obstructive masses, which are often associated with hydronephrosis, hydroureter, or nephrolithiasis. Although noninvasive, imaging studies, which often have low sensitivity, are not used as the sole diagnostic test for UTUC because there are many reasons to explain filling defects other than UTUC (Chlapoutakis et al. "Europe Eur J Radiol .(2010)73:334–8). Urovysion (Sassa et al. American Journal of Clinical Pathology Journal (Am J Clin. Pathol.) (2019) 151(5):469-478), NMP22 (Yafi et al. Urologic Oncology: Seminars and Original Investigations. 2015; 33(2):66.e25-66.e31), and ImmunoCyt (Lodde et al. Urology 58:362-366) have all been reported to have low sensitivity, low specificity, or both in detecting UTUC, and therefore, cystoscopy and ureteroscopy and renal washout or biopsy (all invasive procedures) are the standard of care for evaluating patients with UTUC. Due to the above shortcomings, there is an urgent need to identify sensitive non-invasive molecular markers for detecting UTUC.

[0297] The multiplexed assays described herein have advantages, including reduced costs through lower labor requirements and reagent consumption, and the generation of more data with fewer samples, but the main advantage is the potential to significantly increase clinical test sensitivity and specificity through the combination of multiple biomarkers. Recently, several groups have begun to identify diagnostic biomarker panels for potential bladder cancer applications; Hoque et al. reported methylation of four genes (CDKN2A, ARF, MGMT, GSTP1) ( American Cancer Research Journal (J Natl. Cancer Inst) (2006) 98(14):996-1004), Chung et al. reported the methylation of five selected target genes (MYO3A, CA10, NKX6-2, DBC1, SOX11) (Cancer Epidemiology, Biomarkers, and Prevention (2011) 20(7):1483-91), Hanke et al. reported two mRNA genes (ETS2, uPA) ( Clinical Chemistry Chem.)》(2007) 53(12):2070-7), Mengual et al. reported 12+2 mRNA genes (ANXA10, AHNAK2, CTSE, CRH, IGF2, KLF9, KRT20, MAGEA3, POSTN, PPP1R14D, SLC1A6, TERT and ASAM, MCM10) ( "clinical Cancer Research (2010) 16(9):2624-33). However, only Holyoake et al. from New Zealand reported this finding (Clin Cancer Res 2008 Feb 1;14(3):742-9) and five mRNA genes (CDC2, MDK, IGFBP5, HOXA13; also known as Cxbladder) TM ) with a reported sensitivity of 82% and a specificity of 85%. O'Sullivan et al. Journal of Urology .(2012)188(3):741-7; Kavalieris et al. Journal of Urology. (2017) 197(6):1419-1426. To date, none has been tested in UTUC, and like excreted urine cytology, such testing would require significant amounts of exfoliated tumor cells in the urine.

[0298] Briefly, the multiplexed assay described herein (i.e., liquid biopsy for the detection and management of bladder cancer) is derived from two complementary technologies; gene expression array analysis of exfoliated urothelial cells in urine and glycoproteomic profiling of urine supernatant. Using sophisticated bioinformatics, the two datasets were combined and a cancer-associated signature containing 19 candidate biomarkers was identified. The 19 candidate biomarkers were then reduced to 10 biomarkers using a series of validation cohorts: angiopoietin, ANG; apolipoprotein E, APOE; alpha-1 antitrypsin, A1AT; carbonic anhydrase 9, CA9; interleukin 8, IL8; matrix metallopeptidase 9, MMP9; matrix metallopeptidase 10, MMP10; plasminogen activator inhibitor 1, PAI1; syndecan 1, SDC1; and vascular endothelial growth factor A, VEGFA, and subsequently validated in several late-stage studies, achieving a diagnostic sensitivity of 85-93% and a specificity of 81-95%. The inventors tested the performance of the multiplex assay described herein in subjects with UTUC and noted a sensitivity of 93.5% and a specificity of 75.6%. The sensitivity is comparable to the reported 100% BC-Detection (five target mRNAs; ABL1, CRH, IGF2, UPK1B, ANXA10) was comparable, however the reported specificity was 16.7% (D'Elia et al. Advances in Urology Therapeutics (Ther Adv Urol.)(2022)14).

[0299] Interestingly, Table 9 depicts the diagnostic performance of the multiplex assay described herein in high-grade / low-grade and invasive / non-invasive UTUC. Regardless of grade or invasiveness, the multiplex assay described herein maintains a sensitivity above 88%. This, along with its high NPV of 93.9%, will allow it to be positioned as an exclusion test, i.e., a negative multiplex assay described herein will exclude patients who require cystoscopy with ureteroscopy and renal flushing or biopsy.

[0300] in conclusion

[0301] Diagnosis of UTUC can be difficult requiring invasive procedures to support a diagnosis of the presence of the disease. With advances in diagnostic technology, the development of an accurate and robust urine test for detecting UTUC will benefit both patients and the healthcare system. In a multi-institutional cohort study, the multiplex assay described herein achieved highly encouraging diagnostic performance. The test deploys established multiplex testing technology, enabling rapid uptake in clinical laboratories.

[0302] All documents cited in this specification are incorporated herein by reference as if each document was specifically and individually indicated to be incorporated by reference. The citation of any document is for its disclosure prior to the filing date and should not be construed as an admission that the present disclosure is not entitled to antedate such document by virtue of prior invention.

Claims

1. A method for predicting the likelihood of long-term survival in a bladder cancer patient, the method comprising: (a) obtaining a biological sample from a patient; (b) isolating mRNA from the biological sample; (c) determining the level of mRNA of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA in the biological sample; (d) normalizing the mRNA levels to the level of at least one reference mRNA transcript in the sample to provide normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA levels; (e) comparing the normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA levels to normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA levels in a reference bladder tumor sample; and (f) predicting the likelihood of long-term survival without bladder cancer recurrence, wherein increased levels of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, and VEFGA mRNA indicate a decreased likelihood of long-term survival without bladder cancer recurrence.

2. A method for detecting a bladder cancer biomarker, the method comprising: (a) obtaining a biological sample from a patient; (b) isolating RNA from the biological sample; and (c) determining the levels of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA mRNA in the biological sample.

3. A method for classifying test data, wherein the test data comprises RNA expression data, the method comprising: (a) accessing, using at least one processor, a set of electronically stored training data vectors, each training data vector representing an individual cancer patient and comprising RNA expression data for the corresponding cancer patient, each training data vector further comprising a classification regarding the expression level of a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; (b) training an electronic representation of a classification system using the electronically stored set of training data vectors; (c) receiving, at the at least one processor, test data comprising RNA expression data; (d) using the at least one processor, evaluating the test data using the electronic representation of the classification system; as well as (e) outputting a classification of the test data regarding the likelihood of long-term survival without recurrence of bladder cancer based on the evaluating step.

4. The method of claim 3, wherein the classification system is AdaBoost, an artificial neural network (ANN) learning algorithm, a Bayesian belief network, a Bayesian classifier, a Bayesian neural network, a boosted tree, a case-based reasoning, a classification tree, a convolutional neural network, a decision tree, deep learning, an elastic net, a fully convolutional network (FCN), a genetic algorithm, a gradient boosted tree, a k-nearest neighbor classifier, LASSO, a linear classifier, a naive Bayes, a neural network, a penalized logistic regression, a random forest, a ridge regression, a support vector machine, or a combination thereof.

5. The method according to claim 3 or 4, wherein the classification system is a collection of classification systems.

6. The method according to any one of claims 1 to 5, wherein the mRNA level is determined by microarray analysis, RNAseq, RT-PCR, RT-qPCR, quantitative PCR (qPCR), Northern blot analysis, dot blot, Southern blot analysis, RNA sequencing, fluorescence in situ hybridization (FISH), or a combination thereof.

7. The method according to any one of claims 1 to 6, wherein the mRNA is determined by quantitative PCR (qPCR).

8. The method according to any one of claims 1 to 7, wherein the determining step uses primers selected from the group consisting of: SEQ ID NO: 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 or a combination thereof.

9. The method according to any one of claims 1 to 8, wherein the determining step uses a primer pair selected from the group consisting of: SEQ ID NO: 1 and 2; 3 and 4; 5 and 6; 7 and 8; 9 and 10; 11 and 12; 13 and 14; 15 and 16; 17 and 18; 19 and 20; or a combination thereof.

10. The method according to any one of claims 1 to 9, wherein the determining step uses a labeled nucleic acid probe.

11. The method of claim 10, wherein the label is a radioactive label, a fluorescent label, an enzyme, a chemiluminescent label, a colorimetric label, or a combination thereof.

12. The method according to any one of claims 1 to 11, wherein the RNA is sequenced.

13. The method of any one of claims 1 to 12, wherein the biological sample is blood, serum, whole blood, circulating tumor cells, tumor cells, plasma, urine, tissue, tumor, or a combination thereof.

14. The method according to any one of claims 1 to 13, wherein the biological sample is a tissue, optionally a tumor tissue.

15. The method of claim 13 or 14, wherein the tissue is a fixed wax embedded tissue sample.

16. The method of any one of claims 1 to 15, wherein the levels of amplicons of RNA transcripts of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA are expressed as threshold cycle (Ct) values, and the normalized ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA amplicon levels are expressed as normalized Ct values.

17. The method of any one of claims 1 to 16, wherein the reference bladder cancer samples comprise at least 30 bladder cancer samples.

18. The method according to any one of claims 1 to 17, wherein the method further comprises detecting and quantifying at least one additional biomarker of a urogenital-related cancer type in the biological sample or in a different biological sample.

19. The method of any one of claims 1 to 18, wherein the method further comprises detecting and quantifying at least one additional biomarker of a different cancer type in the biological sample or in a different biological sample.

20. The method of any one of claims 1 to 19, wherein the method is performed at several time points or time intervals as part of monitoring the subject at least one of before, during and after treatment of the cancer.

21. The method of any one of claims 1 to 20, wherein the method further comprises the step of preparing a report indicating that the patient has an increased or decreased likelihood of long-term survival without bladder cancer.

22. A non-transitory computer-readable medium storing an executable program comprising instructions for executing the method according to any one of claims 1 to 21.

23. A system comprising: a server, wherein the server comprises at least one processor and a memory, wherein the memory comprises computer-readable instructions, and wherein the computer-readable instructions, when executed by the processor, cause the processor to perform the following steps: receiving mRNA expression data from a computer terminal located away from the server; and processing the mRNA expression data using a classification system.

24. A method for detecting a biomarker for upper tract urothelial carcinoma (UTUC), the method comprising: (a) obtaining a biological sample from a subject; (b) contacting a biological sample obtained from the subject with a set of binding agents, wherein the set comprises binding agents that bind to and form a complex with a protein selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; and (c) detecting the presence and amount of protein-binding agent complexes formed in the biological sample.

25. A method for classifying test data, wherein the test data comprises protein expression data, the method comprising: (a) accessing, using at least one processor, a set of electronically stored training data vectors, each training data vector representing an individual cancer patient and comprising protein expression data for the corresponding cancer patient, each training data vector further comprising a classification regarding the expression level of a biomarker selected from the group consisting of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof; (b) training an electronic representation of a classification system using the electronically stored set of training data vectors; (c) receiving, at the at least one processor, test data comprising protein expression data; (d) using the at least one processor, evaluating the test data using the electronic representation of the classification system; as well as (e) outputting a classification of the test data regarding the likelihood of upper tract urothelial carcinoma (UTUC) based on the evaluating step.

26. The method of claim 25, wherein the classification system is AdaBoost, an artificial neural network (ANN) learning algorithm, a Bayesian belief network, a Bayesian classifier, a Bayesian neural network, a boosted tree, a case-based reasoning, a classification tree, a convolutional neural network, a decision tree, deep learning, an elastic net, a fully convolutional network (FCN), a genetic algorithm, a gradient boosted tree, a k-nearest neighbor classifier, LASSO, a linear classifier, a naive Bayes, a neural network, a penalized logistic regression, a random forest, a ridge regression, a support vector machine, or a combination thereof.

27. The method of claim 24 or 25, wherein the classification system is a collection of classification systems.

28. The method of any one of claims 24 to 27, wherein the subject is diagnosed with UTUC.

29. The method of any one of claims 24 to 28, wherein the sample is obtained from a subject having at least one symptom of UTUC.

30. The method of any one of claims 24 to 29, wherein the biological sample is blood, serum, whole blood, circulating tumor cells, tumor cells, plasma, urine, tissue, tumor, or a combination thereof.

31. The method of any one of claims 24 to 30, wherein the biological sample is blood, urine, plasma, or a combination thereof.

32. The method of claim 30 or 31 , wherein the biological sample is urine.

33. The method of any one of claims 24 to 32, wherein the binding agent is an antibody or antibody fragment.

34. The method of claim 33, wherein the binding agent is an antibody.

35. The method of any one of claims 25 to 34, wherein the binding agent is a monoclonal antibody.

36. The method of any one of claims 25 to 34, wherein the binding agent is a polyclonal antibody.

37. The method of any one of claims 25 to 34, wherein the biomarkers consist of: ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1 and VEFGA.

38. An array comprising biomarkers selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and combinations thereof immobilized to a substrate.

39. The array of claim 38, wherein the biomarkers are mRNA transcripts.

40. The array of claim 38, wherein the biomarker is cDNA of the mRNA transcript.

41. The array of claim 38, wherein the biomarkers are peptides.

42. A kit comprising specifically binding nucleic acid primers comprising a biomarker selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination.

43. A kit comprising antibodies that specifically bind to a biomarker selected from the group consisting of ANG, A1AT, APOE, CA9, IL8, MMP9, MMP10, PAI-1, SDC1, VEFGA, and a combination.

Images