Method for predicting prognosis of colorectal cancer patient

Abstract

The present disclosure includes a method for predicting the prognosis of a colorectal cancer patient, the method including the measurement of the expression levels of genes of one or more gene groups selected from gene groups (a) to (e): (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12, and CSAG1, and (e) IGF2, NELL2, and RBMAS1 in a cancer sample from a patient; and a kit for predicting the prognosis of a colorectal cancer patient, the kit including a reagent for measuring the expression levels of genes of one or more gene groups selected from the gene groups (a) to (e); etc.

Description

Method for predicting prognosis of colorectal cancer patients

[0001] This application claims priority to Japanese Patent Application No. 2024-017258, the entire contents of which are incorporated herein by reference. The present disclosure includes a method and a kit for predicting the prognosis of a colorectal cancer patient.

[0002] Colorectal cancer (CRC) is the third most common cancer in humans and the second most deadly. Most deaths from CRC are due to metastasis to major organs such as the liver and lungs. As exemplified by the Consensus Molecular Subtyping (CMS) classification, mRNA expression in numerous CRC tissue samples has been analyzed and classified. Most of these analyses have used fresh frozen CRC tissue samples or paraffin-embedded blocks. These samples are composed of "cancer epithelium" and "stroma," which includes blood cells, vascular endothelial cells, inflammatory cells, immune cells, and fibroblasts. This complicates the data, and phenotypic changes in the cancer epithelium itself also affect gene expression. Patient-derived xenografts (PDXs) have also been used to identify CRC "epithelial cell-intrinsic subtypes" (CRIS). This is because stromal cells derived from human tumors are replaced with mouse stromal cells, resulting in a human-derived epithelium. In addition, multiple regions of the same tumor have been analyzed to avoid heterogeneity within individual tumors. However, the prognosis of colorectal cancer patients using gene expression analysis has not yet been fully established.

[0003] The present disclosure aims to provide a method and kit for predicting the prognosis of a colorectal cancer patient.

[0004] In one aspect, the present disclosure provides a method for predicting the prognosis of a patient with colorectal cancer, comprising measuring expression levels of genes from one or more of the following gene groups (a) to (e) selected from a cancer sample from the patient: (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16; (b) DEFA6, DEFA5, and SPINK1; (c) BST2, BATF, RAMP1, and TPM2; (d) MAGEA6, MAGEA3, MAGEA12, and CSAG1; and (e) IGF2, NELL2, and RBMS1.

[0005] In one aspect, the present disclosure provides a kit for predicting the prognosis of a colorectal cancer patient, the kit comprising reagents for measuring the expression levels of genes in one or more gene groups selected from the group of genes (a) to (e): (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12 and CSAG1, and (e) IGF2, NELL2, and RBMS1.

[0006] The present disclosure provides methods and kits for predicting the prognosis of a colorectal cancer patient.

[0007] Flowchart for the discovery of colorectal cancer stem cell (CRC-SC)-specific expression signatures. A, A panel of patient-derived CRC-SCs consisting of 20 previous lines {Yamamoto, 2020} and 37 additional lines. B, Genomic DNA was analyzed for 409 or 50 cancer-associated gene mutation hotspots. C, Five normal colorectal epithelial stem cell (NCE-SC) lines (3N, 50N, 59N, 70N, and 73N) were used as controls. D, Transcriptome analysis was performed using microarray analysis. E and F, Low- and high-expression genes were determined by fold change in expression levels of CRC-SC lines relative to that of control NCE-SC lines. G and H, Search and identification of additional genes within each lead gene subgroup for inclusion in each lead gene signature. Expression levels of the additional genes correlated with each other and with CRC patient survival in three public databases. IL, Finally, the individual normalized signatures were summed, with positive (+) representing a good prognosis and negative (-) representing a poor prognosis, to create a comprehensive colon cancer signature (GCS). M, Preclinical evaluation of the GCS score was performed by transplanting CRC-SC spheroids into the colon of immunodeficient mice. Logarithmic plot showing the rank order of the mean mRNA expression level (Mean TPM) of each gene in three NCE-SC lines. Circles in the figure represent the following gene groups: NCE-SC markers (light gray), lucky 5 genes (white), CRC-SC markers (dark gray), and lead genes (CXCL14, BST2, DEFA6, IGF2, and MAGEA6). Numbers in parentheses indicate the rank order of mRNA expression levels in NCE-SC spheroid cells. ND, below the limit of detection. Single-cell RNA expression analysis of the 70N-NCE-SC line. The cell cycle progression in each region of the single cell population shown in the left panel was determined using Metascape expression analysis {Zhou et al., 2019}. The nine panels on the right show the expression of marker genes (dark gray) in that cell population: MKI67 and PCNA; cell proliferation markers; CD24, CD44, and BMI1; intestinal epithelial stem cell markers; EPCAM; epithelial cell marker.MUC12, PIGR, and PLA2G2A; lucky 5 genes. Cell-type signature enrichment analysis using the top 100 genes highly expressed in NCE-SC under spheroid culture conditions relative to organoid culture conditions. Gene expression in CRC-SC lines compared to NCE-SC lines. (Top left) Flowchart of this study. Transcriptional activity in NCE-SC and CRC-SC spheroids was measured as mRNA expression levels using microarrays. (Top right) Expression levels of CRC-SC relative to NCE-SC expression levels. Gene expression levels are indicated by circles at 1og2 scale. Dark gray, white, and light gray circles represent top, bottom, and lead genes, respectively. Numbers in parentheses indicate the rank order of highly (+) or lowly (-) expressed genes in CRC-SC spheroid lines relative to the reference NCE-SC spheroid line. (Bottom) Heatmap showing the mRNA expression levels of lucky 5 genes and other candidate genes in 57 CRC-SC spheroid lines (1T to 6T) and 5 NCE-SC spheroid lines (3N to 37N; 3N-d in the leftmost column is a cell line differentiated in vitro from 3N by removing the Wnt ligand-containing medium). The squares in the top row indicate the subgroups with the highest relative expression values, as shown in Table 2. Violin plots showing the microarray probe signal values ​​for 36 genes (horizontal axis) in 57 CRC-SC spheroid lines relative to the mRNA expression levels of NCE-SC lines. Lines surrounded by rounded rectangles represent the subgroup with high expression of lucky 5 genes. Lines surrounded by ellipses represent the subgroup represented by the lead genes DEFA6, BST2, MAGEA6, and IGF2. The horizontal wavy line at 0 indicates the average probe signal value for NCE-SC lines. Overall survival (OS), disease-specific survival (DSS), and progression-free survival (PFS) for colorectal cancer patients (all stages) in the TCGA-CRC database (validation cohort 1). The graph on the right shows the hazard ratio (HR) and its 95% confidence interval (CI).OS or recurrence-free survival (RFS) of colorectal cancer patients (all stages) in the GSE39582 database (validation cohort 2). The graphs show the survival times of patients with high (H) or low (L) expression of the lucky 5 gene. The graphs on the right show the HRs and their 95% CIs. OS, DSS, PFS, or RFS of colorectal cancer patients (all stages) in TCGA-CRC (top) or GSE39582 (bottom). The graphs show the survival times of patients with high (High) or low (Low) expression of the lucky 5 signature. The graphs on the right show the HRs and their 95% CIs. Heatmap showing mRNA expression levels in 57 CRC-SC spheroid lines (55T to 83T) and 5 NCE-SC lines (3N to 73N; 3N-d in the leftmost column is a cell line differentiated in vitro from 3N by removing the medium containing Wnt ligand). The boxes in the top row represent colorectal cancer subgroups and indicate the five lead genes with the highest expression in those CRC-SC spheroids (see "Keys" below the figure). OS, DSS, or PFS for colorectal cancer patients (all stages) in the TCGA-CRC database. This shows the survival time of patients with high or low scores for the four lead gene signature. The graph on the right shows the HR and 95% CI for the five signatures, including the lead gene candidate CXCL14. OS or RFS for colorectal cancer patients (all stages) in the GSE39582 database. This shows the survival time of patients with high or low scores for the four lead gene signature. The graph on the right shows the HR and 95% CI for the five signatures, including the lead gene candidate CXCL14. Flowchart for calculating the colorectal cancer comprehensive signature (GCS). a) The five signatures were classified into two categories: long-term survival signatures and short-term survival signatures. b and c) The mean Z scores of the long-term and short-term survival signatures were treated as positive (+) and negative (-) scores, respectively. d) The signature scores were summed to calculate the GCS score. e) A retrospective observational study of the GCS score was performed using the TCGA-CRC-DSS, TCGA-CRC-PFS, or GSE39582-RFS datasets as validation cohorts.(f) Finally, in a prospective mouse model study, CRC-SC lines were divided into Q1-Q4 and orthotopically implanted into the cecal mucosa of NSG mice. After 26 weeks, the primary tumors and liver and lung metastases were examined. Kaplan-Meier plots (left) and HRs (right) show comparisons of DSS or PFS in colorectal cancer patients from the TCGA-CRC database divided by GCS quartiles (Q1-Q4, from lowest to highest GCS score). Ptrend and P values ​​were calculated using the log-rank trend test and log-rank test, respectively. HRs (95% CI) were obtained using the Mantel-Haenszel test. The sample sizes for DSS were Q1: 0-25%, n = 137; Q2: 25-50%, n = 136; Q3: 50-75%, n = 136; and Q4: 75-100%, n = 136. The sample sizes for PFS were Q1 (0-25%, n = 141); Q2 (25-50%, n = 141); Q3 (50-75%, n = 141); and Q4 (75-100%, n = 141). Kaplan-Meier plots (left) and HR comparisons (right) show the RFS comparisons for colorectal cancer patients in the GSE39582 database divided by GCS quartiles (Q1-Q4, from lowest to highest GCS score). The sample sizes were Q1 (0-25%, n = 129); Q2 (25-50%, n = 130); Q3 (50-75%, n = 130); and Q4 (75-100%, n = 130). A radar chart shows the contribution of the five signatures to the GCS score. The results are shown for two colorectal cancer patients (D5-6920 and AG-4021 or CIT478 and CIT295) from TCGA-CRC or GSE39582 whose GCS scores were in the top 25% (Q4) (left) and bottom 25% (Q1) (right) of the total patient population. The numbers -4 to 4 indicate the average Z-score of each signature used in GCS. The good prognostic signatures Lucky 5 and DEFA6 are shown on the bottom, while the poor prognostic signatures IGF2, BST2, and MAGEA6 are shown on the top. The GSC statistics (top) and histograms (bottom) show the distribution of GSCs in CRC-SC spheroids, TCGA-CRC, or GSE39582.Upper left: Kaplan-Meier plot comparing GCS score quartiles of CRC-SC lines orthotopically transplanted into NSG mice. (Number of mice: Q1 = 14, Q2 = 15, Q3 = 14, Q4 = 14.) Upper center: Kaplan-Meier plot comparing GCS scores of Q4 and Q1. Ptrend and P values ​​were calculated using the log-rank test for trend and the log-rank test, respectively. Lower left: Cancer mortality rates in mice orthotopically transplanted with CRC-SC lines. Lower center: Number of cancer deaths in mice orthotopically transplanted with CRC-SC lines. Right: Liver and lung metastases resulting from orthotopic transplantation of two lines from the lower GCS score group (Q1), 8T and 13T, into mice. These tumor-bearing mice died within 26 weeks after transplantation. Upper right: Macroscopic findings of resected livers. Middle and lower right: Histopathological observations of liver and lungs stained with H&E. White scale bar: 10 mm. Black scale bar: 400 μm. Arrowheads represent metastatic lesions. A test was conducted to determine whether NSG mice orthotopically transplanted with CRC-SC lines with GCS scores of Q4 to Q1 died 6 months later. Column L indicates cell lines with high or low lucky 5 signatures, and column P indicates cell lines in which the mice died (Lethal) or survived (Viable). Asterisks indicate cell lines with mucinous histopathology. Flowchart of patient subtype determination using maximum signature score (MSS). a, The maximum score of the four lead gene signature (LGS) scores was defined as the MSS. b, Each patient was divided into lead gene subgroups based on the MSS. ce, Each subgroup was divided into subpopulations with high or low MSS. Patients with high scores were used as the subtype prototype, and patients with low scores from each subgroup were combined into one. A retrospective observational study was conducted using the TCGA-CRC-PFS or GSE39582-RFS dataset as a validation cohort. Heatmaps showing the subtype, stage, and LGS score of CRC patients in CRC-SC (top), TCGA-CRC (middle), and GSE39582 (bottom). Kaplan-Meier plots of PFS or RFS by subtype for colorectal cancer patients (all stages) in TCGA-CRC or GSE39582, and HRs between each LGS subtype and the rest of the patients.Comparison of GCS and ColoGuidePro in terms of RFS for colorectal cancer patients (all stages). Shown are the RFS for patients with Q1-Q4 (left), the highest and lowest scores (Q1 and Q4) (center), or a favorable (80%) and unfavorable (20%) prognosis (right). Note that the ranking of Q1-Q4 is reversed between GCS and ColoGuidePro. Comparison of GCS and ColoGuidePro in terms of RFS for colorectal cancer patients (stage II). Shown are the RFS for patients with Q1-Q4 (left), the highest and lowest scores (Q1 and Q4) (center), or a favorable (80%) and unfavorable (20%) prognosis (right). Comparison of GCS and ColoGuidePro in terms of RFS for colorectal cancer patients (stage III). Shown are the RFS for patients with Q1-Q4 (left), the highest and lowest scores (Q1 and Q4) (center), or a favorable (80%) and unfavorable (20%) prognosis (right).

[0008] Unless otherwise specified, terms used herein have the meanings commonly understood by those skilled in the art of organic chemistry, medicine, pharmacology, molecular biology, microbiology, etc. Definitions of some terms used herein are provided below, but these definitions take precedence over common understandings in this specification.

[0009] In one aspect, the present disclosure relates to a method for predicting the prognosis of a colorectal cancer patient. The prognosis prediction method of the present disclosure comprises measuring the expression levels of genes in one or more gene groups selected from gene groups (a) to (e) shown in Table 1. The expression of genes in each gene group (a) to (e) in colorectal cancer cells is correlated with each other, and each gene group forms a signature. These gene signatures may exhibit stronger statistical significance as a prognostic predictor for colorectal cancer patients than a single gene included in that gene group. Genbank accession numbers for representative mRNA sequences and amino acid sequences encoded by each gene are shown in Table 1. The sequences of each gene in individual colorectal cancer patients are not limited to these sequences and may include natural mutations.

[0010] The expression level of a gene can be measured by measuring the expression of the mRNA or protein encoded by that gene. mRNA or protein expression may be measured by any method. Measurement methods include polymerase chain reaction (PCR), RNA sequencing (RNA-Seq), microarrays, and immunological methods such as immunostaining, ELISA, Western blot, and dot blot. As used herein, the term "PCR" refers to any PCR-based method, including reverse transcription PCR (RT-PCR), quantitative PCR (qPCR), and real-time PCR. Those skilled in the art can appropriately process cancer samples according to the measurement method and can appropriately prepare or select reagents, such as primers, probes, or antibodies, suitable for measurement based on the mRNA sequence or amino acid sequence encoded by each gene. In one embodiment, the expression level of a gene is measured by measuring mRNA expression levels.

[0011] The primers or probes may comprise or consist of a sequence complementary to a portion of the mRNA sequence encoded by each gene. The primers or probes may be, for example, 10 to 50 nucleotides, 15 to 35 nucleotides, or 20 to 30 nucleotides. In one embodiment, the primers or probes comprise or consist of a sequence of 10 to 50 nucleotides, 15 to 35 nucleotides, or 20 to 30 nucleotides complementary to a portion of the mRNA sequence encoded by each gene. In one embodiment, the primers or probes comprise or consist of a sequence of 10 to 50 nucleotides, 15 to 35 nucleotides, or 20 to 30 nucleotides complementary to a portion of any sequence selected from SEQ ID NOs: 1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, and 37.

[0012] In one embodiment, the gene expression levels are measured by PCR using primers specific to each gene. PCR reactions for each gene may be performed in separate vessels, or reactions for multiple genes may be performed in a single vessel. Those skilled in the art can appropriately determine suitable PCR conditions.

[0013] The large intestine is divided into the colon and the rectum, and colorectal cancer includes colon cancer and rectal cancer. Colorectal cancer is divided into stages 0, I, II, III, and IV depending on the degree of progression. Colorectal cancer in the present disclosure includes colorectal cancer of any stage. In one embodiment, the colorectal cancer patient is a patient with stage I, II, III, or IV. In one embodiment, the colorectal cancer patient is a patient with stage II, III, or IV. In one embodiment, the colorectal cancer patient is a patient with stage III or IV.

[0014] The cancer sample may be any sample that can be used to examine gene expression levels in a patient's colon cancer tissue. The cancer sample may be, for example, a surgical sample, a biopsy sample, or a cancer spheroid. In one embodiment, the surgical sample is a sample obtained at the time of primary tumor resection. As used herein, "cancer spheroid" refers to a cell mass containing tumor-initiating cells, and "tumor-initiating cells" refers to cells that form tumors when transplanted into an immunodeficient animal. Cancer spheroids are created from a patient's cancer cells or tissue. Cancer spheroids may be created by any method, for example, by the method described in International Publication No. WO 2019 / 111998.

[0015] In one embodiment, the method of the present disclosure calculates a Z-score (Z) for each gene from the expression level of each gene in a patient using formula (I): (where x is the expression level of the gene in the patient, and μ and σ are the mean and standard deviation of the expression levels of the gene in a plurality of colorectal cancer patients, respectively); determining a gene expression signature score for each gene group; and comparing the gene expression signature score to a cutoff value.

[0016] μ and σ are the mean and standard deviation of the expression levels of the gene in question in multiple colorectal cancer patients, and are predetermined values ​​for each gene. The μ and σ values ​​can be determined, for example, from the gene expression levels of multiple colorectal cancer patients measured using the same method as used to measure the patient's gene expression levels. Alternatively, the μ and σ values ​​can be determined based on gene expression data of multiple colorectal cancer patients registered in a database. Available databases include TCGA-CRC (The Cancer Genome Atlas) (https: / / www.cancer.gov / about-nci / organization / ccg / research / structural-genomics / tcga) and GSE39582 in the Gene Expression Omnibus (GEO) of the National Center for Biotechnology Information (NCBI) (https: / / www.ncbi.nlm.nih.gov / geo / ). The μ and σ values ​​can also be determined based on gene expression data of cancer spheroids from multiple colorectal cancer patients. The number of colorectal cancer patients may be, for example, but is not limited to, 50 to 1000, 50 to 500, 50 to 200, or 50 to 100. Those skilled in the art can appropriately determine the number of cases required, and can also correct the values ​​of μ and σ when the number of cases increases (e.g., each time the number of cases doubles). The gene expression levels used to calculate the Z-score may be values ​​that reflect the expression level of each gene, and may be measured values, relative values, or logarithms of these, or may be appropriately corrected or standardized values.

[0017] The gene expression signature score is a value indicating the expression level of the genes in each gene group, and is the sum of the Z scores of the genes in the gene group, or the average Z score of the genes in the gene group. For example, the gene expression signature score of gene group (a) (lucky 5 signature) can be determined by adding up the Z scores of five genes, MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, or by dividing the sum of the Z scores of these genes by 5, which is the number of genes. In one embodiment, the gene expression signature score is the average Z score of the genes in each gene group.

[0018] The gene expression signature score is compared with a cutoff value. The cutoff value is a value of the gene expression signature score that can divide multiple colorectal cancer patients with known prognosis into a good prognosis group and a poor prognosis group with a statistically significant difference. The cutoff value can be, for example, the value that minimizes the P value when multiple colorectal cancer patients are divided into a good prognosis group and a poor prognosis group in a survival curve according to the Kaplan-Meier method. The statistical significance between the good prognosis group and the poor prognosis group can be determined, for example, by a logrank test.

[0019] The gene expression signature score may be compared with multiple cutoff values. For example, for multiple colorectal cancer patients whose prognosis is known, the prognosis of each patient may be ranked into three or more levels, and two or more gene expression signature scores that can separate each rank group with a statistically significant difference may be used as cutoff values. Statistical significance can be determined, for example, by the log-rank test. Alternatively, the cutoff values ​​may be multiple values ​​that can separate multiple colorectal cancer patients at a certain ratio in order of their gene expression signature scores, for example, the quartiles of the gene expression signature scores of multiple colorectal cancer patients. By comparing with multiple cutoff values, colorectal cancer patients can be stratified according to their prognosis.

[0020] The comparison of the gene expression signature score with the cutoff value may or may not involve statistical analysis, such as one-way ANOVA, two-way ANOVA, Dunnett's test, or Tukey's test.

[0021] The gene expression signature score of the gene group (a) is compared with a cutoff value for the gene group. If the gene expression signature score of the gene group (a) is higher than the cutoff value, the patient is predicted to have a good prognosis.

[0022] The gene expression signature score of gene group (b) is compared with a cutoff value for the gene group. If the gene expression signature score of gene group (b) is higher than the cutoff value, the patient is predicted to have a good prognosis.

[0023] The gene expression signature score of the gene group (c) is compared with a cutoff value for the gene group. If the gene expression signature score of the gene group (c) is higher than the cutoff value, the patient is predicted to have a poor prognosis.

[0024] The gene expression signature score of the gene group (d) is compared with a cutoff value for the gene group. If the gene expression signature score of the gene group (d) is higher than the cutoff value, the patient is predicted to have a poor prognosis.

[0025] The gene expression signature score of the gene group (e) is compared with a cutoff value for the gene group. If the gene expression signature score of the gene group (e) is higher than the cutoff value, the patient is predicted to have a poor prognosis.

[0026] In one embodiment, the method of the present disclosure comprises measuring the expression levels of genes in two gene groups selected from gene groups (a) to (e). In one embodiment, the method of the present disclosure comprises measuring the expression levels of genes in three gene groups selected from gene groups (a) to (e). In one embodiment, the method of the present disclosure comprises measuring the expression levels of genes in four gene groups selected from gene groups (a) to (e). In one embodiment, the method of the present disclosure comprises measuring the expression levels of genes in five gene groups selected from gene groups (a) to (e).

[0027] In one embodiment, the method of the present disclosure includes measuring the expression levels of genes in the gene groups (a) to (e); and calculating a colorectal cancer comprehensive signature (GCS) score according to formula (II): (In the formula, is a coefficient indicating the polarity of prognosis, where the coefficient of gene group (a) to (b) is 1, the coefficient of gene group (c) to (e) is -1, is the average Z-score of the genes in each gene group), and comparing the GCS score to a cutoff value.

[0028] The GCS score is a comprehensive signature that integrates five types of signatures from gene groups (a) to (e). is a coefficient indicating the polarity of the prognosis of each gene group (i.e., each gene signature). The coefficient of gene groups (a) to (b) correlated with a good prognosis is 1, and the coefficient of gene groups (c) to (e) correlated with a poor prognosis is -1. is the average Z score of the genes in each gene group, and is calculated by adding up the Z scores of the genes in each gene group and dividing by the number of genes. In other words, the GCS score is calculated for gene groups (a) to (e) by multiplying the average Z score of the genes in each gene group by a coefficient indicating the polarity of prognosis and then adding these values ​​together.

[0029] The GCS score is compared with a cutoff value. The cutoff value is a GCS score value that can statistically significantly divide multiple colorectal cancer patients with known prognosis into a good prognosis group and a poor prognosis group. The cutoff value can be, for example, the value that minimizes the P value when multiple colorectal cancer patients are divided into a good prognosis group and a poor prognosis group in a Kaplan-Meier survival curve. The statistical significance between the good prognosis group and the poor prognosis group can be determined, for example, by the log-rank test. If the colorectal cancer comprehensive score is higher than the cutoff value, the patient is predicted to have a good prognosis.

[0030] The GCS score may be compared with multiple cutoff values. For example, for multiple colorectal cancer patients whose prognosis is known, the prognosis of each patient may be ranked into three or more levels, and two or more colorectal cancer comprehensive scores that can separate each ranked group with a statistically significant difference may be used as cutoff values. Statistical significance can be determined, for example, by the log-rank test. Alternatively, the cutoff values ​​may be multiple values ​​that can separate multiple colorectal cancer patients at a certain ratio in order of GCS score, such as the quartiles of the colorectal cancer comprehensive scores of multiple colorectal cancer patients. By comparing with multiple cutoff values, colorectal cancer patients can be stratified according to their prognosis.

[0031] The comparison of the GCS score with the cutoff value may or may not involve statistical analysis, such as one-way ANOVA, two-way ANOVA, Dunnett's test, or Tukey's test.

[0032] Treatment guidelines for colorectal cancer stipulate treatment strategies for each stage. For example, chemotherapy is generally not recommended for stage II cancer, but chemotherapy can be administered if the method of the present disclosure predicts a poor prognosis. Furthermore, chemotherapy is generally recommended for stage III cancer, but if the method of the present disclosure predicts a good prognosis, chemotherapy can be avoided or a drug with fewer side effects can be used. If the method of the present disclosure predicts a poor prognosis, a drug with stronger effects can be used. In other words, the prognosis prediction method of the present disclosure contributes to optimizing treatment methods.

[0033] Thus, a colorectal cancer patient whose prognosis has been predicted by the method of the present disclosure can be treated with a therapeutic agent or therapy selected based on the prediction result. Therefore, in one aspect, the present disclosure provides a method for treating a colorectal cancer patient, the method comprising predicting the prognosis of the colorectal cancer patient by the prognosis prediction method of the present disclosure, and administering to the patient a therapeutic agent or therapy selected based on the prediction result. The therapeutic agent includes anticancer drugs such as 5-fluorouracil (5-FU), tegafur, tegafur-uracil (UFT), doxifluridine (5'-DFUR), carmofur, tegafur-gimeracil-oteracil potassium (S-1), capecitabine (Cape), regorafenib, trifluridine-tipiracil hydrochloride (FTD / TPI), mitomycin C, irinotecan (IRI), oxaliplatin (OX), and vascular endothelial growth factor (VGF). Examples of molecular targeted drugs include VEGF inhibitors (e.g., bevacizumab, aflibercept), vascular endothelial growth factor receptor (VEGFR) inhibitors (e.g., ramucirumab), FGFR inhibitors (e.g., erdafitinib, rogaratinib, AZD4547, infigratinib, fisogatinib, LY2874455, futibatinib, ASP5878, derazantinib), and EGFR inhibitors (e.g., cetuximab, panitumumab, erlotinib). Treatments include chemotherapy such as FOLFOX therapy (intravenous 5-FU + levofolinate + OX), FOLFIRI therapy (5-FU + levofolinate + IRI), FOLFOXIRI therapy (5-FU + levofolinate + OX + IRI), CapeOX therapy (Cape + OX), SOX therapy (S-1 + OX), and IRIS therapy (S-1 + IRI), as well as endoscopic resection, surgical resection, and radiation therapy.

[0034] In one aspect, the present disclosure relates to a kit for predicting the prognosis of a colorectal cancer patient, the kit comprising reagents for measuring the expression levels of genes in one or more gene groups selected from the group of genes (a) to (e): (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12 and CSAG1, and (e) IGF2, NELL2, and RBMS1.

[0035] The reagents can be prepared or selected based on the mRNA sequence or amino acid sequence encoded by each gene, as described in the prognosis prediction method of the present disclosure. In one embodiment, the reagents are primers or probes. In one embodiment, the reagents are PCR primers. In addition to the reagents, the kit may also include a measurement container, a buffer solution, instructions for use, etc.

[0036] Exemplary embodiments of the present disclosure are described below. [1] A method for predicting the prognosis of a colorectal cancer patient, comprising measuring the expression levels of genes in a group of one or more genes selected from the group of genes (a) to (e) in a cancer sample from the patient: (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12, and CSAG1, and (e) IGF2, NELL2, and SPINK1. [2] The Z-score (Z) of each gene is calculated using formula (I): (wherein x is the expression level of the gene in the patient, and μ and σ are the mean and standard deviation of the expression levels of the gene in a plurality of colorectal cancer patients, respectively); determining a gene expression signature score for each gene group, wherein the gene expression signature score for each gene group is the sum of the Z-scores of the genes in each gene group or the mean of the Z-scores of the genes in each gene group; and comparing the gene expression signature score to a cutoff value. [3] The method of 2 above, wherein the gene expression signature score for each gene group is the mean of the Z-scores of the genes in each gene group. [4] The method of either 2 or 3 above, comprising comparing the gene expression signature score of gene group (a) to a cutoff value, wherein the patient is predicted to have a good prognosis if the gene expression signature score is higher than the cutoff value. [5] The method according to any one of 2 to 4 above, comprising comparing the gene expression signature score of gene group (b) with a cutoff value, and predicting that the patient will have a good prognosis if the gene expression signature score is higher than the cutoff value. [6] The method according to any one of 2 to 5 above, comprising comparing the gene expression signature score of gene group (c) with a cutoff value, and predicting that the patient will have a poor prognosis if the gene expression signature score is higher than the cutoff value. [7] The method according to any one of 2 to 6 above, comprising comparing the gene expression signature score of gene group (d) with a cutoff value, and predicting that the patient will have a poor prognosis if the gene expression signature score is higher than the cutoff value. [8] The method according to any one of 2 to 7 above, comprising comparing the gene expression signature score of gene group (e) with a cutoff value, and predicting that the patient will have a poor prognosis if the gene expression signature score is higher than the cutoff value. [9] The method according to any one of 1 to 8, wherein the one or more gene groups are gene groups (a) to (e).

[10] The method according to any one of 1 to 8, wherein the one or more gene groups are gene groups (a) to (e), and the colorectal cancer comprehensive signature (GCS) score is calculated using formula (II): (In the formula, is a coefficient indicating the polarity of prognosis, where the coefficient of gene group (a) to (b) is 1, the coefficient of gene group (c) to (e) is -1,

[11] The method according to any one of 1 to 10 above, wherein the cancer sample is a surgical sample or a biopsy sample.

[12] The method according to any one of 1 to 10 above, wherein the cancer sample is a cancer spheroid.

[0037]

[13] A kit for predicting the prognosis of a colorectal cancer patient, comprising reagents for measuring the expression levels of genes in one or more gene groups selected from gene groups (a) to (e): (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12 and CSAG1, and (e) IGF2, NELL2, and RBMS1.

[14] The kit according to 13, wherein the one or more gene groups are gene groups (a) to (e).

[15] The kit according to 13 or 14, wherein the reagents are primers or probes.

[0038] 13. A method for treating a patient with colorectal cancer, comprising: predicting the prognosis of the patient with colorectal cancer by the method described in any one of 1 to 12; and administering to the patient a therapeutic agent or therapy selected based on the prediction result.

[0039] The present invention will be further described below with reference to examples, but the present invention is not limited to the following examples in any sense.

[0040] 1. Materials and Methods Human Specimens A total of 216 tumor specimens were collected from 208 patients with colorectal cancer who underwent primary tumor resection at Kyoto University Hospital (KUHP) between October 2014 and August 2018. Their diagnoses were confirmed as well-, moderately, or poorly differentiated, mucinous, or papillary colorectal cancer through histopathological examination by a pathologist at KUHP. The research protocol for human specimens was approved by the Medical Ethics Committee of Kyoto University, Kyoto, Japan (approval numbers R0915 and R0857, 2015 to 2024), and patients provided written informed consent for the use of their specimens and data analysis.

[0041] Isolation and culture of patient-derived CRC-SCs as spheroids was previously described {Miyoshi et al., 2018}. Briefly, two tumor tissue fragments (approximately 1 cm each) were collected from the protruding portion of the colon cancer lesion, and one normal mucosal tissue fragment (approximately 1 cm) was collected from the surrounding normal colon tissue. The tissue fragments were transported to the laboratory in ice-cold wash medium [DMEM / F12 with HEPES and L-glutamine (Nacalai tesque, Kyoto, Japan) containing 100 units / ml penicillin (Nacalai tesque), 0.1 mg / ml streptomycin (Nacalai tesque), and 10% fetal bovine serum (Sigma)]. Tissues were processed within 24 hours after surgery. Colon cancer tissue fragments and normal colon mucosa samples were separately minced with scissors on 60-mm Petri dishes. Each specimen was digested in 2 ml of collagenase solution (washing medium supplemented with 0.2% collagenase type I (Thermo Fisher Scientific, Waltham, MA, USA) and 50 μg / ml gentamicin (Thermo Fisher)) at 37°C for 40–60 min, and the cells were separated by pipetting. CRC and NCE cells were then filtered once through a 100 μm cell strainer (Corning Inc., Corning, NY) and collected by centrifugation. The isolated cells were then suspended in Matrigel (Corning) and placed in the center of each well of a 12-well cell culture plate (TPP, Trasadingen, Switzerland) (30 μl per well).After Matrigel gelation at 37°C, cells were cultured in Advanced DMEM / F12 (Thermo Fisher Scientific) supplemented with penicillin (100 units / ml, Nacalai tesque), streptomycin (0.1 mg / ml, Nacalai tesque), L-glutamine (2 mM), Y27632 (10 μM), SB431542 (1 μM), EGF (50 ng / ml, Peprotech), and basic FGF (100 ng / ml, Peprotech) in 5% CO2 at 37°C. NCE-SC cells were cultured in eL-WRN medium as previously described {Miyoshi et al., 2018}.

[0042] RNA sequencing analysis. RNA sequencing (RNA-seq) analysis of the NCE-SC spheroid line (GSE189212) was performed by Macrogen (Seoul, Korea) using a previously reported method {Yamamoto, Miyoshi et al. 2020, Kitano, Yamamoto et al. 2022}. Briefly, synthesized libraries were sequenced from both ends (2 × 100 bases) using an Illumina HighSeq 2000 sequencer at a depth of approximately 40 million reads per sample. Sequence reads were mapped to the hg19 reference genome using the HISAT2 program {Kim, Langmead et al. 2015, Kim, Paggi et al. 2019}, and transcripts were integrated using the StringTie program {Pertea, Pertea et al. 2015}. Read counts were determined using the featureCounts program {Liao, Smyth et al. 2014}.

[0043] Database analysis of human colorectal cancer datasets. We used RNA-seq and clinical data from TCGA-CRC {TCGA Network, 2012} (https: / / www.cancer.gov / about-nci / organization / ccg / research / structural-genomics / tcga) using FPKM (fragments per kilobase of exon per million reads mapped) based on the Human Protein Atlas {Uhlen et al., 2017} version 21.0 and Ensembl version 103.38 (https: / / www.proteinatlas.org / about / download). We also used RNA-seq and clinical data from TCGA-CRC normalized using RSEM (RNA-Seq by Expectation Maximization) from cBioportal {Cerami et al., 2012} (https: / / www.cbioportal.org / datasets). We collected microarray and clinical data for GSE39582 {Marisa et al. 2013} from the Gene Expression Omnibus (GEO) of the National Center for Biotechnology Information (NCBI) (https: / / www.ncbi.nlm.nih.gov / geo / ).

[0044] Logarithmic transformation of RNA-seq data. In this study, we initially performed all statistical calculations using FPKM data values, and then finally double-checked the results using RSEM data. To calculate FPKM data, raw RNA-seq expression levels were logarithmically scaled (log2) using a previously reported method. Because Log2(0) cannot be defined, FPKM was added to the logarithmic scale before being transformed: Log2(FPKM+0.1) {Reinhold et al., 2019}. However, because some genes in the RSEM output still had negative values ​​even after adding 0.1, we added 1 to the output values ​​and then transformed them into logarithmic scales, converting all values ​​to positive values: Log2(RSEM+1), as previously reported {Jones et al., 2019}. We confirmed that similar results were obtained for GCS in TCGA-CRC colorectal cancer patients using both Log2(FPKM+0.1) and Log2(RSEM+1) (data not shown).

[0045] Microarray Analysis: NCE-SC or CRC-SC were cultured as described above. After aspirating the medium, lysis buffer (TaKaRa Bio, Kusatsu, Japan) was added directly to each well of the spheroid culture. Total RNA from NCE-SC and CRC-SC spheroid lines was prepared using the NucleoSpin RNA II kit (TaKaRa Bio) according to the manufacturer's protocol. Microarray analysis was performed by Macrogen Inc. (Seoul, Republic of Korea). Briefly, RNA labeling and hybridization were performed using the Agilent One-Color Microarray-Based Gene Expression Analysis protocol (Agilent Technology, V 6.5, 2010), which exhibits a high dynamic range. 100 ng of total RNA from each specimen was linearly amplified and labeled with Cy3-dCTP. Labeled cRNA was purified using the RNAeasy Mini Kit (Qiagen). The concentration and specific activity of the labeled cRNA preparation (pmol Cy3 / μg cRNA) were measured using a NanoDrop ND-1000 (NanoDrop, Wilmington, DE, USA). Each labeled cRNA sample (600 ng) was fragmented with 5 μl 10x blocking agent and 1 μl 25x fragmentation buffer, then heated at 60°C for 30 min. Finally, 25 μl of 2x GE hybridization buffer was added to dilute the labeled cRNA. 40 μl of hybridization solution was poured into a gasket slide and loaded onto Agilent SurePrint G3 Human GE 8x 60K, V3 Microarrays (Agilent). The slides were incubated at 65°C for 17 hours in an Agilent hybridization oven and then washed at room temperature using the Agilent One-Color Microarray-Based Gene Expression Analysis protocol (Agilent Technology, V 6.5, 2010).Hybridized array slides were immediately scanned using an Agilent Microarray Scanner D (Agilent Technologies, Inc.). Raw data were extracted using Agilent Feature Extraction software (v11.0.1.1). To create a raw data text file containing expression data for each gene probe on the array slide, raw data for the same gene were automatically compiled using the Agilent feature extraction protocol. gProcessedSignal values ​​were log-transformed and normalized using the quantile method. Statistical significance of expression data was determined using the false discovery rate (q < 0.1) and fold change (FC > 2). Data analysis was performed using R3.0.2 (www.r-project.org), GraphPad Prism software (Dotmatics, San Diego, USA), and Microsoft Excel software (Microsoft, Tokyo, Japan).

[0046] Lead genes and their associated genes: We identified genes with extremely variable mRNA expression (SD > 6.5) in CRC-SC lines. Among these genes, we selected five lead genes encoded on different chromosomes and independently regulated; their expression levels were not correlated with each other (Spearman correlation coefficient |r| < 0.5). We classified CRC-SC lines into six subgroups based on lead gene expression, including five lead gene subgroups and an unclassified subgroup. We then identified 12 lead gene-associated genes whose expression correlated with the levels of specific lead gene mRNA species. We also investigated the survival time of colorectal cancer patients.

[0047] Gene expression signature scores Gene expression signature scores were calculated as previously reported {Jackstadt et al., 2019}. First, the Z-score Zij for the jth specific gene in the ith patient in the dataset was calculated using the following formula: Here, Xij (uppercase letters) indicates the logarithmic value of the mRNA expression level after log2 transformation for each patient's tumor. The GSE39582 data was measured using microarrays and was originally displayed as log2 values, whereas the TCGA-CRC data used in this study was RNA-seq data, so values ​​were displayed without logarithmic transformation. Therefore, for the TCGA-CRC RNA-seq data, a constant was added while log2 transformation was performed using the following formula: Here, xij (lowercase) denotes the mRNA level of the jth gene in the ith patient data (mRNA abundance measured by RNA-seq). The constant α was set to 0.1 for FPKM (data normalized to account for differences in gene length and correct for fragment number) and 1 for RSEM. Adding a constant during log2 transformation is performed to prevent negative values ​​from being obtained after log2 transformation. This routine method ensures all transformed values ​​are positive, and is recommended as a standard procedure by cBioportal (https: / / docs.cbioportal.org / 5.1-data-loading / data-loading / file-formats / z-score-normalization-script). The constant value was adjusted due to differences in normalization methods between FPKM and RSEM. μj and σj denote the mean and standard deviation of the logarithmic expression level of the jth gene in each dataset, respectively. μj and σj were calculated using the following formulas: Here, N represents the number of patients in the dataset. The cutoff score was determined so as to minimize the statistical P value.

[0048] The gene expression signature score was the average Z score calculated using the following formula: Specifically, the lucky 5 gene expression signature score was calculated by adding the Z scores of the jth genes of the five genes, and the lead gene score was calculated by adding the Z scores of the lead gene and its jth gene correlated with expression in the subgroup, and then dividing the result by the number of genes (m) (i.e., 5 for lucky 5, 3-4 for lead gene signature). To examine overall survival (OS), disease-specific survival (DSS), progression-free survival (PFS), and recurrence-free survival (RFS) for these signatures, colorectal cancer patients in the TCGA-CRC and GSE39582 databases were divided into high and low signature score groups. The cutoff score was determined to minimize the statistical P value.

[0049] General Colorectal Cancer Score (GCS) score Individual mean Z scores were multiplied by positive or negative prognostic coefficients and added to obtain the GCS score. where: is a coefficient indicating the polarity of the prognosis (good prognosis is +1, poor prognosis is -1), is the mean Z-score for each signature. That is, the Z-score is divided by the number of genes in the signature group, such as 5 for the lucky 5 subgroup, 3 for the DEFA6, CXCL14, and IGF2 subgroups, and 4 for the BST2 and MAGEA6 subgroups. For quartile analysis, CRC specimens were divided into four equal subgroups according to GCS score and analyzed using the Kaplan-Meier method followed by the log-rank test. The 95% CI and HR for the other quartile groups were calculated using the Mantel-Haenszel method in the GraphPad Prism software package, with the poorest prognosis quartile as the reference.

[0050] Gene enrichment analysis. False discovery rate (FDR; q) and mean (A) values ​​obtained by comparing the NCE-SC spheroid line (GSE189212) and NCE-SC organoid line (GSE125472) were calculated using TCC-GUI {Su et al., 2019}. Differentially expressed genes (DEGs) were filtered for q < 0.1 and A > 1. Enrichment analysis using cell type signatures {Subramanian et al., 2005} was performed in Metascape {Zhou et al., 2019}. Gene set enrichment analysis (GSEA) using hallmark gene sets was previously described {Subramanian et al., 2005}.

[0051] Hotspot Mutation Analysis. Matrigel-enclosed spheroids were suspended in cell recovery solution (Corning) and collected in a 1.5 mL tube. Matrigel was dissolved at 4°C for 30 minutes with rotation and mixing. Spheroids were centrifuged at 4°C for 5 minutes and washed twice with PBS. Genomic DNA from the spheroids was purified using the DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol. Hotspot mutations in 50 or 409 cancer-related genes in CRC-SC lines were detected by Macrogen Inc. (Seoul, Republic of Korea). A list of sequenced genes and mutations was available on the manufacturer's website (https: / / tools.thermofisher.com / content / sfs / brochures / CO25560_Ion_AmpliSeq_Comprehensive_Cancer_Panel_Gene_List_final9062012.pdf). The dataset was analyzed using Integrative Genomics Viewer software (Broad Institute).

[0052] k-means analysis. Among the genes highly expressed in CRC-SC lines compared to NCE-SC lines, the 100 most differentially expressed genes were subjected to gene cluster analysis using Integrated Differential Expression and Pathway analysis v0.951 (iDEP.951) {Ge, Son et al. 2018}.

[0053] Single-cell RNA-seq (scRNA-seq) library preparation, data processing, and quality check. scRNA-seq library preparation was performed by Rhelixa Inc. (Tokyo, Japan) as previously described {Joanito, Wirapati et al. 2022}. Briefly, fresh 70N-NCE-SC were cultured in single suspension and loaded into a Chromium system (10X Genomics) at 5,000 cells per well. Barcoded sequencing libraries were generated using the Chromium Single Cell 3' v3.1 reagent Kit Dual (10X Genomics). The libraries were sequenced on an Illumina NovaSeq 6000 until full saturation was achieved. Immediately after detection, sequence information from the sequence reads was matched to the hg38 reference genome and further processed using CellRanger v6.1.2 (10X Genomics) to generate an expression matrix based on unique molecular identifier (UMI) values ​​(representing the number of single RNA molecules detected) for each gene in each cell (droplet). Droplets with fewer than 2500 detected gene expressions, more than 6000 detected genes, or more than 15% mitochondrial genome were excluded as droplets containing no target genes, multiplets, or low-quality / dead cells, respectively. UMI values ​​were normalized using the LogNormalize function in Seurat 4.0 {Hao, Hao et al. 2021} to result in a total of 10,000 UMIs for all genes in each cell. Normalized values ​​were then log-transformed after adding a pseudocount of 1.

[0054] Dimensionality Reduction Analysis Using Unsupervised Clustering As previously described {Wang, Dang et al. 2021}, we used Seurat 4.0 {Hao, Hao et al. 2021} on the normalized gene-cell matrix to identify highly variable genes for unsupervised cell clustering. To identify highly variable genes, we used the MeanVarPlot method in the Seurat package to calculate the mean-variance relationship between normalized expression counts per cell. A set of 2,000 highly variable genes was selected. An elbow plot was created using the PCelbowPlot function in Seurat, and the number of significant principal components was determined based on this. Different resolution parameters were then tested for unsupervised clustering to determine the optimal number of clusters. In this study, the first 10 principal components and highly variable genes identified by Seurat were used for unsupervised clustering at a resolution of 0.5, resulting in a total of seven cell populations. To visualize these, we further reduced the dimensionality using the UMAP method with the Seurat function RunUMAP, which showed that the 70N-NCE-SC cell population was roughly a single population (Fig. 2B, left). The principal components used to calculate the embedding were the same as those used for clustering. The plot of single cells with respect to the marker gene group (Fig. 2B, right) was colored by the Seurat function FindMarkers.

[0055] Lead gene subtype assignment for 57 CRC-SC strains. The subtypes of the 57 CRC-SC strains were determined by the most highly expressed lead gene, as calculated in Table 2. That is, the highest value among the five lead gene candidates was used as the determinant of the subtype, as shown in the top box in Figure 5A. CRC-SC strains showing a lead gene fold change of less than 70-fold were assigned to the unclassified "Others" category.

[0056] Isolation of the Lucky 5 Genes: We compared mRNA expression data from CRC-SC spheroids to NCE-SC spheroids in the discovery cohort and evaluated the relationship between the lowest-expressing mRNAs in CRC-SC and NCE-SC spheroids in the validation cohort and patient survival. In the discovery cohort, 19 candidate markers for favorable colorectal cancer outcome were identified using the 20 lowest-ranked microarray probes for fold change in CRC-SC compared with NCE-SC transcriptomes. In the validation cohort, the first evaluation criterion was a log-rank P value of 0.05 or less for the comparison of patient survival between high and low expression levels of the 19 candidate genes in the TCGA-CRC-OS (n = 597), TCGA-CRC-DSS (n = 545), and TCGA-CRC-PFS (n = 564) datasets. Eleven genes met the criteria: CLCA4, ZG16, AQP8, CA2, CEACAM7, PIGR, SLC26A3, MUC12, NXPE4, SLC4A4, and PLA2G2A. The second evaluation criterion in the validation cohort was that the log-rank test P value for patient survival between high and low expression levels of the 11 candidate genes in the GSE39582-OS (n = 556) and GSE39582-RFS (n = 564) datasets must be ≤0.05. Five genes met this criterion: ZG16, PIGR, MUC12, SLC4A4, and PLA2G2A. The expression threshold for the candidate genes was determined using maximally selected log-rank statistics (MSRS) {Hothorn, Lausen et al. 2003}.

[0057] Patient-derived spheroid orthotopic transplantation (PDSOX) model. Ten-week-old NSG mice were anesthetized using an isoflurane anesthesia system (NARCOBI-E, Natsume Seisakusho) on a heat mat (KN-475, Natsume Seisakusho). Abdominal hair was shaved with clippers (ER807PP-A, Panasonic). The shaved abdomen was disinfected with 80% ethanol (#699749, Marusan Sangyo) and povidone-iodine (#46026800, Kenei Pharmaceutical). The abdomen was then covered with a small animal drape (Cleyera). An abdominal incision was made slightly left of the center with sterilized scissors (Daiyu Medical Industries). The end of the cecum was extracted from the abdominal cavity using ring forceps (Daiyu Medical Industries) and placed on the drape. At this point, the mouse's body was positioned so that its head was facing forward and its tail was facing backward. The tip of the cecum was held in place with the index finger of the left hand, and the smooth muscle layer of the cecum was incised with a 26-gauge needle (TERUMO) in the right hand, creating a space between the mucosa and the smooth muscle. The space was then further widened with an 18-gauge needle. Under a long-focus microscope (M651-1, Leica Microsystems), a cluster of spheroids (approximately 10,000 CRC-SC cells) was transplanted into the smooth muscle pocket using tweezers (11252-30 Dumont #5, Fine Science Tools). The cecum was carefully returned to the abdominal cavity, and the incised abdominal wall and epidermis were sutured. The PDSOX mice were observed for 26 weeks after transplantation. Moribund PDSOX mice were euthanized for pain relief and considered to have died from cancer. Subsequently, they were analyzed macroscopically and pathologically. Survival time was determined as the average survival time of two PDSOX mice. All animal experiments were conducted in accordance with protocols approved by the Kyoto University Animal Care and Use Committee.

[0058] Statistical analysis was performed using GraphPad Prism software (Dotmatics, San Diego, USA) or Microsoft Excel software (Microsoft, Tokyo, Japan). Data were analyzed using Kaplan-Meier survival curves, the log-rank test, the Mantel-Haenszel test, the Spearman correlation test, the Mann-Whitney U test, the Kruskal-Wallis test, the Welch t test, and the Benjamini-Hochberg method.

[0059] 2. Results Gene Expression Analysis of Normal Colon Epithelial Stem Cells (NCE-SCs) Human NCE-SCs have been cultured using several methods. We established NCE-SCs from over 160 colorectal cancer patients using serum-containing medium. We first measured mRNA expression levels in five NCE-SC spheroid cell lines using microarray hybridization-based mRNA quantification and RNA sequencing (data not shown). The resulting data revealed a unique pattern: the complexity and levels of mRNA expression were essentially identical across different patients, with only gender differences observed for genes located on the sex chromosomes. This contrasts with the colorectal cancer stem cell (CRC-SC) spheroid line population, which showed a broader distribution of mRNA levels.

[0060] On the other hand, analysis using the Metascape program revealed that our NCE-SC spheroid line conspicuously expressed a higher proportion of mRNA species expressed in the fetal digestive system compared with cells from organoid culture {Michels et al., 2019} (Figure 2C). Furthermore, Gene Set Enrichment Analysis (GSEA) results revealed significant differences in several gene sets (data not shown). These results were more similar to NCE-SC organoids cultured in "refined medium" than in the "original medium" described in the literature {Sato, T. et al. 2011}. To be sure, we analyzed the cell type composition of the NCE-SC we established and used using single-cell RNA (scRNA) analysis, confirming that it essentially consisted of a single undifferentiated cell type (Figure 2B).

[0061] High mRNA expression levels in colorectal cancer stem cells correlate with favorable prognosis; the lucky pentad (5) signature. We previously reported approximately 160 stem cell (also known as tumor-initiating cell) lines cultured as spheroids from primary colorectal cancer tumors. From these CRC-SC cell lines and recent additions, we selected 57 lines, relatively enriched for clinically relevant stage III and IV tumors (79%) (epithelial stem cells, not tumor stroma), for further characterization of intrinsic markers of CRC-SC. The male / female ratio was 34:23. They represented a wide variety of driver gene mutations and stages (data not shown). These colorectal cancers harbored mutations in expected driver genes, such as APC, TP53, and RAS, as well as mutations in SMAD4 and FBXW7. The histopathological findings of the primary tumors were mostly well- to moderately- or poorly differentiated adenocarcinomas, including some myxoid types, two cases likely derived from serrated adenomas with BRAF V600E mutations, and one case with microsatellite DNA instability (MSI-H). The low frequency of MSI-H cases is due to the fact that 75% of them were in stage I / II disease, and none of the 216 cases were in stage III / IV disease.

[0062] We hypothesized that a subset of genes expressed in the CRC-SC transcriptome would provide prognostic data even when measured using whole-cancer tissue samples from patients. To test this hypothesis, we developed a novel strategy using CRC-SC as a prognostic biomarker discovery cohort (Fig. 3A). Specifically, we compared mRNA expression data from CRC-SC with those from NCE-SC (Fig. 3A). We then focused on the 20 gene probes ranked at the bottom and top of the resulting differentially expressed gene list, i.e., genes whose expression levels were significantly decreased and increased compared to NCE-SC (Table 2).

[0063] In the first set of analyses, we first examined the genes with the lowest expression levels in CRC-SC as a whole compared with NCE-SC to identify a subgroup of cell lines. Specifically, we found that 19 genes were expressed at low levels in most CRC-SC spheroids compared with NCE-SC, but their expression levels were increased to levels similar to those in NCE-SC in a subset of CRC-SC lines (Fig. 3A and Table 2). From these 19 genes, we first selected 11 as candidate markers for favorable prognosis based on the TCGA-CRC database (n = approximately 600), and further narrowed this down to five genes that also showed favorable prognosis in the GSE39582 database (n = approximately 600). These genes were MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16 (Fig. 2A and Fig. 3B).

[0064] As expected, high expression of these five genes was a favorable prognostic factor for patients in both TCGA-CRC and GSE39582, and each gene showed statistically significant differences in not only OS data but also PFS / RFS / DSS data (Figures 4A and 4B). Therefore, we named these five genes the "lucky pentad (5)."

[0065] To assess the clinical significance of these five gene expressions collectively, we normalized and integrated these standardized expression levels by Z-score transformation. The resulting statistical significance of the prognostic difference between the high and low Lucky 5 signature score expression groups showed low HRs and P values ​​for both TCGA-CRC and GSE39582: TCGA-CRC-OS / DSS / PFS HR 0.48 / 0.48 / 0.6 (P = 0.002 / 0.01 / 0.01), and GSE39582-OS / RFS HR 0.46 / 0.39 (P = 0.003 / 0.0007) (Figure 4C).

[0066] Furthermore, these genes are highly expressed in specific cell types that compose the normal colonic mucosa: MUC12 and PIGR in colonocytes, PLA2G2A and ZG16 in goblet cells, and SLC4A4 in CA1-positive colonocytes.

[0067] Furthermore, among these five genes, MUC12, SLC4A4, and ZG16 were expressed at even higher levels when NCE-SCs were induced to differentiate by Wnt ligand withdrawal in cell culture (Fig. 3A (bottom), leftmost row 3N-d). The expression levels of PIGR and PLA2G2A were comparable between NCE-SCs and their differentiated cells. Therefore, NCE-SCs cultured as spheroids in our medium appear to be partially differentiated toward fetal colon and small intestine, as suggested by the Metascape analysis data (Fig. 2C). On the other hand, in our CRC-SC spheroids, the expression levels of EMT (epithelial-mesenchymal transition) signature genes were comparable to those of NCE-SC stem cell markers (Fig. 3B).

[0068] A novel prognostic signature candidate represented by high, non-intersecting mRNA expression of DEFA6, CXCL14, BST2, MAGEA6, and IGF2 in human colorectal cancer stem cell spheroid cell lines. In a second analysis, we searched for genes that were expressed at higher levels in independent, non-intersecting subsets of CRC-SC spheroids than in NCE-SC. We found five lead gene candidates representing novel molecular subgroups (Table 2 and Figure 5A). These gene candidates are DEFA6, CXCL14, BST2, MAGEA6, and IGF2. These lead gene candidates are only moderately expressed in NCE-SC, and largely do not intersect with each other or with the lucky 5-gene signature expression group (Figure 2A, Figure 5A).

[0069] Although all five subtypes represented by these lead genes were found in left-sided and right-sided colon cancers as well as rectal cancers, primary tumors expressing BST2 were statistically significantly more prevalent in colon cancers (P = 0.0001), and those expressing IGF2 were more prevalent in rectal cancers (P = 0.002).

[0070] To assess the clinical significance of these lead gene candidates in the five molecular subtypes of CRC-SC spheroids, we analyzed the TCGA-CRC and GSE39582 databases to examine their correlation with patient prognosis and survival, similar to the Lucky 5 analysis described above. Specifically, we first performed Kaplan-Meier analysis of the five lead gene candidates to examine the difference in prognosis between high- and low-expressing patients in the public database (data not shown). Next, we selected genes (excluding the lead gene) that were highly expressed within each subgroup compared with NCE-SC from the list of the top 20 genes (Table 2), and then selected genes whose expression correlated with each other and with patient prognosis. Finally, we calculated Z-scores for these genes (including the lead gene) as subgroups to examine whether they formed statistically stronger prognostic signatures compared with the data for the lead gene alone (Figures 5B and 5C). These candidate genes were (i) DEFA6, DEFA5, and SPINK1; (ii) CXCL14, MME, and MSX1; (iii) BST2, BATF, RAMP1, and TPM2; (iv) MAGEA6, MAGEA3, MAGEA12, and CSAG1; and (v) IGF2, NELL2, and RBMS1. In a preliminary study, we used the mRNA expression levels from the TCGA-CRC database without logarithmic transformation to calculate a gene expression signature score by summing the Z scores of the genes in each gene group, and found a similar correlation with patient prognosis (data not shown).

[0071] To statistically examine the method used here, we used the k-means algorithm, a representative non-hierarchical statistical method, to classify the data into five subgroups. The five lead gene candidates mentioned above, namely, DEFA6, CXCL14, BST2, MAGEA6, and IGF2, were found in each independent cluster, consistent with the classification described above (data not shown). We further increased the number of clusters used in the k-means algorithm to 10, but the five lead gene candidates still belonged to independent clusters. The lead genes in the additional clusters were not related to non-coding RNAs or patient survival (data not shown). These results suggest that our method is essentially exhaustive in terms of identifying genes specific to CRC-SC that predict outcome.

[0072] Below we explain the five molecular subtypes represented by individual lead genes.

[0073] For the lead gene signatures (LGS) DEFA6, BST2, and MAGEA6, the differences in prognosis (OS / DSS / PFS / RFS) between patients with high and low expression of TCGA-CRC and GSE39582 genes showed statistically significant differences with small P values ​​(<0.05). For the CXCL14 signature, patients with high expression in TCGA-CRC-OS tumors showed a favorable prognosis, but PFS was reversed to a poor prognosis. Similarly, GSE39582-OS and RFS also showed a reversal of favorable and poor prognosis. Because these statistical results do not meet the criteria for a prognostic predictor, we have decided to exclude CXCL14 and its signature from our list of prognostic biomarkers. The questionable prognostic value of CXCL14 has been discussed in a recent review {Westrich 2020}. IGF2 and its signature significantly correlated with poor prognosis in both TCGA-CRC and GSE39582. Although the statistical power was weaker in GSE39582-OS (P = 0.1), the trend toward poor prognosis was consistent, and IGF2 was retained as the lead gene signature. Similarly, the MAGEA6 signature showed a P = 0.07 in the GSE39582-OS dataset, but was consistent across all other data sets, and was therefore retained as the lead gene signature. In summary, the DEFA6, BST2, MAGEA6, and IGF2 signatures were ultimately adopted as the lead gene signature for prognosis prediction, while CXCL14 and its signature were excluded.

[0074] Colon Cancer Comprehensive Signature (GCS): A Novel Prognostic Parameter Integrating Lucky 5, DEFA6, BST2, MAGEA6, and IGF2 Signatures. As described above, we identified five molecular subgroups and their signatures that correlate with patient prognosis. Two of these, Lucky 5 and DEFA6, are associated with a favorable prognosis, while the remaining three, BST2, MAGEA6, and IGF2, are associated with a poor prognosis. Our established CRC stem cell spheroid lines belonging to these five molecular subgroups included some typical lines, but also borderline lines and lines that did not fit into any subgroup (Figure 5A). Furthermore, some CRC-SC spheroid lines expressing Lucky 5 also expressed high levels of BST2 and IGF2, two of the four lead genes mentioned above, which are associated with poor prognosis (Figure 3A (bottom), see the subgroups in the top row). However, predicting patient prognosis is always important in any clinical setting.

[0075] To this end, we devised a novel mRNA expression signature that integrated the five signatures described above according to the scheme shown in Figure 6A. This signature was derived by summing the individual normalized signatures, assigning a positive polarity to favorable prognosis and a negative polarity to unfavorable prognosis, and is called the General Colorectal Cancer Signature (GCS). The colorectal cancer stem cell spheroid lines we analyzed for total mRNA as our discovery cohort were a set of expanded stage III and IV cancers, taking into account clinical significance. However, the number of lines was limited to 57. As a validation cohort, we first analyzed the correlation with overall survival (OS) using the TCGA-CRC cancer tissue mRNA database, which contains nearly 600 cases.

[0076] Statistical analysis of patients by quartile method revealed a stratification of patients with progressive decline in TCGA-CRC-DSS score (Q4-Q1, Ptrend = 0.001, Fig. 6B), with a significant difference between the highest quartile (Q4) and the lowest quartile (Q1) (P = 0.004, HR, 0.39; 95% CI, 0.20-0.74; Fig. 6B).

[0077] These results suggest that the GCS can be a very useful prognostic signature: by determining which quartile an individual patient falls in, a unified prognosis can be predicted, and relatively mild chemotherapy may be sufficient for patients with high GCS values, while more intensive chemotherapy may be recommended for patients with low GCS values, if possible.

[0078] We also analyzed the OS / RFS data for the mRNA data from GSE39582, which has a similar number of colorectal cancer patients to the TCGA-CRC-DSS and TCGA-CRC-PFS data. Although the statistical power of the P values ​​was weaker due to the smaller number of cases compared to the TCGA-CRC-OS and TCGA-CRC-PFS data, we obtained trends similar to those of the TCGA-CRC-PFS data (Figure 6C).

[0079] To intuitively visualize the GCS features of each patient's tumor, we used a radar chart (Figure 6D) as an example. Tumors from patients in TCGA-CRC Q4 (highest ranking) showed high signature scores for Lucky 5 and DEFA6, while tumors from patients in Q1 (lowest ranking) showed significantly higher values ​​for the MAGEA6 signature. Interestingly, patients in the same GSE39582 group had characteristics quite similar to those of TCGA-CRC (Figure 6D).

[0080] We examined the distribution of GCS scores for the discovery cohort (n = 57) and the TCGA-CRC and GSE39582 datasets. Unsurprisingly, the mean GCS score (μ) was close to 0 in all three datasets. Meanwhile, the median GCS score was positive (0.24 and 0.20) in the two databases, while negative (−0.45) in the discovery cohort's CRC stem cells (Figure 6E, top). This is likely due to the intentionally high proportion of stage III and IV cases in the discovery cohort. Meanwhile, the GCS score distributions for the larger sample sizes of TCGA-CRC (n = 597) and GSE39582 (n = 566) were close to normal, demonstrating the ease of interpretation of the GCS and its applicability to larger patient populations in general clinical practice (Figure 6E, bottom).

[0081] We developed a novel xenograft model system, called patient-derived spheroid orthotopic xenografts (PDSOX), as an animal model for predicting the prognosis and outcome of postoperative colorectal cancer patients. Briefly, approximately 10,000 limited CRC-SC cells were implanted under the cecal epithelium of immunodeficient mice. CRC-SC lines from the above-mentioned GCS score quartiles Q1 to Q4 were then tested. These xenografts formed primary tumors in the cecal mucosa. Four of the 14 CRC-SC lines derived from the poor-prognosis Q1 quartile developed multiple metastatic lesions in the liver and lungs over the 26-week observation period (Figure 6F, right). On the other hand, mice implanted with the good-prognosis Q4 quartile CRC-SC line survived throughout the observation period (Figure 6F, upper left), and consistent with this, no metastatic lesions were detected in the liver or lungs at autopsy. Two CRC-SC lines from each of the Q2 and Q3 quartiles died within 26 weeks after implantation (Figure 6F, lower left). Furthermore, many of the CRC-SC lines in Q4 had high levels of the Lucky 5 signature, a signature associated with a favorable prognosis (Fig. 6G). Although the possibility remains that the tumor microenvironment may have an additional influence on these results, these results strongly suggest that the GCS score may play an important role in the outcome of colorectal cancer patients, reflecting the unique growth and metastatic characteristics of individual CRC-SCs.

[0082] Four typical lead gene subtypes account for approximately 40% of the patient population. Up to this point, we have focused on developing diagnostic tools applicable to all individual patients. However, when we limited our analysis to patients with high lead gene signature (LGS) scores, it became clear that highly typical molecular subtypes exist. To this end, we analyzed the same two public databases using a different statistical analysis method to identify such unique molecular subtypes. Specifically, as shown in Figure 7A, we artificially assigned each individual patient to the LGS subgroup with the highest score out of four LGS scores. This resulted in a strict classification of all patients into four distinct subgroups (this method has been described in a recent paper). Each of these four subgroups was then divided into high- and low-expressing subgroups using the MSRS method. Only the high-expressing subgroup was defined as the typical subtype, while the low-expressing subgroup was combined with patients from the low-expressing subgroups of the other subgroups in the database (Figure 7B). The four typical subtypes thus constructed accounted for approximately 40% of TCGA-CRC and GSE39582 patients, respectively, with the remaining 60% being relatively poorly characterized subgroups (Figure 7B).

[0083] Based on these results, we examined progression-free survival (PFS) for typical subtypes in the TCGA-CRC data. We found a stratification of DEFA6, BST2, MAGEA6, and IGF2 subtypes, which showed progressively decreased survival (log-rank test P = 0.02). As expected, the DEFA6 subtype tended to have a longer PFS (HR, 0.60; 95% CI, 0.35-1.0), while the BST2, MAGEA6, and IGF2 subtypes tended to have a shorter PFS (Figure 7C, left). These results suggest that these subtypes may form distinctive signatures. That is, patients in the DEFA6 subtype may be able to receive relatively mild chemotherapy, while patients in the BST2, MAGEA6, or IGF2 subtypes may be recommended more intensive chemotherapy than usual. Similarly, when we analyzed the RFS of the mRNA data of GSE39582, we found a trend very similar to that of the TCGA-CRC-PFS data (logrank test P = 0.003) (Fig. 7C, right).

[0084] Comparison of GCS with existing colon cancer prognosis prediction methods Prognosis prediction by GCS was compared with an existing colon cancer prognosis prediction method, ColoGuidePro (Inven2, Oslo, Norway) {Sveen et al. 2012}. We combined the RFS data from GSE14333 {Jorissen et al. 2009}, GSE17536 {Smith et al. 2009}, GSE17537 {Smith et al. 2009}, GSE33113 {de Sousa E Melo et al. 2011}, GSE37892 {Laibe et al. 2012}, GSE38832 {Tripathi et al. 2014}, and GSE39582 to show the RFS of patients with Q1–Q4 (left), the highest and lowest scores (Q1 and Q4) (center), or a favorable (80%) and poor (20%) prognosis (right) in a cohort of patients with colorectal cancer of all stages (Figure 8A), stage II (Figure 8B), or stage III (Figure 8C). GCS showed a much stronger statistical significance than ColoGuidePro.

[0085] These results indicate that each gene signature and the GCS can be very useful signatures for predicting the prognosis of colorectal cancer patients. In particular, the GCS, which reflects the integrated analysis of each gene signature, can be an extremely practical prognostic prediction tool. As a result, the GCS score is expected to facilitate understanding of the disease condition for colorectal cancer patients and their treating physicians, and to help determine the most appropriate treatment plan.

[0086]

[0087] References Cerami, E., et al. (2012). "The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data." Cancer Discovery 2(5): 401-404. Ge, SX., et al. (2018). " iDEP: an integrated web application for differential expression and pathway analysis of RNA-Seq data." BMC Bioinformatics 19(534): 1-24. Hao., Y., et al. (2021). "Integrated analysis of multimodal single-cell data." Cell 184:3573-3587. Hothorn., Y., et al. (2003). "On the exact distribution of maximally selected rank statistics." Comput. Stat. Data Anal. 43:121-137. Jackstadt., I., et al. (2019). " Epithelial Notch signaling rewires the tumor microenvironment of colorectal cancer to drive poor-prognosis subtypes and metastasis." Cancer Cell 36: 319-336. Joanito., et al. (2022). " Single-cell and bulk transcriptome sequencing identifies two epithelial tumor cell states and refines the consensus molecular classification of colorectal cancer." Nat. Genet. 54(7): 963-975. Jones, W., et al. (2019)."Deleterious effects of formalin-fixation and delays to fixation on RNA and miRNA-seq profiles." Sci. Reports 9: 6980. Jorissen, R.N. et al. (2009) "Metastasis-associated gene expression changes predict poor outcomes in patients with Dukes stage B and C colorectal cancer." Clin Cancer Res 15, 7642-7651. Kim, D., et al. (2015). "HISAT: A fast spliced aligner with low memory requirements." Nat. Methods 12: 357-360. Kim, D., et al. (2019). "Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype." Nat Biotechnol 37: 907-915. Kitano, S., et al. (2022). "A novel parameter for cancer chemosensitivity to FGFR inhibitors." Cancer Sci. 113(11): 4005-4010 Laibe, S. et al. (2012). "A seven-gene signature aggregates a subgroup of stage II colon cancers with stage III." OMICS 16, 560-565. Liao, Y., et al. (2014). "featureCounts: an efficient general purpose program for assigning sequence reads to genomic features." Bioinformatics 30(7): 923-930. Loboda, A., et al. (2011)."EMT is the dominant program in human colon cancer." BMC Medical Genomics 4(9): 1-10. Marisa, L., et al. (2013). "Gene expression classification of colon cancer into molecular subtypes: characterization, validation and prognostic value." PLoS Med 10: e1001453. Miyoshi, H., et al. (2018). "An improved method for culturing patient-derived colorectal cancer spheroids." Oncotarget 9(31): 21950-21964. The Cancer Genome Atlas Research Network (2012). "Comprehensive molecular characterization of human colon and rectal cancer." Nature 489(7407): 330-337. Pertea, M., et al. (2015). "StringTie enables improved reconstruction of a transcriptome from RNA-seq reads." Nat. Biotechnol. 33: 290-295. Reinhold, W. C., et al. (2019). "RNA sequencing of the NCI-60: integration into CellMiner and CellMiner CDB." Cancer Res. 79(13): 3514-3524. Smith, J.J. et al. (2009). "Experimentally derived metastasis gene expression profile predicts recurrence and eath in patients with colon cancer." Gastroenterology 138, 958-968. de Sousa E Melo, F. et al. (2011). "Methylation of cancer-stem-cell-associated Wnt target genes predicts poor prognosis in colorectal cancer patients." Cell Stem Cell 9, 476-485. Su, W., et al. (2019). "TCC-GUI: a Shiny-based application for differential expression anaslysis of RNA-Seq count data." BMC Research Notes 12(133). Subramanian, A., et al. (2005). "Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles." PNAS 102: 15545-15550. Sveen, A. et al. (2012). "ColoGuidePro: A prognostic 7-gene expression siganture for stage III colorectal cancer patients." Clin. Cancer Res. 11. Tripathi, M.K. et al. (2014). "Nuclear factor of activated T-cell activity is associated with metastatic capacity in colon cancer." Cancer Res 74, 6947-6957. Uhlen, M., et al. (2017). "A pathology atlas of the human cancer transcriptome." Science 357(6352): eaan2507. Yamamoto, T., et al. (2020)."Chemosensitivity of Patient-Derived Cancer Stem Cells Identifies Colorectal Cancer Patients with Potential Benefit from FGFR Inhibitor Therapy." Cancers 12(8): 2010. Wang, R., et al. (2021). " Single-cell dissection of intratumoral heterogeneity and lineage diversity in metastatic gastric adenocarcinoma." Nat. Med. 27(1): 141-151. Zhou, Y., et al. (2019). "Metascape provides a biologist-oriented resource for the analysis of systemes-level datasets." Nat. Comm. 10(1523).

[0088]

Claims

1. A method for predicting the prognosis of a colorectal cancer patient, comprising measuring the expression levels of one or more genes selected from the group of genes (a) to (e) in a cancer sample from the patient: (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16; (b) DEFA6, DEFA5, and SPINK1; (c) BST2, BATF, RAMP1, and TPM2; (d) MAGEA6, MAGEA3, MAGEA12 and CSAG1; and (e) IGF2, NELL2, and RBMS1.

2. The Z-score (Z) of each gene is calculated using the formula (I):

10. The method of claim 1, further comprising: determining a gene expression signature score for each gene group by: (where x is the expression level of the gene in the patient, and μ and σ are the mean and standard deviation of the expression levels of the gene in a plurality of colorectal cancer patients, respectively); determining a gene expression signature score for each gene group, wherein the gene expression signature score for each gene group is the sum of the Z-scores of the genes in each gene group or the mean of the Z-scores of the genes in each gene group; and comparing the gene expression signature score to a cutoff value.

3. The method of claim 2, wherein the gene expression signature score for each gene group is the average of the Z scores of the genes in each gene group.

4. The method of claim 2, further comprising comparing the gene expression signature score of the gene group (a) with a cutoff value, wherein if the gene expression signature score is higher than the cutoff value, the patient is predicted to have a good prognosis.

5. The method of claim 2, further comprising comparing the gene expression signature score of gene group (b) with a cutoff value, wherein if the gene expression signature score is higher than the cutoff value, the patient is predicted to have a good prognosis.

6. The method of claim 2, further comprising comparing the gene expression signature score of the gene group (c) with a cutoff value, wherein if the gene expression signature score is higher than the cutoff value, the patient is predicted to have a poor prognosis.

7. The method of claim 2, further comprising comparing the gene expression signature score of the gene group (d) with a cutoff value, wherein if the gene expression signature score is higher than the cutoff value, the patient is predicted to have a poor prognosis.

8. The method of claim 2, comprising comparing the gene expression signature score of the gene group (e) with a cutoff value, wherein if the gene expression signature score is higher than the cutoff value, the patient is predicted to have a poor prognosis.

9. The method of claim 1, wherein the one or more gene groups are gene groups (a) to (e).

10. The one or more gene groups are gene groups (a) to (e), and the colorectal cancer comprehensive signature (GCS) score is calculated using formula (II): (In the formula, is a coefficient indicating the polarity of prognosis, where the coefficient of gene group (a) to (b) is 1, the coefficient of gene group (c) to (e) is -1, 10. The method of claim 1, further comprising: determining the GCS score by: (where, Z-score = 1, where Z-score = 1, is the average of the Z-scores of the genes in each gene group); and comparing the GCS score to a cutoff value.

11. The method of any one of claims 1 to 10, wherein the cancer sample is a surgical sample or a biopsy sample.

12. The method according to any one of claims 1 to 10, wherein the cancer sample is a cancer spheroid.

13. A kit for predicting the prognosis of a colorectal cancer patient, comprising reagents for measuring the expression levels of genes in one or more gene groups selected from the group consisting of (a) to (e): (a) MUC12, PIGR, PLA2G2A, SLC4A4, and ZG16, (b) DEFA6, DEFA5, and SPINK1, (c) BST2, BATF, RAMP1, and TPM2, (d) MAGEA6, MAGEA3, MAGEA12 and CSAG1, and (e) IGF2, NELL2, and RBMS1.

14. The kit according to claim 13, wherein the one or more gene groups are gene groups (a) to (e).

Images