Gastric cancer patient classification and methods of treatment

Abstract

The present disclosure provides methods for classifying genomic subtypes of gastric tumor using digital gene expression profiles. The methods include selecting a therapeutic modality for treating patients based on their genomic subtype of gastric tumor. Also provided herein are kits including primers or probes directed to one or more genes associated with a gastric tumor subtype.

Description

GASTRIC CANCER PATIENT CLASSIFICATION AND METHODS OF TREATMENTCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of priority under 35 U.S.C. § 119(e) of U.S. Provisional Application Nos. 63 / 531,520, filed August 8, 2023 and 63 / 535,022, filed August 28, 2023. The disclosures of the prior applications are considered part of and are herein incorporated by reference in the disclosure of this application in their entirety.FIELD OF THE DISCLOSURE

[0002] The present disclosure generally relates to classification and treatment of cancer using genomic analysis, and more specifically, to classifying gastric tumors in patients based on the gene expression profile and identifying therapeutic strategies based on the sub class of the tumor.BACKGROUND INFORMATION

[0003] Gastric adenocarcinoma (GAC) is the fifth most commonly diagnosed cancer and the third leading cause of cancer death in the world, following only lung and colorectal cancers in overall mortality. Despite recent advances in treatment including immunotherapy, the prognosis for patients with advanced GAC remains poor. The development of GAC is a multistep process driven by genomic and epigenomic alterations that lead to the transformation of epithelial cells in the stomach and then the growth and invasion of cancer cells. GAC is a highly heterogeneous disease, as reflected by its wide range of clinical outcomes in terms of overall survival, early recurrence, and resistance to treatment. Therefore, a robust classification of GACs into clinically and molecularly homogeneous subgroups could significantly improve treatment selection. Many investigators have devoted considerable effort in establishing such classifications by various approaches. However, there are still considerable gaps in translating molecular subtypes into clinical practice.

[0004] The clinical relevance of genomic and molecular subtypes of GAC has been demonstrated by several studies showing their significant association with overall survival, early recurrence, and treatment response. However, their translation to the clinic has been hindered by discrepancies in classification methods that might be attributable to ethnic differences of patients, platforms used for collecting genomic data, sample collection and processing methods, and the prediction algorithms used. Despite these discrepancies, there is substantial overlap among the subtypes identified previously. Therefore, finding the consensus GAC subtypes would be highly useful for uncovering shared oncogenic / immunogenicmolecular alterations that would lead to better selection of therapeutic targets and identification of robust biomarkers for predicting prognosis or treatment response.SUMMARY OF THE DISCLOSURE

[0005] The present disclosure is based on the key discovery that gastric tumors can be classified based on their digital gene expression profiles and that such classification provides information on therapeutic modalities that are suitable to administer to the patients in that subtype.

[0006] In one embodiment, the present disclosure provides a method of classifying a genomic subtype of a gastric tumor. In one aspect, the method includes obtaining a digital gene expression profile of the gastric tumor. In one aspect, the method includes analyzing the digital gene expression profile using a first computer system for classifying the digital gene expression profile, wherein the first computer system classifies the genomic subtype of the gastric tumor as a CGS1 subtype tumor, a CGS2 subtype tumor, a CGS3 subtype tumor, a CGS4 subtype tumor, a CGS5 subtype tumor, or a CGS6 subtype tumor. In one aspect, the digital gene expression profile of the gastric tumor is compared to a reference digital gene expression profile of the genomic subtype of the gastric tumor or a reference database of digital gene expression profiles. In one aspect, the digital gene expression profile is obtained by normalizing gene expression data to one or more housekeeping genes using a second computer system. In one aspect, the CGS1 subtype tumor includes expression of DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and / or H19. In one aspect, the CGS2 subtype tumor includes expression of AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and / or SIOOP. In one aspect, the CGS3 subtype tumor includes expression of OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 1 OB, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and / or CXCL9. In one aspect, the CGS4 subtype tumor includes expression of PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6 and / or SERPINH1. In one aspect, the CGS5 subtype tumor includes expression ofZIC2, CXCL1, SERPINB5, METRN, HOXA9, CRYBG1, MYB, GSKIP, SPP1, NOXO1, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and / or DOCK5. In one aspect, the CGS6 subtype tumor includes expression of GBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D,PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and / or CD247. In one aspect, the method further includes analyzing one or more of a digital proteomic expression profile, a copy number profile, a mutation profile, and / or methylation profile for the genomic subtype of the gastric tumor. In one aspect, the CGS1 subtype tumor includes a poor prognosis, a high stem cell characteristic, a high IGF1 expression, and / or a low level of genomic alteration. In one aspect, the CGS2 subtype tumor includes canonical epithelial gene expression. In one aspect, the CGS3 subtype tumor includes a high copy number alteration, a low immune reactivity, and / or HER2 activation. In one aspect, the CGS4 subtype tumor includes a high copy number alteration, a low immune reactivity, SALL4 expression and / or KRAS activation. In one aspect, the CGS5 subtype tumor includes a high mutation burden, moderately high immune reactivity and / or micros atellite instability. In one aspect, the CGS6 subtype tumor includes a high level of methylation, a high immune reactivity and / or is positive for Epstein Barr virus. In one aspect, the CGS1 subtype tumor includes a stem score of about 0.8. In one aspect, the CGS2 subtype tumor includes a stem score of about 0.4 to 0.5. In one aspect, the CGS3 subtype tumor includes a stem score of about 0.2 to 0.3. In one aspect, the CGS4 subtype tumor includes a stem score of about 0.4. In one aspect, the CGS5 subtype tumor includes a stem score of about 0.1 to 0.2. In one aspect, the CGS6 subtype tumor includes a stem score of about 0.2 to 0.4.

[0007] In one aspect, the CGS1 subtype tumor includes a CD8TRS score of about 0.4 to 0.6. In one aspect, the CGS2 subtype tumor includes a CD8TRS score of about 0.2. In one aspect, the CGS3 subtype tumor includes a CD8TRS score of about 0.2. In one aspect, the CGS4 subtype tumor includes a CD8TRS score of about 0.3. In one aspect, the CGS5 subtype tumor includes a CD8TRS score of about 0.3 to 0.4. In one aspect, the CGS6 subtype tumor includes a CD8TRS score of about 0.6.

[0008] In one aspect, the CGS1 subtype tumor includes an anti-PDLl response score of about 0.4 to 0.5. In one aspect, the CGS2 subtype tumor includes an anti-PDLl response score of about 0.3. In one aspect, the CGS3 subtype tumor includes an anti-PDLl response score of about 0.3. In one aspect, the CGS4 subtype tumor includes an anti-PDLl response score of about 0.4. In one aspect, the CGS5 subtype tumor includes an anti-PDLl response score of about 0.3 to 0.4. In one aspect, the CGS6 subtype tumor includes an anti-PDLl response score of about 0.3 to 0.4.

[0009] In one embodiment, the present disclosure provides a method of selecting a therapeutic modality for treating a patient with a gastric tumor. In one aspect, the method includes classifying the genomic subtype of the gastric tumor according to the methodsdescribed herein. In one aspect, the method includes selecting a therapeutic modality based on the genomic subtype of the gastric tumor. In one aspect, the method includes treating the patient with the CGS3 subtype tumor with a radiation therapy, a ferroptosis inducer, a HER2 targeting therapy or a combination thereof. In one aspect, the HER2 targeting therapy is neratinib, trastuzumab, fam-trastuzumab deruxtecan, hyaluronidase, dacomitinib, lapatinib, tucatinib, pertuzumab, margetuximab, ado-trastuzumab emtansine or a combination thereof. In one aspect, the ferroptosis inducer is erastin, RSL5, DIDS, TMZ, MMRi62, Sorafenib, RSL3, ML162, DPI7, DPI10, FIN56, BSO, MEII, PE, AE, SAS, NDP4928, FIN56, Statin, Brequinar, Dexamethasone or a combination thereof. In one aspect, the method includes treating the patient with the CGS1 subtype tumor, the CGS4 subtype tumor, the CGS5 subtype tumor, or the CGS6 subtype tumor with an immunotherapy. In one aspect, the immunotherapy is a monoclonal antibody, a CAR T cell therapy, a checkpoint inhibitor, a cancer vaccine, a cytokine, an immune system modulator or a combination thereof. In one aspect, the checkpoint inhibitor is atezolizumab, ipilimumab, pembrolizumab, nivolumab or a combination thereof. In one aspect, the cytokine is interleukin-2 (IL-2), interferons-alpha (IFN-alpha), interleukin- 12 (IL-12), interleukin- 15 (IL-15), or a combination thereof. In one aspect, the method includes treating the patient with the CGS1 subtype tumor with an IGF1R targeting therapy. In one aspect, the IGF 1R targeting therapy is ganitumab, xentuzumab, teprotumumab, teprotumumab, Linsitinib (OSI-906), BMS-754807, Picropodophyllin, NT157, an IGF-1R antisense oligonucleotide, an IGF1R small interfering RNA (siRNA) or a combination thereof. In one aspect, the method includes treating the patient with the CGS4 subtype tumor with a SALL4 or KRAS targeting therapy. In one aspect, the KRAS targeting therapy is AMG510 (sotorasib), MRTX849 (adagrasib), ARS-3248, LY3499446, D-1553, JDQ443, GDC-6036 or a combination thereof. In one aspect, the SALL4 targeting therapy is a SALL4 antisense oligonucleotide, a SALL4 small interfering RNA (siRNA), thalidomide, or a combination thereof. In one aspect, the methods include treating the patient with the CGS6 subtype tumor with an antiviral agent or a PI3K inhibitor. In one aspect, the antiviral agent is acyclovir, desciclovir, ganciclovir, adenine arabinoside, phosphonoacetic acid or a combination thereof. In one aspect, the PI3K inhibitor is Copanlisib, Idelalisib, Umbralisib, Duvelisib, Alpelisib or a combination thereof.

[0010] In one embodiment, provided herein is a kit for classifying a genomic subtype of a gastric tumor that includes a probe or a primer for one or more genes associated with a gastric tumor subtype. In one aspect, the one or more genes are DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1,F0XF1, VIM, PBX3, CYR61, CTGF, and H19 associated with a CGS1 gastric tumor subtype. In one aspect, the one or more genes are AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and / or SIOOP associated with a CGS2 subtype tumor. In one aspect, the one or more genes are, OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 1 OB, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and CXCL9 associated with a CGS3 subtype tumor. In one aspect, the one or more genes are PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and SERPINH1 associated with a CGS4 subtype tumor. In one aspect, the one or more genes are ZIC2, CXCL1, SERPINB5, METRN, H0XA9, CRYBG1, MYB, GSKIP, SPP1, N0X01, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and DOCK5 associated with a CGS5 subtype tumor. In one aspect, the one or more genes are GBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and CD247 associated with a CGS6 subtype tumor.

[0011] In one embodiment, provided herein is a kit for determining a therapeutic modality for treating a patient with a gastric tumor that includes a probe or a primer for one or more genes associated with a gastric tumor subtype. In one aspect, the one or more genes are DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and H19 associated with a CGS1 gastric tumor subtype. In one aspect, the one or more genes are AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and SIOOP associated with a CGS2 subtype tumor. In one aspect, the one or more genes are, OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP10B, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and CXCL9 associated with a CGS3 subtype tumor. In one aspect, the one or more genes are PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and SERPINH1 associated with a CGS4 subtype tumor. In one aspect, the one or more genes are ZIC2, CXCL1, SERPINB5, METRN, HOXA9, CRYBG1, MYB, GSKIP, SPP1, NOXO1, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and DOCK5 associated with a CGS5 subtype tumor. In one aspect, the one or more genes are GBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1,CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and CD247 associated with a CGS6 subtype tumor.BRIEF DESCRIPTION OF THE DRAWINGS

[0012] FIGS. 1A-1G. FIG. 1A is a table summarizing the genomic data used in the study. FIG. IB is a table summarizing the genomic subtypes of gastric cancer used for COCA analysis. FIG. 1C is a picture of a hierarchical dendrogram of supercluster subtypes identified by cluster-of-cluster assignment (COCA). GAC tumors were grouped according to shared features of independently discovered molecular subtypes. Tumors from 6 GAC cohorts (Korea, Samsung, TCGA, Shanghai, Shanxi, and Singaporel, n = 1427) were stratified by 8 genomic classification methods, and subtype information was used for clustering with ConsensusClusterPlus (v.3.12). Columns and rows represent GAC tumors in the discovery / training set. FIG. ID is a graph of delta area plot. The plot shows the relative change in area under the cumulative distribution function (CDF) curve comparing k and k - 1. FIG. IE is a graph of a tracking plot. The plot shows the cluster assignment of tumors (columns) for each k (rows). This plot provides a view of item cluster membership across different values of k. FIG. IF, left panel is a picture of a circular ideogram of pairwise comparisons of 2 classification systems comparing CGS vs. MP-EP subtype, the ribbons indicate matched samples between 2 classification systems. Consensus subtypes are shown on the left, and matched subtypes from previous studies are displayed on the right. Graphics were built using Circos (http: / / circos.ca / ). MP, mesenchymal phenotype; EP, epithelial phenotype; GDIF, genomic diffuse; GINT, genomic intestinal; TCGA, The Cancer Genome Atlas; GS, genome stable; MSI, microsatellite instable; CIN, chromosome instable; EBV, Epstein Barr virus. FIG. IF middle panel is a picture of a circular ideogram of pairwise comparisons of 2 classification systems comparing CGS vs Intrinsic subtype, the ribbons indicate matched samples between 2 classification systems. Consensus subtypes are shown on the left, and matched subtypes from previous studies are displayed on the right. Graphics were built using Circos (http: / / circos.ca / ). MP, mesenchymal phenotype; EP, epithelial phenotype; GDIF, genomic diffuse; GINT, genomic intestinal; TCGA, The Cancer Genome Atlas; GS, genome stable; MSI, microsatellite instable; CIN, chromosome instable; EBV, Epstein Barr virus. FIG. IF right panel is a picture of a circular ideogram of pairwise comparisons of 2 classification systems comparing CGS vs. TCGA subtype, the ribbons indicate matched samples between 2 classification systems. Consensus subtypes are shown on the left, and matched subtypes from previous studies are displayed on the right. Graphics were built using Circos (http: / / circos.ca / ). MP, mesenchymalphenotype; EP, epithelial phenotype; GDIF, genomic diffuse; GINT, genomic intestinal; TCGA, The Cancer Genome Atlas; GS, genome stable; MSI, microsatellite instable; CIN, chromosome instable; EBV, Epstein Barr virus. FIG. 1G is a graph showing the clinical significance of 6 consensus subtypes of GAC in discovery / training cohorts (n = 1185). Patients in the Korea, TCGA, Shanxi, and Samsung cohorts were included in the analysis; clinical data from the Shanghai and Singaporel cohorts were not available. OS, overall survival. P-value indicates the significance from log-rank test.

[0013] FIGS. 2A-2D. FIG. 2A is a picture showing the expression patterns of the prediction signatures for 6 consensus subtypes of GAC in training and validation cohorts. FIG. 2B is a picture showing the 2-sample t-tests were conducted to identify subtype-specific genes in gene expression data from this cohort. For selection of subtype CGS1 genes, 5 2-sample t-tests (CGS1 vs. CGS2, CGS1 vs, CGS3, CGS1 vs. CGS4, CGS1 vs. CGS5, and CGS1 vs. CGS6 comparisons) were carried out. Of the genes whose expression significantly differed in 4 of the 5 t-tests (P < 0.01), the top 20 genes for each subtype were selected. Expression patterns of 120 selected genes from TCGA data are presented. FIG. 2C is a diagram demonstrating how to use the GPICS120 web-based predictor. Expression of 120 genes from gastric cancer tumors can be directly loaded to the Shiny webserver (https: / / kasahal.shinyapps.io / gpicsl20 / ). The webbased GPICS120 predictor provides outcomes of predictions, a heatmap of the query data set, and circular faction plots. Gene expression data from the validation set are provided as an example / practice data set. FIG. 2D is a graph showing prognostic significance of consensus subtypes in validation cohorts (n = 1100). OS, overall survival; RFS, recurrence-free survival.

[0014] FIGS. 3A-3J. FIG. 3A left panel is graph of violin and box plots of the number of nonsynonymous mutations in the consensus subtypes (n = 338). CGS5 showed a significantly higher mutation burden than other subtypes (P < 0.001 in all comparisons by Student t-test). Within each violin, the horizontal distance between the left and right curved boundaries represents the distribution of mutations in each subtype. In the box plots, the boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10th and 90thpercentiles. FIG. 3 A right panel is a graph of box and violin plots of Fraction of the genome altered by copy number gain and loss, estimated by gistic2 analysis for each tumor (n = 382). CGS3 and CGS4 had significantly higher proportions of copy number alterations than the other subtypes (P < 0.001 in all comparisons by Student t-test). FIG. 3B is a picture showing a heat map of copy number alterations of the 22 autosomes (n = 382). FIG. 3C is a picture showing the profile of somatic mutations in consensus subtypes in TCGA cohort. The top panel shows individual tumormutation rates; the bottom panel shows genes with statistically significant levels of mutation (MutSig suite, false discovery rate, 0.1 and >7% mutation rate). Mutation types are indicated in the legend at the top. FIG. 3D is a graph showing somatic mutations associated with consensus subtypes in TCGA cohort. Mutation rates of each gene are presented as fractions within subtypes. FIG. 3E is a graph of mutation rates of ERBB3 and ERBB4 are presented as percentages within subtypes. Mutation rates of these genes were significantly higher in CGS5 than in other subtypes (P < 0.01 by %2 test). The combined mutation rate was 45% in CGS5. FIG. 3F is a picture showing distribution of ERBB3 and ERBB4 in consensus subtypes. FIG. 3G is a picture showing Methylation patterns specific to the 6 consensus subtypes of gastric cancer in the TCGA cohort. Multiple 2-sample t-tests were used to identify subtype-specific methylation patterns in the TCGA cohort (n = 377). For the selection of subtype CGS1, 5 2- sample t-tests (CGS1 vs. CGS2, CGS1 vs. CGS3, CGS1 vs. CSG4, CGS1 vs. CGS5, and CGSl vs. CGS6) were carried out. Only subtypes CGS1, CGS5, and CGS6 showed significant (P < 0.001) alterations of DNA methylation. FIG. 3H is a picture showing miRNA expression in the consensus subtypes in the TCGA cohort. Multiple 2-sample t-tests were carried out to identify subtype-specific miRNAs in the TCGA cohort (n = 368). For the selection of subtype CGS1, 5 2-sample t-tests (CGS1 vs. CGS2, CGS1 vs. CGS3, CGS1 vs. CSG4, CGS1 vs. CGS5, and CGS1 vs. CGS6) were carried out. A total of 286 miRNAs had subtype-specific expression (P < 0.01): 217 for CGS1, 6 for CGS2, 9 for CGS3, 14 for CGS4, 8 for CGS5, and 32 for CGS6. FIG. 31 is a picture showing Protein expression in the consensus subtypes in the TCGA cohort. Multiple 2-sample t-tests were carried out to identify subtype-specific protein features in the TCGA cohort (n = 215). For the selection of subtype CGS1, 5 2-sample t-tests (CGS1 vs. CGS2, CGS1 vs. CGS3, CGS1 vs. CSG4, CGS1 vs. CGS5, and CGS1 vs. CGS6) were carried out. A total of 100 protein features had subtype specific expression (P < 0.05): 33 for CGS1, 2 for CGS2, 6 for CGS3, 4 for CGS4, 24 for CGS5, and 31 for CGS6. FIG. 3J is a graph showing box and violin plots of Tumor purity of the consensus subtypes in the TCGA cohort. Each tumor’s purity was estimated by hematoxylin and eosin (H&E) staining and mRNA expression (ESTIMATE). Within each violin, the horizontal distance between the left and right curved boundaries represents the distribution of the percentage of stroma, tumor, or ESTIMATE scores. In the boxes, the boundaries of each box indicate the 25th to 75thpercentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10th and 90th percentiles.

[0015] FIGS. 4A-4F. FIG. 4A is a Venn diagram for selection of gastric stem cell genes from mouse pylorus cells. First, genes whose expression was associated with Lgr5 expression level(Lgr5-high vs. Lgr5-low) in pylorus cells from genetically engineered Lgr5-eGFP-IRES- CreERT2 and Lgr5-DTR-eGFP mice (GEM) were selected as Lrg5-genes. Second, genes whose expression was associated with Aqp5 expression level (Aqp5-high vs. Aqp5-low) in pylorus cells from wild-type (WT) mice were selected as Aqp5-genes. Genes whose expression was significantly different (P < 0.05, by Student t-test) in 2 groups of each GEM or WT mice were selected as the gastric stem cell signature (356 genes). FIG. 4B is a heatmap of gastric stem cell (STEM) signature from mouse pylorus cells. Data are presented in matrix format; each row represents an individual gene, and each column represents a sample. Each cell in the matrix represents the expression level of a gene in an individual tissue sample. FIG. 4C is a picture showing a schematic diagram of prediction model for generating sternness scores in GC tumors. (BCCP, Bayesian compound covariate predictor; LOOCV, leave-one-out cross validation.) FIG. 4D left panel is a graph showing distribution of sternness (STEM) scores in the consensus subtypes in the training cohort. The boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10thand 90th percentiles. FIG. 4D right panel is a graph showing distribution of sternness (STEM) scores in the consensus subtypes in the validation cohort. The boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10thand 90th percentiles. FIG. 4E is a graph showing subtype probabilities from GPICS120 in the validation cohort plotted against their sternness scores. The significance is estimated by Pearson correlation coefficient. FIG. 4F is a graph showing association of Association of sternness score with recurrence risk score. Sternness score was significantly correlated with previously developed TCGA recurrence risk scores (RS) in training and validation cohorts. The significance is estimated by Pearson correlation coefficient.

[0016] FIGS. 5A-5J. FIG. 5A is a picture showing that YTN2 mouse gastric cancer tumors spontaneously regress in xenograft-bearing mice in a Cd8 T cell-dependent manner. Depletion of Cd8 T cells by administration of anti-Cd8 antibodies to mice makes YTN2 cells form tumors in mice. YTN16 cells form tumors well without depletion of Cd8 T cells. FIG. 5B is a picture that shows gene expression data were generated from untreated mouse tumors harvested at 1 week to 3 weeks after transplantation, and 931 genes whose expression was significantly different between YTN2 and YTN16 tumors were selected (P < 0.01 and 2-fold difference) as the Cd8 T cell-reactive signature (CD8TRS) in gastric cancer. Heatmap shows CD8TRS gene expression data at the indicated time points. Data are presented in matrix format; each row represents an individual gene, and each column represents a sample. Each cell in the matrixrepresents the expression level of a gene in an individual tissue sample. FIG. 5C is a picture showing a schematic diagram of prediction model for generating CD8TRS scores in gastric cancer tumors. BCCP, Bayesian compound covariate predictor; LOOCV, leave-one-out cross validation. FIG. 5D is a graph of box plots of CD8 T-cell immunity scores (CD8TRS) in consensus subtypes. The boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10th and 90th percentiles. *P < 0.001 by Student t-test. FIG. 5E is a graph showing correlation of anti-cancer immunity scores with genome instability in gastric cancer. Cd8 T-cell immunity scores were negatively correlated with genome instability in the TCGA gastric cancer cohort (n = 384). Genome instability is represented by summation of absolute gistic2 scores in individual tumors. FIG. 5F is a picture showing gene expression signature associated with response to anti-PD-Ll antibody (atezolizumab) treatment in gastroesophageal tumors. 33 patients with resectable gastroesophageal cancer were treated with neoadjuvant chemoradiotherapy combined with intravenous atezolizumab (1200 mg) (PERFECT trial, NCT03087864). A pathological complete response (pCR) was observed in 30% (10 / 33) of patients. Gene expression data were generated from pretreatment biopsy samples from 29 patients, and 422 genes whose expression was significantly associated with pCR after treatment were selected as the anti-PDLl response signature (P < 0.05 and 0.5-fold difference). FIG. 5G is a picture of a schematic diagram of the prediction model for generating anti-PD-Ll response scores in GAC tumors. BCCP, Bayesian compound covariate predictor; LOOCV, leave-one- out cross validation. FIG. 5H is a graph of box plots of anti-PD-Ll response scores in consensus subtypes. The boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10th and 90th percentiles. *P < 0.001 by Student t-test. FIG. 51 is a graph of scatter plots of CD8 T-cell immunity scores and anti-PD-Ll response scores in training and validation cohorts. Lines indicate loess regression. FIG. 5J is a graph showing correlation of anti-cancer immunity scores with genome instability in gastric cancer. Anti-PD-Ll response scores were negatively correlated with genome instability in the TCGA gastric cancer cohort (n = 384). Genome instability is represented by summation of absolute gistic2 scores in individual tumors.

[0017] FIGS. 6A-6E. FIG. 6A is a picture showing a flow chart depicting selection of patients in ARTIST cohort for applying GPICS 120 and number of patients with each consensus subtype of gastric cancer. FIG. 6B is a graph showing disease-free survival (DFS) of patients (n = 331) included in GPICS120 analysis by type of adjuvant therapy received. XP, capecitabine and cisplatin; XPRT, XP + concurrent radiotherapy. FIG. 6C is a picture showing a schematicdiagram for design of custom probes and generation of gene expression data from ARTIST tumor tissues with the NanoString nCounter platform. FIG. 6D is a picture showing expression patterns of GPICS120 genes in gastric tumor tissues from ARTIST cohort (n = 331). Tumors were stratified according to GPICS120. Subtype-specific gene expression patterns were well conserved in the tumors from the ARTIST cohort. FIG. 6E is a graph showing forest plots showing hazard ratio (HR) for recurrence for adjuvant XPRT treatment over XP treatment in consensus subtypes of GAC tumors from patients in the ARTIST clinical trial. A Cox proportional hazard regression model was used to analyze the interaction between DFS of patients with CGS3 tumors and the rest of patients (non-CGS3 tumors and XPRT treatment. The dotted line represents the 95% confidence interval (CI) of HRs. XP, capecitabine + cisplatin; XPRT, XP + radiotherapy.

[0018] FIGS. 7A-7H. FIG. 7A is a graph showing copy number alteration, mRNA expression, and promoter methylation of HER2 in consensus subtypes of GAC in TCGA cohort. FIG. 7B is a graph showing scatter plots of HER2 copy number alteration (CNA), mRNA expression, and promoter methylation in TCGA cohort. The significance is estimated by Pearson correlation coefficient. FIG. 7C is a graph showing SALL4 mRNA expression in consensus subtypes of GAC in TCGA cohort. FIG. 7D is a graph showing scatter plots of SALL4 copy number alteration (CNA), mRNA expression, and promoter methylation in TCGA cohort. The significance is estimated by Pearson correlation coefficient. FIG. 7E is a graph showing KRAS amplification in consensus subtypes of GAC in TCGA cohort. FIG. 7F is a graph showing scatter plots of KRAS copy number alteration (CNA), mRNA expression, and promoter methylation in TCGA cohort. The significance is estimated by Pearson correlation coefficient. FIG. 7G is a graph showing IGF1 expression in consensus subtypes of GAC in TCGA cohort. FIG. 7H is a graph showing scatter plots of IGF 1 copy number alteration (CNA), mRNA expression, and promoter methylation in TCGA cohort. The significance is estimated by Pearson correlation coefficient.

[0019] FIGS. 8A-8J. FIG. 8A is a picture showing expression patterns of GPICS120 genes in gastric cancer PDX tumors. PDX tumors (n = 114) were stratified according to GPICS120. Subtype-specific gene expression patterns were well conserved in the PDX models. FIG. 8B is a graph showing passage number of PDX tumors for each consensus subtype. The subtypes’ PDX passage numbers did not differ significantly. The boundaries of each box indicate the 25th to 75th percentile, and the black line within the box marks the mean. Whiskers above and below the box indicate the 10th and 90th percentiles. FIG. 8C is a picture showing consensus subtypes in GAC cell lines (n = 37 in CCLE data set). GAC cell lines were stratified accordingto nearest shrunken centroids probability from the GPICS120 predictor. FIG. 8D is a graph showing Log-linear plot demonstrating the relative sensitivity of the 5 GAC cell lines (AGS, NUGC3, SNU668, MKN1, and Hs746T) to ionizing radiation over the range of 0 to 8 Gy as determined by a colony-counting assay. Each point represents the mean of 5 replicates. ** P < 0.001 for AGS or NUSC3 vs. other cells. FIG. 8E is a graph showing categories of upstream effectors or regulators sharing gene networks with CGS3 subtype. Ingenuity Pathway Analysis was performed for CGS3-specific genes (n = 1188) to find potential upstream regulators or effectors sharing networks of genes with the CGS3 subtype. Predicted upstream effectors with significant activation (z-score > 2) were selected for analysis. FIG. 8F is a table showing chemical drugs sharing gene networks with CGS3 subtypes. FIG. 8G is a picture showing upstream effectors of CGS3-specific genes. The lines and arrows represent functional and physical interactions with statin drugs and directions of regulation. FIG. 8H is a panel of graphs showing increased sensitivity to ionizing radiation with statins. Colony-counting assays showed that atorvastatin and simvastatin treatment both sensitized SNU668 and Hs746T non- CGS3 gastric cancer cells to radiation. Seeded cells were exposed to ionizing radiation over the range of 0 to 8 Gy as indicated. P values indicate the significance of the differences at the 8-Gy dose in each experiment. Means (± SEM) of at least 3 experiments are shown. FIG. 81 is a panel of graphs showing colony-counting assays showed that ferrostatin-1 (fer-1), an inhibitor of ferroptosis, abrogates statin-mediated radio sensitization in radiation resistant SNU668 and Hs746T gastric cancer cells. FIG. 8J is a panel of graphs showing that fer-1 makes radiation-sensitive AGS and NUGS3 gastric cancer cells resistant to ionizing radiation.

[0020] FIG. 9 is a picture showing summary of the characteristics of the 6 consensus subtypes of GAC.DETAILED DESCRIPTION OF THE DISCLOSURE

[0021] Gastric cancer is the fifth most common cancer with 1.1 million cases per year. The standard treatment for gastric cancer is surgery, but many patients experience recurrence after surgery. Radiation therapy has been used to prevent recurrence, but it is not always effective. Biomarkers can help to identify patients who are more likely to benefit from radiation therapy. However, single marker do not provide enough prediction power and lacks robustness. There is a growing need for methods and compositions that can help to improve the diagnosis and treatment of the disease.

[0022] The present disclosure provides systematic analysis of genomic and proteomic data of GAC which uncovered 6 subtypes that are significantly associated with benefit of standardand experimental treatments. The methods described herein also identified a gastric cancer subtype (CGS3) which is more sensitive to radiation therapy owing to high basal level of ferroptosis. The newly identified subtypes likely offer new and enhanced information regarding the underlying biology and clinical behavior of GAC. The inventors of the present disclosure also identified genetic and epigenetic alterations associated with the 6 subtypes, which will provide opportunities to develop new intervention for GAC.

[0023] The present disclosure provides methods for classifying genomic subtypes of gastric tumors. In one embodiment, the gastric tumor is a gastric adenocarcinoma. In one embodiment, the methods involve obtaining a digital gene expression profile of the gastric tumor.

[0024] In one embodiment, the methods of the disclosure involve isolating nucleic acids or proteins from a subject, e.g., from a biopsy sample from a subject, such as from a snap-frozen biopsy sample from a subject or a formalin-fixed and paraffin-embedded (FFPE) biopsy sample from a subject. In one embodiment, the nucleic acid is total cellular mRNA, rRNA, cDNA obtained by reverse transcription of total cellular mRNA, or a protein. In one embodiment, the nucleic acids are obtained from an FFPE biopsy sample using methods that are known in the art or are commercially available (such as, but not limited to, QIAamp DNA FFPE Tissue Kit, RNAEASY™ FFPE Kit (Qiagen, Venlo, Netherlands); and MAGMAX™ FFPE DNA Isolation Kit (Life Technologies, Carlsbad, Calif)).

[0025] In one embodiment, the methods of the disclosure include obtaining a digital gene expression profile. “Digital gene expression profile,” as used herein, refers to gene expression information that is based on the generation of sequence tags. In one embodiment, digital gene expression data is obtained and analyzed using a variety of methods known in the art, such as, for example, serial analysis of gene expression (SAGE) (see, e.g., Velculescu et al., Science, 270(5235): 484-487 (1995)), SuperSAGE (see e.g., Matsumura et al., Proc. Natl. Acad. Sci. USA, 100 (26): 15718-15723 (2003)), digital northern analysis (see, e.g., Cao et al., Breast Cancer Research, 10: R91 (2008)), and RNA-seq (see, e.g., Mortazavi et al. Nat Methods, 5(7):621-628 (2008)). In one embodiment, the digital gene expression data is obtained using the NCOUNTER™ gene expression assay available from NanoString Technologies, Inc. The NCOUNTER™ assay detects the expression of up to 800 genes in a single reaction with high sensitivity and linearity across a broad range of expression levels. The NCOUNTER™ assay is based on direct digital detection of mRNA molecules of interest using target-specific, color- coded probe pairs, and does not require the conversion of mRNA to cDNA by reverse transcription or the amplification of the resulting cDNA by PCR. Each target gene of interest is detected using a pair of reporter and capture probes carrying 35- to 50-nucleotide target-specific sequences. In addition, each reporter probe carries a unique color code at the 5' end that enables the molecular barcoding of the genes of interest, while the capture probes all carry a biotin label at the 3' end that provides a molecular handle for attachment of target genes to facilitate downstream digital detection. After solution-phase hybridization between target mRNA and reporter-capture probe pairs, excess probes are removed and the probe / target complexes are aligned and immobilized in an NCOUNTER™ cartridge, which is then placed in a digital analyzer for image acquisition and data processing. Hundreds of thousands of color codes designating mRNA targets of interest are directly imaged on the surface of the cartridge. The expression level of a gene is measured by counting the number of times the color-coded barcode for that gene is detected, and the barcode counts are then tabulated. In one embodiment, the gene expression profile refers to the expression of a group of coordinately expressed genes. In one embodiment, the coordinately expressed genes are associated with a particular genomic subtype of gastric tumor. In one embodiment, the gene expression profile includes genes expressed in a specific cell lineage, stage of differentiation, or during a particular biological response. In one embodiment, the genes reflect biological aspects of the tumors in which they are expressed, such as the cell of origin of the cancer, the nature of the non-malignant cells in the biopsy, and the oncogenic mechanisms responsible for the cancer.

[0026] In one embodiment, the methods of the disclosure involve analyzing the digital gene expression profile using a computer system. In one aspect, the computer system includes at least one processor and a memory storing at least one program for execution by the at least one processor. In one embodiment, the program includes instructions for applying the digital gene expression profile to generate a genomic subtype classification for the gastric tumor. In one embodiment, the program includes a cluster-of-clusters assignment (COCA) which takes as input the binary vectors that represent each previously established subtype and reclusters the tumors according to those vectors. In one embodiment, the program is a GAC predictor of integrated consensus subtype with 120 genes (GPICS120).

[0027] In one embodiment, the digital gene expression profile is classified by the computer system as consensus genomic subtype 1 (CGS1) subtype tumor, a CGS2 subtype tumor, a CGS3 subtype tumor, a CGS4 subtype tumor, a CGS5 subtype tumor, and / or a CGS6 subtype tumor.

[0028] In one embodiment, the digital gene expression profile of the gastric tumor is compared to a reference digital gene expression profile of the genomic subtype of the gastric tumor. In one embodiment, the reference digital gene expression profile includes expression of genes whose expression is not expected to be change among different gastric cancer subtypes.In one embodiment, the reference digital gene expression profile includes expression of housekeeping genes. As a non-limiting example, the housekeeping genes are ACTB, CLTC, GAPDH, GUSB, HPRT1, RPL29, and / or TBP. In one embodiment, the digital gene expression profile is obtained by normalizing the gene expression data to one or more housekeeping genes by using a computer system. In one embodiment, the computer system can include at least one program for execution. In one embodiment, the program includes CodeSet content normalization method, which uses housekeeping genes for normalization (nSolver v 4.0 software from NanoString Technologies). In one embodiment, the digital gene expression profile of the gastric tumor is compared to a reference database.

[0029] Each gastric tumor subtype is associated with expression of one or more genes.

[0030] In one embodiment, the CGS1 subtype tumor includes expression of DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and / or H19.

[0031] In one embodiment, the CGS2 subtype tumor includes expression of AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and / or SI OOP.

[0032] In one embodiment, the CGS3 subtype tumor includes expression of, OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 10B, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and / or CXCL9.

[0033] In one embodiment, the CGS4 subtype includes expression of PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and / or SERPINH1.

[0034] In one embodiment, the CGS5 subtype tumor includes expression of ZIC2, CXCL1, SERPINB5, METRN, HOXA9, CRYBG1, MYB, GSKIP, SPP1, NOXO1, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and / or DOCK5.

[0035] In one embodiment, the CGS6 subtype tumor includes expression of GBP4, HLA- DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and / or CD247.

[0036] In one embodiment, gastric tumors are further analyzed to determine, digital proteomic expression profile, copy number profile, mutation profile, and / or methylation profile. The digital proteomic expression profile is determined by protein arrays, immunologic assay, ELISA, mass spectrometry, isobaric or isotope tagging method. In one embodiment, the subtype-specific digital proteomic expression profile is determined using a program in acomputer system. In one embodiment, the program in the computer system performs a differential analyses using 2-sample t-tests comparing each subtype with other subtypes. In one embodiment, the miRNA profile of the tumor subtypes is also characterized. In one embodiment, the subtype-specific miRNA profile is determined using a program in a computer system. In one embodiment, the program in the computer system performs a differential analyses using 2-sample t-tests comparing each subtype with other subtypes. In one embodiment, the methylation profile of the gastric tumor is determined using Illumina Infinium HumanMethylation450 Array platform. In one embodiment, the subtype-specific methylation expression profile is determined using a program in a computer system. In one embodiment, the program in the computer system performs a differential analyses using 2-sample t-tests comparing each subtype with other subtypes. In one embodiment, the mutation profile includes analysis of somatic mutations and indel (insertion and deletion) calls from exome sequencing of GAC tumors and copy number alteration data. Silent mutations are not included.

[0037] In one embodiment, the CGS1 subtype tumor includes a poor prognosis, a high stem cell characteristic, a high IGF1 expression, and / or a low level of genomic alteration. In one embodiment, genomic alteration is measured as the mutation profile of the CGS1 subtype tumor. In one embodiment the CGS1 subtype is characterized by high expression of MYH11 and collagen VI and low expression of E-cadherin and a-catenin,

[0038] In one embodiment, the CGS2 subtype tumor includes canonical epithelial gene expression. In one embodiment, the CGS2 subtype is characterized by high expression of GKN1, GKN2, LIPF, MUC5AC, and / or MUC6.

[0039] In one embodiment, the CGS3 subtype tumor includes a high copy number alteration, a low immune reactivity, and / or HER2 activation. In one embodiment, the HER2 activation is driven by high HER2 expression due to genomic copy number amplification and promoter demethylation.

[0040] In one embodiment, the CGS4 subtype tumor includes a high copy number alteration, a low immune reactivity, SALL4 expression and / or KRAS activation. In one embodiment, high expression of SALL4 in CGS4 is attributed by both gene amplification and promoter demethylation.

[0041] In one embodiment, the CGS5 subtype tumor includes a high mutation burden, moderately high immune reactivity and / or microsatellite instability.

[0042] In one embodiment, the CGS6 subtype tumor includes a high level of methylation, a high immune reactivity and / or is positive for Epstein Barr virus.

[0043] In one embodiment, a stem score (also herein sternness score) is prepared for each gastric tumor. As used herein, “a stem score or sternness score” is a probability that a particular tumor has a gastric stem cell gene expression profile. In one embodiment, the stem score is calculated using a Bayesian compound covariate predictor (BCCP) model. In one embodiment, the stem score is about 0.1 to 0.2, 0.2 to 0.3, 0.3 to 0.4, 0.4 to 0.5, 0.5 to 0.6, 0.6 to 0.7, 0.7 to 0.8, 0.8 to 0.9, or 0.9 to 1.0. In one embodiment, the stem score is about 0.1 to 0.3, 0.2 to 0.4, 0.3 to 0.5, 0.4 to 0.6, 0.5 to 0.7, 0.6 to 0.8, 0.7 to 0.9, or 0.8 to 1.0. As a non-limiting example, the CGS1 subtype tumor has a stem score of about 0.8. As a non-limiting example, the CGS2 subtype tumor has a stem score of about 0.4 to 0.5. As a non-limiting example, the CGS3 subtype tumor has a stem score of about 0.2 to 0.3. As a non-limiting example, the CGS4 subtype tumor has a stem score of about 0.4. As a non-limiting example, the CGS5 subtype tumor has a stem score of about 0.1 to 0.2. As a non-limiting example, the CGS6 subtype tumor has a stem score of about 0.2 to 0.4.

[0044] In one embodiment, a Cd8 T cell-reactive signature (CD8TRS) score is assigned to a gastric tumor. As used herein, the CD8TRS score refers to a probability that a particular tumor will be sensitive to host immunity, specifically to CD8 T cells. The BCCP algorithm is applied to data sets to generate CD8TRS scores.

[0045] In one embodiment, the CD8TRS score is about 0.1 to 0.2, 0.2 to 0.3, 0.3 to 0.4, 0.4 to 0.5, 0.5 to 0.6, 0.6 to 0.7, 0.7 to 0.8, 0.8 to 0.9, or 0.9 to 1.0. In one embodiment, the CD8TRS score is about 0.1 to 0.3, 0.2 to 0.4, 0.3 to 0.5, 0.4 to 0.6, 0.5 to 0.7, 0.6 to 0.8, 0.7 to 0.9, or 0.8 to 1.0. As a non-limiting example, the CGS1 subtype tumor has a CD8TRS score of about 0.4 to 0.6. As a non-limiting example, the CGS2 subtype tumor has a CD8TRS score of about 0.2. As a non-limiting example, the CGS3 subtype tumor has a CD8TRS score of about 0.2. As a non-limiting example, the CGS4 subtype tumor has a CD8TRS score of about 0.3. As a nonlimiting example, the CGS5 subtype tumor has a CD8TRS score of about 0.3 to 0.4. As a nonlimiting example, the CGS6 subtype tumor has a CD8TRS score of about 0.6.

[0046] In one embodiment, an anti-PD-Ll score is assigned to a gastric tumor. As used herein, an anti PD-L1 score is defined as a probability that a particular tumor will be responsive to anti-PD-Ll antibodies. To identify genes whose expression is associated with response to anti-PD-Ll antibodies (atezolizumab), digital gene expression profile of pretreated gastric tumors, in which gastric cancer patients were treated with atezolizumab after conventional chemoradiotherapy in the neoadjuvant setting are analyzed. A pathological complete response (pCR) was observed in 30% of patients. Gene expression data are generated from pretreatment biopsies of 29 patients by using the HiSeq4000 platform from Illumina (GSE165252). In someembodiments, the anti-PD-Ll score are calculated based on 422 genes whose expression is significantly associated with pCR after treatment (P < 0.05 by 2-sample t-test and 0.5-fold difference); these genes were considered as an anti-PD-Ll response signature. The BCCP algorithm is applied to data sets to generate anti-PD-Ll response scores.

[0047] In one embodiment, anti-PD-Ll response score is about 0.1 to 0.2, 0.2 to 0.3, 0.3 to 0.4, 0.4 to 0.5, 0.5 to 0.6, 0.6 to 0.7, 0.7 to 0.8, 0.8 to 0.9, or 0.9 to 1.0. In one embodiment, the anti-PDLl response score is about 0.1 to 0.3, 0.2 to 0.4, 0.3 to 0.5, 0.4 to 0.6, 0.5 to 0.7, 0.6 to 0.8, 0.7 to 0.9, or 0.8 to 1.0. As a non-limiting example, the CGS1 subtype tumor has an anti- PD-Ll response score of about 0.4 to 0.5. As anon-limiting example, the CGS2 subtype tumor has an anti-PD-Ll response score of about 0.3. As a non-limiting example, the CGS3 subtype tumor has an anti-PD-Ll response score of about 0.3. As a non-limiting example, the CGS4 subtype tumor has an anti-PD-Ll response score of about 0.4. As a non-limiting example, the CGS5 subtype tumor has an anti-PD-Ll response score of about 0.3 to 0.4. As a non-limiting example, the CGS6 subtype tumor has an anti-PD-Ll response score of about 0.3 to 0.4.

[0048] The methods of the disclosure also provide for selecting a treatment option or a therapeutic modality for treating a patient with a gastric tumor based on the subject's gastric tumor classification. Accurate determination of gastric tumor subtype in a subject allows for better selection and application of therapeutic methods. Knowledge about the exact gastric tumor subtype affecting a subject allows a clinician to select therapies or treatments that are most appropriate and useful for that subject, while avoiding therapies that are nonproductive or even counterproductive.

[0049] In one embodiment, any of the gastric genomic subtypes described herein are treated with chemotherapeutic agents alone or in combination with other therapeutic modalities. Non limiting examples of chemotherapeutic agents include, cisplatin, capecitabine, paclitaxel, carboplatin, docetaxel, irinotecan, oxaliplatin, fluorouracil, trifluridine and / or tipiracil (Lonsurf).

[0050] In one embodiment, a patient with a CGS3 subtype is treated with a radiation therapy, a ferroptosis inducer, a HER2 targeting therapy or a combination thereof. In one embodiment, the HER2 targeting therapy reduces the expression of HER2, and / or blocks signaling downstream of HER2. Non-limiting examples of HER2 targeting therapies include, neratinib, trastuzumab, fam-trastuzumab deruxtecan, hyaluronidase, dacomitinib, lapatinib, tucatinib, pertuzumab, margetuximab, ado-trastuzumab emtansine or a combination thereof. In one embodiment, ferroptosis inducer is erastin, RSL5, DIDS, TMZ, MMRi62, Sorafenib, RSL3, ML162, DPI7. DPI10, FIN56, BSO, MEII, PE, AE, SAS, NDP4928, FIN56, Statin, Brequinar,Dexamethasone or a combination thereof. In one embodiment, the CGS3 subtype is treated with a statin drug. The present disclosure provides data that show that the CGS3 subtype is enriched in gene set regulated by statins that induce ferroptosis.

[0051] In one embodiment, a patient with CGS1 subtype tumor, the CGS4 subtype tumor, the CGS5 subtype tumor, and / or the CGS6 subtype tumor is treated with an immunotherapy. The immunotherapy treatment. In one embodiment, the immunotherapy is a monoclonal antibody, a CAR T cell therapy, a checkpoint inhibitor, a cancer vaccine, a cytokine, an immune system modulator or a combination thereof. In one embodiment, the monoclonal antibody and / or CAR T cell therapy are directed to one or more genes whose expression is associated with a CGS1 subtype tumor, the CGS4 subtype tumor, the CGS5 subtype tumor, and / or the CGS6 subtype tumor. In one embodiment, the checkpoint inhibitor is atezolizumab, ipilimumab, pembrolizumab, nivolumab or a combination thereof. In one embodiment, the cytokine is interleukin-2 (IL-2), interferons-alpha (IFN-alpha), interleukin- 12 (IL-12), interleukin- 15 (IL- 15), or a combination thereof. In one embodiment, more than one type of immunotherapy is administered to the subject. In one embodiment, patients with CGS4 subtype tumors are treated with cancer vaccines against cancer testis antigens (CTAs) such as MAGEA3, and MAGEA12 which are expressed in CGS4 subtype tumors.

[0052] In one embodiment, a patient with a CGS1 subtype is treated with an IGF1R targeting therapy. In one embodiment, the IGF1R targeting therapy can include a therapy for reducing the expression of IGF1R or signaling downstream of IGF1R. Non-limiting examples of IGF1R targeting therapy include, but are not limited to, ganitumab, xentuzumab, teprotumumab, teprotumumab, Linsitinib (OSI-906), BMS-754807, Picropodophyllin, NT157, an IGF-1R antisense oligonucleotide, an IGF1R small interfering RNA (siRNA) or a combination thereof.

[0053] In one embodiment, a patient with a CGS4 subtype is treated with a SALL4 and / or a KRAS targeting therapy. In one embodiment, the KRAS targeting therapy is AMG510 (sotorasib), MRTX849 (adagrasib), ARS-3248, LY3499446, D-1553, JDQ443, GDC-6036 or a combination thereof. In one embodiment, the SALL for targeting therapy is a SALL4 antisense oligonucleotide, a SALL4 small interfering RNA (siRNA), thalidomide, or a combination thereof.

[0054] In one embodiment, a patient with a CGS6 subtype tumor is treated with an antiviral agent targeting the Epstein Barr Virus and / or a PI3K pathway inhibitor. As a non-limiting example, the antiviral agent is acyclovir, desciclovir, ganciclovir, adenine arabinoside, phosphonoacetic acid or a combination thereof. In one embodiment, the PI3K inhibitor is Copanlisib, Idelalisib, Umbralisib, Duvelisib, Alpelisib or a combination thereof.

[0055] The present disclosure also provides kits for classifying genomic subtypes of a gastric tumor. Kits useful in the methods of the disclosure include components useful in any of the methods described herein, including for example, primers for nucleic acid amplification, hybridization probes, restriction enzymes (e.g., for RFLP analysis), allele-specific oligonucleotides, means for amplification of nucleic acids, means for analyzing the nucleic acid sequence of a nucleic acid, means for analyzing the amino acid sequence of a polypeptide encoded by a nucleic acid associated with genomic subtype of gastric tumor. In some embodiments, the kits include necessary buffers, nucleic acid primers for amplifying nucleic acids (e.g., nucleic acids that include one or more of the polymorphic markers as described herein), and reagents such as primers and necessary enzymes (e.g., DNA polymerase).

[0056] In one embodiment, the disclosure pertains to a kit for assaying a sample from a subject to classify the genomic subtype of a gastric tumor. In one embodiment, the kit can include primers or probes that hybridize to one or ore genes associated with a particular gastric tumor subtype. In one embodiment, the disclosure pertains to a kit for determining therapeutic modality for treating a patient with a gastric tumor subtype.

[0057] In one embodiment, the kit includes primers or probes for detection of DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and H19 associated with a CGS1 gastric tumor subtype. In one embodiment, the kit includes primers or probes for detection of AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and SIOOP associated with a CGS2 subtype tumor. In one embodiment, the kit includes primers or probes for detection of, OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 10B, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and CXCL9 associated with a CGS3 subtype tumor. In one embodiment, the kit includes primers or probes for detection of PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and SERPINH1 associated with a CGS4 subtype tumor. In one embodiment, the kit includes primers or probes for detection of ZIC2, CXCL1, SERPINB5, METRN, H0XA9, CRYBG1, MYB, GSKIP, SPP1, N0X01, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and DOCK5 associated with a CGS5 subtype tumor. GBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and CD247 associated with a CGS6 subtype tumor.

[0058] In one embodiment, the kit includes one or more labeled nucleic acids capable of detection of one or more genes associated with a particular gastric tumor subtype. Suitable labels include, e.g., a radioisotope, a fluorescent label, an enzyme label, an enzyme co-factor label, a magnetic label, a spin label, an epitope label.

[0059] In one embodiment, the kit further includes a set of instructions for using the reagents included in the kit. In certain embodiments, the kit further includes a collection of digital gene expression data assessed by the kit and classification to a gastric tumor subtype. In one embodiment, the collection of data are provided on any suitable format. In one embodiment, the collection of data is provided on a computer-readable format.EXAMPLES

[0060] The following examples are provided to further illustrate the embodiments of the present disclosure but are not intended to limit the scope of the disclosure. While they are typical of those that might be used, other procedures, methodologies, or techniques known to those skilled in the art may alternatively be used.

[0061] Example 1: Materials and Methods

[0062] Gene expression data from gastric cancers

[0063] TCGA RNA sequencing data for GAC was downloaded from the University of California, Santa Cruz, Genomics Institute on July 11, 2017 (xenabrowser.net / ). RSEM- normalized data were log-transformed. Gene expression data from the Korea cohort were generated in earlier studies; the data are publicly available from the NCBI’s GEO database (GSE26899, GSE2690, and GSE13861). Gene expression data from the Samsung Medical Center (GSE66229), ACRG (GSE66229), Shanxi Hospital (GSE29272), Ruijin Hospital (GSE54129), Singaporel (GSE29998), Singapore2 (GSE15459), KNCC (GSE14208, GSE14209), and Yonsei Hospital (GSE84437) cohorts are available from the NCBI’s GEO database.

[0064] Clinically defined gastric cancer subtypes and associated gene expression signatures

[0065] The intrinsic subtype of gastric cancer, Hippo pathway subtypes, TCGA4 subtype, TCGA risk score, ACRG subtype, Yonsei subtype, mesenchymal subtype, and IncRNA subtypes and their associated gene signatures were described in earlier studies as listed in FIG. IB.

[0066] Statistical analysis of genomic and clinical data

[0067] The BRB Array Tools software program (linus. nci.nih.gov / BRB-ArrayTools.html) was used for analysis of the gene expression data and construction of a prediction model. Aheat map was generated using the Cluster and TreeView software programs, and further statistical analysis was performed using the R language (r-proj ect.org).

[0068] Classification of tumors and cluster of clusters approach (COCA)

[0069] Gastric tumors in 6 discovery cohorts were stratified into subtypes according to the prediction models and algorithms used in the original studies by using the BRB Array Tools software program. Before gene expression data in the discovery sets were collated for the construction of the prediction models, the expression levels of genes in each data set were independently standardized by transforming each gene’s expression level to a mean of 0 and a standard deviation of 1.

[0070] A COCA was used to find consensus subtypes of gastric tumors. The advantage of this approach is that it could combine data from across the 8 classification methods without the need for normalization steps prior to clustering, and each classification method influenced the final integrated result with a weight proportional to the number of subtypes. Briefly, the algorithm takes as input the binary vectors (1 or 0) that represent each of the genomic subtypes and re-clusters the samples according to those vectors by using ConsensusClusterPlus algorithm (v.3.12). On the basis of delta are plot and tracking plot of consensus clustering, k=6 is selected for optimal number of subgroups. Relationships between consensus subtypes and previous classifications were visualized using Circos.

[0071] Analysis of genomic and proteomic data. Somatic mutations and indel (insertion and deletion) calls from exome sequencing of GAC tumors and copy number alteration data were obtained from cbioportal (cbioportal.org / ). Silent mutations were not included. Gene expression data and copy number data were available for 384 tumors.

[0072] For characterization of the 6 consensus subtypes by DNA methylation status, level 3 P-values were used from the Illumina Infinium HumanMethylation450 Array platform. The data set consisted of 377 tumors. To identify subtype-specific methylation patterns, differential methylation analyses on the basis of 2-sample t-tests, comparing each consensus subtype with the other subtypes was performed.

[0073] For miRNA characterization of the 6 consensus subtypes, level 3 reads per million miRNAs mapped (RPM) data from the TCGA data portal were obtained. The RPM data were log2-transformed after adding one pseudocount. To identify subtype-specific miRNAs, differential analyses using 2-sample t-tests, comparing each consensus subtype with the other subtypes was performed.

[0074] For reverse-phase protein array analyses of protein levels, normalized measurements of 220 proteins from 215 GAC tumors were downloaded from the TCPA website(appl.bioinformatics.mdanderson.org / tcpa / ). To identify subtype-specific protein features, differential analyses using of 2-sample t-tests, comparing each consensus subtype with the other subtypes was performed.

[0075] Gastric stem cell signature and sternness scores in gastric tumors

[0076] To identify genes whose expression was significantly associated with gastric stem cells, gene expression data of mouse pylorus cells collected from genetically engineered mice (GEMs) (GSE121803) by using the intestinal stem cell marker Lgr5 and the newly identified gastric stem cell marker Aqp5 was analyzed. The Lgr5-EGFP-IRES-creERT2 GEM model expresses EGFP and a CreERT2 fusion protein from the Lgr5 promoter / enhancer elements. The Lgr5-DTR-EGFP GEM model expresses EGFP and a DTR fusion protein from the Lgr5 promoter / enhancer elements. Lgr5-positive cells from GEM models were isolated by EGFP fluorescence-activated cell sorting (FACS), and Aqp5-positive cells from mouse pylorus tissue were collected by using FACS with anti-Aqp5 antibodies from wild-type mice. Gene expression data were generated by using the Affymetrix Mouse Gene 2.0 ST Array platform. A 2-step selection process was used to select the gastric stem cell signature. First, genes whose expression was associated with Lgr5 expression (Lgr5-high vs. Lgr5-low) in pylorus cells from each GEM model were selected as the Lrg5-genes for each model (P < 0.05, by Student t-test), generating 2 gene lists (2038 genes for Lgr5-EGFP-IRES-creERT2 mice and 3624 genes for Lgr5-DTR-EGFP mice). Likewise, genes whose expression was associated with Aqp5 expression (Aqp5-high vs. Aqp5-low) in pylorus cells from wild-type mice were selected as Aqp5-genes (P < 0.05, by Student t-test, 4564 genes). Second, genes whose expression significantly differed between the low and high groups in both GEM models and wild-type mice were selected as a gastric stem cell signature (356 genes). A Bayesian compound covariate predictor (BCCP) model was used to generate the probability (i.e., a sternness score) that a particular primary tumor would have a gastric stem cell gene expression signature. The BCCP was applied to standardized and combined data sets from mouse tissues (training set) and patients’ primary tumors (test set).

[0077] Immune-response signatures of gastric tumors. To identify genes whose expression was associated with sensitivity or resistance to host immunity, gene expression data of xenografted mouse gastric tumors established from 2 mouse gastric cancer cell lines, YTN2 and YTN16 was analyzed. YTN2 mouse gastric cancer tumors spontaneously regress in xenograft-bearing mice in a Cd8 T cell-dependent manner. Unless host Cd8 T cells are depleted by administration of anti-Cd8 antibodies, YTN2 tumors regress 3 weeks after transplantation in syngeneic mice. YTN16 cells spontaneously form tumors without depletionof Cd8 T cells. Gene expression data were generated from untreated mouse tumors harvested at 1 to 3 weeks after transplantation by using the HiSeq X Ten or NovaSeq 6000 platforms from Illumina. Normalized gene expression data were obtained from the Gene Expression Omnibus database (GSE146027) and analyzed to find genes whose expression was significantly different between the 2 mouse tumors, yielding 931 genes (P < 0.01 by 2-sample t-test and 2-fold difference) containing the Cd8 T cell-reactive signature (CD8TRS) in gastric cancer. The BCCP algorithm was applied to standardized and combined data sets from mouse tumors (training set) and patients’ primary tumors (test set) to generate CD8TRS scores as described in the section on the generation of sternness scores.

[0078] To identify genes whose expression was associated with response to anti-PD-Ll antibodies (atezolizumab), gene expression data of pretreated gastric tumors from the PERFECT trial, in which gastric cancer patients were treated with atezolizumab after conventional chemoradiotherapy in the neoadjuvant setting (NCT03087864) was analyzed. A pathological complete response (pCR) was observed in 30% of patients. Gene expression data were generated from pretreatment biopsies of 29 patients by using the HiSeq4000 platform from Illumina (GSE165252). Analysis revealed 422 genes whose expression was significantly associated with pCR after treatment (P < 0.05 by 2-sample t-test and 0.5-fold difference); these genes were considered as an anti-PD-Ll response signature. The BCCP algorithm was applied to standardized and combined data sets from the PERFECT trial’s tumors (training set) and primary tumors (test sets) to generate anti-PD-Ll response scores.

[0079] ARTIST cohort for assessing efficacy of adjuvant chemoradiation in the consensus subtype

[0080] The ARTIST study included 458 patients with radically resected (R0) nonmetastatic gastric cancer were enrolled and randomized to either adjuvant XP (capecitabine plus cisplatin) or XPRT (XP for 2 cycles followed by radiotherapy concurrently with capecitabine, followed by 2 additional XP cycles). The planned treatment was completed in 172 (75.4%) patients in the XP arm and 188 (81.7%) patients in the XPRT arm. Of the patients who completed treatment, tissues for NanoString nCounter experiments were available from 158 (XP) and 173 (XPRT) patients.

[0081] Generation and analysis of gene expression data with NanoString nCounter platform

[0082] A customized panel for NanoString nCounter experiments (120 genes and 7 housekeeping reference genes) was compiled based on the gene list in the GPICS120 prediction model and stably expressed reference genes in gastric cancer tissues. Reference genes included ACTB, CLTC, GAPDH, GUSB, HPRT1, RPL29, and TBP. Custom codesets of 127 geneswere designed according to NanoString’s standard protocol. RNA was isolated from 10 pM- thick formalin-fixed, paraffin-embedded tissue samples mounted on slides using the Roche High Pure RNA Paraffin Kit according to the manufacturer’s protocol. Concentration and fragmentation of purified RNA were assessed by using an Invitrogen qubit fluorometer and an Agilent bioanalyzer at the Advanced Technology Genomics Core at The University of Texas MD Anderson Cancer Center. Hybridization of RNA to NanoString probes was carried out according to the manufacturer’s protocol. After incubating at 65 °C for hybridization for 20 to 22 hours, the reaction products were loaded to a GEN2 Prep Station for washing and immobilizing signals to a cartridge. The data were collected by Digital Analyzer (NanoString Technologies). Gene expression was processed and normalized by using the CodeSet content normalization method, which uses housekeeping genes for normalization (nSolver v 4.0 software from NanoString Technologies). The GPICS120 algorithm was applied to normalized gene expression data to stratify patients in the ARTIST cohort into consensus subtypes.

[0083] Gene expression data from GC PDX models

[0084] Gastric cancer PDX tumors were established. mRNA expression data from PDX tumors were generated on the Illumina HiSeq2500 platform. For bioinformatic analysis of transcriptome sequencing data, raw RNA sequencing data were first cleaned up by removing contamination reads that preferentially mapped to the mouse genome (UCSC MM10). Clean reads were mapped to reference genes (ENSEMBL GRCh37.66) by Bowtie, and gene expression was calculated by MMSEQ. The GPICS120 predictor was applied to gene expression data from PDX models to stratify them into the 6 consensus subtypes.

[0085] Radiation treatment of gastric cancer cells. AGS, MKN1, SNU668, Hs746T, NUGC3 cells were grown and maintained in RP MI-1640 medium (Hy clone, Logan, UT, USA) supplemented with 10% heat-inactivated bovine serum (Hyclone) (56 °C for 30 min) and 1% penicillin-streptomycin (Coming Inc., Coming, NY, USA). Cells were maintained at 37 °C in a humidified incubator containing 5% CO2. Short tandem repeat DNA profiling was performed to authenticate cell lines used in the experiments.

[0086] To determine the radiation sensitivity of gastric cancer cells, colony-counting assays were conducted. Cells were first trypsinized, diluted, and enumerated; next, they were seeded onto 6-well plates in triplicate. After 24 h of incubation, the cells were exposed to different doses of X-ray radiation (0, 2, 4, 6, and 8 Gy) using an XStrahl RS225 cabinet at room temperature with 195 kV / 15 mA X-rays, producing a dose rate of 1.6 Gy per minute. Irradiated and untreated control cells were subsequently cultured for 8-21 days. The surviving cell-derived colonies were stained with crystal violet solution (0.5% crystal violet in 50% methanol) and counted.

[0087] Atorvastatin (S5715) and Fer-1 (S7243) were purchased from Selleckchem (Houston, TX, USA). Simvastatin (S6196) was purchased from Sigma (St. Louis, MO, USA). Cells were treated with atorvastatin, simvastatin (Hs746T: 10 nM; SNU668:100 nM), or fer-1 (50 nM) 12 h after seeding. After 24 h of incubation, the cells were irradiated and cultured for 9-24 days. To determine whether fer-1 could abrogate statin-induced radio sensitization, cells were treated with 50 nM fer-1 24 h after irradiation. Irradiated and drug-treated cells were cultured, stained, and counted as described above.

[0088] Example 2: Meta- analysis of genomic subtypes of primary GAC

[0089] For discovery and validation of consensus genomic subtypes (CGSs) of GAC, 10 GAC data sets (N = 2527) were collected and divided the data into 2 groups: a discovery / training set (Korea, Samsung, TCGA, Shanxi, Ruijin, and Singaporel cohorts, n = 1427) and a validation set (ACRG, Korea National Cancer Center [KNCC], Yonsei Medical Center, and Singapore2 cohorts, n = 1100) (FIG. 1A). GAC tumors in the discovery set were stratified according to the 8 genomic subtyping algorithms (FIG. IB) that were independently developed and validated in previous studies. FIG. 1A summarizes the strategy and workflows of the analyses. In order to comprehensively integrate the subtype information from 8 classification methods and identify consensus subtypes, a previously established method, cluster-of-clusters assignment (COCA), was used, which takes as input the binary vectors that represent each subtype of the 8 classifications and reclusters the tumors according to those vectors as described in method. COCA of binary vectors revealed 6 CGSs of GAC (FIG. 1C). While the delta area plots indicated that 5 or 6 was the optimal number of subtypes, tracking plots showed that one subtype had only one sample when k=5 (FIG. ID, FIG. IE). Therefore, k=6 was selected as the optimal number of subtypes.

[0090] Not surprisingly, some of the 6 CGSs were subsets of previously recognized genomic subtypes. For example, the MD Anderson mesenchymal phenotype (MP) subtype was highly similar to CGS1. However, the MD Anderson epithelial phenotype (EP) subtype was split among CGS2, CGS3, CGS4, CGS5, and CGS6 (FIG. IF left panel). Likewise, the genomic diffuse (GIDF) subtype was split among CGS1 and CGS4, and the genomic intestinal (GINT) subtype was split among CGS2, CGS3, and CGS5 (FIG. IF middle panel). In addition, the genomically stable (GS), microsatellite instability (MSI), and Epstein Barr virus (EBV) subtypes of TCGA4 classification were similar to CGS1, CGS5, and CGS6, respectively, while the chromosome instability (CIN) subtype was split among CGS2, CGS3, and CGS4 (FIG. IFright panel). Taken together, these relationships suggest that the COCA approach not only rediscovered subtypes similar to previously recognized subtypes (CGS1, CGS5, and CGS6) but also discovered novel subtypes that were not recognized by previous studies (CGS2, CGS3, and CGS4). Overall, new analysis increased confidence in these subtypes because of the largest sample size analyzed to date. Further confirming the concordance of the 6 consensus subtypes with previous classifications, the 6 consensus subtypes were significantly associated with patient prognosis (P = 1.5 x 10, log-rank test; FIG. 1G).

[0091] Example 3: Validation of 6 CGSs

[0092] Because the TCGA study provided the most comprehensive genomic profiling among the included studies, its data were selected for development of the prediction model. By applying multiple statistical tests to the gene expression data from the TCGA cohort, 20 genes per subtype as a subtype-specific gene expression signature were selected (FIG. 2A, FIG. 2B). To validate the 6 subtypes and their clinical significance in independent cohort of GAC patients, the “GAC predictor of integrated consensus subtype with 120 genes” (GPICS120) was constructed by using a previously established algorithm. For this validation, gene expression and clinical data from the validation set (n = 1100) were used. When patients in the validation cohort were classified according to GPICS120 (FIG. 2A, FIG. 2C), the 6 consensus subtypes were all significantly associated with overall survival (OS) rate (P = 9.6 x 10, by the log-rank test) (FIG. 2D). As in the discovery set, CGS1 and CGS4 were associated with the worst OS rates. In contrast, CGS5 and CGS6 were associated with the best OS rates. Patients in CGS2 and CGS3 had OS rates that were worse than those of CGS5 and CGS6 but better than CGS1 and CGS4. Recurrence-free survival rates of the CGSs were highly similar to OS rates (P = 3.2 x 10).

[0093] Example 4: Genomic, epigenomic, and proteomic characteristics of the CGSs

[0094] Next, the association of genomic and epigenomic characteristics with the 6 CGSs in the TCGA cohort was assessed to gain additional insight into each subtype’s biology. Mutation burden was significantly higher in CGS5 than in all other subtypes (P < 0.001 in all comparisons, FIG. 3A left panel). Copy number alterations also significantly differed among the subtypes, with CGS3 and CGS4 having the most (FIG. 3A right panel, FIG. 3B). Both of these findings suggest that differences in genomic alterations are well reflected in the consensus subtypes.

[0095] Next mutations significantly associated with the subtypes were identified (FIG. 3C). Owing to its high mutation burden, CGS5 had the highest number of associated mutated genes, including ARID 1 A, XYLT2, RPL22, and PGM5 (FIG. 3D). More interestingly, the mutationrate of 2 tyrosine kinases, ERBB3 and ERBB4, was significantly higher in CGS5 than in the other subtypes (FIG. 3E, FIG. 3F). TP53 was the most frequently mutated gene, with relatively high mutation rates in all subtypes, but TP53 mutation rate was highest in CGS3. CGS1 was most associated with CDH1 mutations, and CGS3 was most associated with CDH11 mutations. Interestingly, PIK3CA mutations were highly enriched in CGS6, suggesting that the phosphatidylinositol 3-kinase (PI3K) pathway is highly activated in CGS6.

[0096] Furthermore, an analysis of TCGA data revealed subtype-specific methylation patterns (FIG. 3G). Most strikingly, CGS6 was the subtype most strongly associated with hypermethylation, indicating a potential connection between PI3K pathway activation and hypermethylation in GAC. CGS5 was also associated with hypermethylation, albeit to a lesser degree. Interestingly, CGS1 displayed both hypermethylation and hypomethylation. However, the CGS2, CGS3, and CGS4 subtypes did not show any specific methylation pattern, suggesting that epigenetic alterations play limited roles in the development of these subtypes.

[0097] Analysis of microRNA (miRNA) expression data showed that CGS1 had the most dramatic alterations in miRNA expression (FIG. 3H). Of 286 subtype-specific miRNAs, 75% (n = 217) were CGS1 -specific. While the vast majority of CSG1 -specific miRNAs were downregulated, miRNAs miR-99a-5p, miR-99a-3p, miR-100-5p, miR-199-5p, miR-199-3p, and miR-133a-3p were overexpressed in CGS1. Interestingly, most miRNAs that were specific to the other consensus subtypes were upregulated. CGS3-specific miRNAs included miR-192- 5p, miR-192-3p, miR-194-3p, and miR-194-5p. The hypermethylated subtype CGS6 was characterized by overexpression of miR-31-3p and miR-31-5p, indicating a potential connection of these miRNAs to epigenetic regulation.

[0098] Reverse-phase protein array analysis revealed CGS 1 , CGS5, and CGS6 to have highly distinctive proteomic characteristics (FIG. 31). Interestingly, the CGS1 subtype was characterized by high expression of MYH11 and collagen VI and low expression of E-cadherin and a-catenin, which play essential roles in cell-to-cell interaction of epithelial cells, suggesting that CGS1 tumors undergo EMT that might account for the poor clinical outcomes associated with this subtype. CGS5 was characterized by high expression of claudin 7, ASNS, and 4E- BP. Immune-related signaling proteins such as STAT5, Syk, and Lek were highly expressed in CGS6, suggesting that this subtype’s high EBV load might trigger massive activation of immune cells.

[0099] In good agreement with the high likelihood of EMT in CGS1, hematoxylin and eosin staining analysis of TCGA tumors revealed CGS1 to have the highest stromal fraction and lowest tumor fraction in the tumor mass (FIG. 3 J), suggesting that EMT may trigger infiltrationof stromal cells into the tumor mass. mRNA expression-based estimations of tumor purity (ESTIMATE) yielded consistent results.

[0100] Example 5: Stem-cell features of the subtypes

[0101] To determine the stem-cell characteristics of each GAC subtype, sternness scores of GAC tumors were generated by applying a gastric stem-cell gene expression signature, which was extracted by analyzing gene expression data from mouse gastric stem cells, to gene expression data from GAC tumors (FIG. 4A, FIG. 4B, FIG. 4C). Interestingly, the subtype with the poorest prognosis, CGS 1, had the highest sternness score, whereas the subtype with the best prognosis, CGS5, had the lowest sternness score in both the training and validation cohorts (FIG. 4D left and right panel). Consistently, a correlation analysis of the sternness score with GPICS120 probability for each subtype showed that the CGS1 subtype was significantly positively correlated with the sternness score (r = 0.567, P = 1.3 x 10), whereas the CGS5 subtype was significantly but negatively correlated with the sternness score (r = -0.402, P = 5.5 x 10, FIG. 4E). In agreement with this, the sternness score was significantly correlated with the TCGA recurrence risk score that was previously developed and validated for predicting recurrence after treatment (FIG. 4F), further supporting significant association of a sternness phenotype with poor prognosis.

[0102] Example 6: Clinical implications of the consensus subtypes for immunotherapy

[0103] Immunotherapy with antibodies that the block inhibitory immune checkpoint regulators has yielded encouraging results in GAC patients. However, not all patients benefit from immunotherapy. Because this unequal benefit might be caused by intrinsic sensitivity or resistance of cancer cells to host immune activity, these features were assessed in each subtype by using gene expression signatures from mouse gastric tumors that are sensitive or resistant to CD8 T cell-dependent anticancer immunity (CD8 T-cell reactive signature [CD8TRS] scores (FIG. 5A, FIG. 5B, FIG. 5C). CGS6, with the highest fraction of EBV-positive tumors, had the highest CD8TRS score, i.e., the highest sensitivity to CD8 T cell-dependent anticancer immunity (FIG. 5D). Interestingly, CGS1, with the poorest prognosis and highest sternness, had the second highest CD8TRS score, suggesting that tumors in this subtype might be sensitive to CD8 T cell-dependent anticancer immunity or related immunotherapy. CGS3 and CGS4, which had high genomic instability, had lower CD8TRS scores. Interestingly, CSG1 and CGS6, which had high CD8TRS scores, had very low genomic instability. In agreement with this observation, CD8TRS score was significantly negatively correlated with genomic instability (FIG. 5E).

[0104] To further validate these findings, a gene expression signature significantly associated with response to treatment with a neoadjuvant anti-PD-Ll antibody (atezolizumab) in patients with gastroesophageal cancer was applied to generate anti-PD-Ll response scores (FIG. 5F, FIG. 5G). Consistent with the CD8TRS scoring results, CGS3 and CGS4 subtypes had the low potential response to anti-PD-Ll or anit-PD-1 agents, while CGS1 had the highest potential response (FIG. 5H). However, contrary to its CD8TRS score, the CGS6 subtype had only a moderate potential response to anti-PD-Ll treatment. Nonetheless, overall, CD8TRS score was significantly correlated with anti-PD-Ll response score (FIG. 51), supporting the conclusion that these 2 independently generated immunity signatures reflect anticancer immunity in GACs. As was the CD8TRS score, anti-PD-Ll response score was significantly but negatively correlated with genomic instability (FIG. 5J, strongly supporting a potential mechanistic connection between chromosomal instability and host immune response. Taken together, these results indicate that the consensus subtypes are strongly associated with potential response to immunotherapy.

[0105] Example 7: Clinical implications of the consensus subtypes for chemoradiation therapy

[0106] Although adjuvant chemotherapy has become the standard treatment for GAC after surgery, the benefit of adjuvant chemoradiation therapy is not clearly defined. Thus, the goal was to determine whether the consensus subtypes were associated with clinical benefit from adjuvant chemoradiation therapy. To do this, the NanoString nCounter platform was used to generate gene expression data of GACs from patients in the ARTIST trial who were treated with adjuvant capecitabine and cisplatin (XP) or XPRT (XP followed by radiotherapy) and followed for disease-free survival (DFS) as the primary endpoint. Of the registered patients, only those who completed treatment and had available tissues in the analysis were included (n = 331, FIG. 6A). Similar to the original report, addition of radiotherapy to XP did not significantly reduce recurrence after surgery in the included patients (FIG. 6B, FIG. 6C). Application of GPICS120 to the gene expression data of the 331 ARTIST tumors stratified them into the 6 consensus subtypes (55 for CGS1, 84 for CGS2, 46 for CGS3, 60 for CGS4, 51 for CGS5, and 35 for CGS6, FIG. 6D). Most importantly, subtype CGS3 showed a significant benefit from XPRT treatment compared to XP treatment (FIG. 6E). For CGS3, the hazard ratio (HR) for recurrence in patients receiving XPRT treatment compared to XP treatment was 0.22 (95% confidence interval [CI], 0.058-0.81, P = 0.02, FIG. 6E). However, no significant benefit from XPRT treatment was observed in the rest of the subtypes, suggestingthat the benefit of the addition of radiotherapy might be attributable to differences in the underlying biology among subtypes.

[0107] Example 8: Potential therapeutic targets in consensus subtypes

[0108] Because previous studies showed that trastuzumab, which targets HER2 (ERBB2), is the most effective treatment for GAC with genomic copy number amplification or high expression of HER2, copy number alterations of HER2 in the consensus subtypes was assessed. Interestingly, CGS3 had the highest amplification of HER2 among the 6 subtypes (FIG. 7A, FIG. 7B). In agreement with this, mRNA expression of HER2 was also highest in CGS3, and its expression and copy number were significantly positively correlated More interestingly, methylation of the HER2 promoter was lowest in CGS3, and HER2 expression and promoter methylation were significantly but negatively correlated, suggesting that HER2 expression in CGS3 is increased by both genetic and epigenetic alterations.

[0109] Interestingly, similar to HER2, expression of SALL4 was positively correlated with copy number but negatively correlated with promoter methylation (FIG. 7C, FIG. 7D), indicating that its expression in GAC is also regulated by both genetic and epigenetic alterations. However, unlike HER2, SALL4 expression was highest in CGS4. Next, because a recent study showed that amplification is the major activation mechanism of KRAS in GAC cancer and that KRAS amplification is significantly associated with resistance to MAPK blockade, copy number alterations of KRAS by subtype were assessed. KRAS amplification was highest in CGS4 and was significantly associated with KRAS mRNA expression (r = 0.81, P = 1.2 * 10, FIG. 7E, FIG. 7F). However, unlike that of HER2 and SALL4, KRAS promoter methylation was not significantly associated with mRNA expression, indicating that KRAS activation in GAC is mainly driven by copy number alteration.

[0110] One of the genes most significantly overexpressed in the poorest-prognosis CGS1 subtype was IGF1 (FIG. 7G, FIG. 7H), a ligand for IGF1R that is a known potential therapeutic target for many cancers. mRNA expression of IGF1 was negatively correlated with IGF1 promoter methylation (r = -0.459, P = 2.1 x 10) but not with copy number alteration, indicating that activation of IGF1 in GAC is largely mediated by epigenetic mechanisms.

[0111] Example 9: Consensus subtypes in preclinical GAC models

[0112] Next GPICS120 was applied to the gene expression data of 114 GAC patient-derived xenograft (PDX) tumors. Of these tumors, 6, 14, 26, 31, 30, and 7 were classified as subtypes CGS1, CGS2, CGS3, CGS4, CGS5, and CGS6, respectively (FIG. 8A). Importantly, the expression patterns of the GPICS120 genes were well conserved in PDX tissues, suggesting that PDX tumors recapitulate the molecular and biological characteristics of primary GACwell. Interestingly, most PDX tumors appeared to be stable, as the passage number of PDX models differed only slightly among subtypes, except for CGS1, which was much lower (FIG. 8B).

[0113] Because established cancer cell lines are the most practical experimental models, GPICS120 was applied to gene expression data of GAC cell lines from the Cancer Cell Line Encyclopedia (CCLE) data set and stratified 37 cell lines according to GPICS120 (FIG. 8C). Next the associations of the consensus subtypes with sensitivity to ionizing radiation was examined in these cell lines. In agreement with the clinical association of CGS3 subtype with radiation sensitivity in the ARTIST cohort, 2 cell lines of the CGS3 subtype, AGS and NUGC3, were more sensitive to radiation than were cell lines of other subtypes (FIG. 8D), suggesting that these cell lines may recapitulate the biology of radiation sensitivity at a cellular level. To uncover the underlying biology associated with sensitivity to radiation in CGS3 tumors, network analysis with genes specific to the CGS3 subtype was performed. Since the 20 genes specific to CGS3 in GPICS120 were too few to run network analysis, genes that were differentially expressed in CGS3 tumors compared to non-CGS3 tumors, yielding 1188 genes were re-selected. Next, gene network analysis was performed to identify potential upstream effectors or regulators of 1188 genes. Interestingly, a large fraction of predicted effectors of gene networks in CGS3 were chemical drugs (FIG. 8E). In particular, multiple statins predicted to regulates genes specific to CGS3 subtype (FIG. 8F, FIG. 8G), indicating that the cellular state of CGS3 tumors is similar to that of tumor cells treated with statins. This association of statins with genes specific to the radiation-sensitive CGS3 subtype suggests that statins may sensitize GAC tumors to radiation treatment.

[0114] To test this possibility, non-CGS3 cell lines were treated with statins during radiation exposure. In agreement with the predictions of the network analysis, both atorvastatin and simvastatin sensitized 2 non-CGS3 cell lines (SNU668 andHs746T) to ionizing radiation (FIG. 8H). Since statins are known to promote ferroptosis by inhibiting synthesis of coenzyme Q10, which is the end product of the mevalonate pathway and a major agent for reducing oxidized fatty acids, it was postulated that statins sensitize GAC cells to radiation by enhancing baseline ferroptosis. Indeed, treatment with ferrostatin-1 (fer-1), an inhibitor of ferroptosis, abrogated statin-induced radiation sensitivity, strongly suggesting that radiation sensitivity is mediated by an increase of ferroptosis (FIG. 81). In good agreement with this, fer-l-treated CGS3- subtype cells (AGS and NUGC3) were resistant to radiation (FIG. 8 J), further supporting the notion that sensitivity to radiation in the CGS3 subtype is mediated by ferroptosis.

[0115] Example 10: Discussion

[0116] To form a consensus from independently identified molecular subtypes of gastric cancer, subtype information for 8 previously discovered genomic subtypes was integrated and reanalyzed. This analysis identified 6 distinct genomic subtypes of GAC (FIG. 9). Importantly, clinical outcomes, including overall survival rates and potential responses to treatment, differed significantly among the subtypes. Results indicate that these 6 consensus subtypes likely offer new and enhanced information regarding the underlying biology and clinical behavior of GAC.

[0117] CGS1 is characterized by the poorest prognosis, strongest stem cell features, and a high potential response to immunotherapy. While gene expression in CGS1 tumors was the most similar among the subtypes to that from mouse gastric stem cells, it is currently unknown if this high similarity reflects the origin of cancer cells or the high fraction of cancer stem cells in the tumor mass. Because acquisition of EMT is a common event in poor-prognosis tumors with stem-cell features, the high expression of mesenchymal homeobox transcription factor BARX1, one of CGSl-specific genes in GPICS120, suggests that BARX1 might contribute to EMT and the stem cell phenotype. In fact, previous studies showed that BARX1 is an essential factor for stomach development during organogenesis. In agreement with this, the mesenchymal marker VIM was another CGSl-specific gene in GPICS120, providing further support that EMT is a maj or phenotype of CGS1. Importantly, CGS1 is significantly associated with CD8 T cell— mediated eradication of cancer cells and potential response to anti-PD-Ll or anti-PD-1 treatment, although this finding needs further validation in preclinical and clinical studies.

[0118] The CGS2 subtype has gene expression patterns similar to those of normal gastric epithelial cells, as reflected by its high expression of GKN1, GKN2, LIPF, MUC5AC, and MUC6. Because the sternness score of the CGS2 subtype is not the lowest among the 6 subtypes, it is unlikely that differentiation status is not fully accountable for expressing normal gastric epithelial genes, but likely that CGS2 GAC cells partially maintain a normal physiologic transcriptome program even after malignant transformation. This subtype also was associated with a better prognosis than CGS1 but a worse one than CGS5. Notably, the CGS2 subtype was similar to a normal-like subtype in breast cancer that also expressed normal epithelial genes and has a mid-range prognosis. Although major drivers of the CGS2-subtype have not been identified yet, the shared cellular features in 2 different tumor types indicate that maintaining a normal transcriptomic program may provide a certain advantage to cancer cells, for instance by allowing them to evade anticancer immune activity.

[0119] The CGS3 subtype is characterized by low expression of the cytotoxic T-cell marker CD 8 A, immune cytolytic genes such as GZMB and GZMH, and chemokine ligands such as CCL5, CXCL9, and CXCL10, suggesting that anticancer immunity is suppressed in this subtype. In fact, the CGS3 subtype has the lowest CD8TRS and anti-PD-Ll response scores among the 6 subtypes. Although its mutation burden is relatively low, its overall genomic instability is very high, as reflected in its high number of copy number alterations. Because previous studies indicated that high genomic instability is significantly associated with poor response to anti-CTLA-4 immunotherapy, high genomic instability might account for the low immune activity in the CGS3 subtype. In contrast to the low immune activity in this subtype, CGS3 tumors have high expression of HER2 because of both genomic copy number amplification and promoter demethylation, indicating that tumors of the CGS3 subtype might be more sensitive to targeting of HER2 with trastuzumab. Importantly, the CGS3 subtype is significantly associated with a benefit from the addition of radiation therapy to adjuvant chemotherapy, suggesting that CGS3-subtype cancer cells might be more sensitive to radiation than other subtypes. The high sensitivity of CGS3 GAC cells to radiation was further demonstrated in a cell culture model. Analysis of genomic data suggested that the CGS3 subtype may have a high basal level of ferroptosis, as reflected by the enrichment of a gene set regulated by statins that induces ferroptosis in cancer cells. In good agreement with these predictions, statin treatment sensitized radiation-resistant CGS3 GAC cells to ionizing radiation. Therefore, a novel therapeutic approach using a statin drug in addition to chemoradiation would warrant further investigation in preclinical models.

[0120] Like CGS3, CGS4 has high genomic instability. However, in CGS4 tumors, KRAS is one of the most highly amplified and highly expressed. Previous observations also showed that KRAS is frequently activated via amplification in gastroesophageal cancer. Like HER2 in CGS3, high expression of SALL4 in CGS4 is attributed by both gene amplification and promoter demethylation. SALL4 is a C2H2 zinc finger transcription factor that is essential for embryonic development and frequently activated in many cancers. Because recent studies identified SALL4 as a neosubstrate of the molecular glue thalidomide and its derivatives, it will be interesting to determine the efficacy of thalidomide in CGS4 tumors in future preclinical models. In addition to the KRAS and SALL4 oncogenes, CGS4 tumors highly express cancertestis antigens (CTAs) such as MAGEA3 and MAGEA12, indicating that they might be good candidates for testing CTA-mediated cancer vaccine therapy.

[0121] CGS5 and CGS6 are characterized by high mutation rates and genomic hypermethylation, respectively. Interestingly, while CGS5 showed the highest number ofsubtype-specific mutations owing to its high overall mutation burden, PI3KCA is most highly mutated in the CGS6 subtype, suggesting that CGS6 tumors might be more sensitive to inhibitors of the PI3K pathway. Similar to the CGS1 subtype, the CGS6 subtype has high basal level immune activity as reflected in its high CD8TRS scores.

[0122] Sung, H., et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA: a cancer journal for clinicians 71, 209-249 (2021).

[0123] Ajani, J.A., et al. Gastric adenocarcinoma. Nature reviews. Disease primers 3, 17036 (2017).

[0124] Nagaraja, A.K., Kikuchi, O. & Bass, A.J. Genomics and Targeted Therapies in Gastroesophageal Adenocarcinoma. Cancer discovery 9, 1656-1672 (2019).

[0125] Joshi, S.S. & Badgwell, B.D. Current treatment and recent progress in gastric cancer. CA: a cancer journal for clinicians 71, 264-279 (2021).

[0126] Wang, Q., Liu, G. & Hu, C. Molecular Classification of Gastric Adenocarcinoma. Gastroenterology research 12, 275-282 (2019).

[0127] TCGAConsortium. Comprehensive molecular characterization of gastric adenocarcinoma. Nature 513, 202-209 (2014).

[0128] Oh, S.C., et al. Clinical and genomic landscape of gastric cancer with a mesenchymal phenotype. Nat Commun 9, 1777 (2018).

[0129] Cheong, J.H., et al. Predictive test for chemotherapy response in resectable gastric cancer: a multi-cohort, retrospective analysis. Lancet Oncol 19, 629-638 (2018).

[0130] Cristescu, R., et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med 21, 449-456 (2015).

[0131] Cho, J.Y., et al. Gene expression signature-based prognostic risk score in gastric cancer. Clin Cancer Res 17, 1850-1857 (2011).

[0132] Sohn, B.H., et al. Clinical Significance of Four Molecular Subtypes of Gastric Cancer Identified by The Cancer Genome Atlas Project. Clin Cancer Res 23, 4441-4449 (2017).

[0133] Lee, J., et al. Nanostring-based multigene assay to predict recurrence for gastric cancer patients after surgery. PloS one 9, e90133 (2014).

[0134] Wang, G., et al. Comparison of global gene expression of gastric cardia and noncardia cancers from a high-risk population in China. PloS one 8, e63826 (2013).

[0135] Holbrook, J.D., et al. Deep sequencing of gastric carcinoma reveals somatic mutations relevant to personalized medicine. Journal of translational medicine 9, 119 (2011).

[0136] Ooi, C.H., et al. Oncogenic pathway combinations predict clinical prognosis in gastric cancer. PLoS genetics 5, el000676 (2009).

[0137] Kim, H.K., et al. A gene expression signature of acquired chemoresistance to cisplatin and fluorouracil combination chemotherapy in gastric cancer patients. PloS one 6, el 6694 (2011).

[0138] Tan, LB., et al. Intrinsic subtypes of gastric cancer, based on gene expression pattern, predict survival and respond differently to chemotherapy. Gastroenterology 141, 476-485 (2011).

[0139] Lee, K.W., et al. Development and Validation of a Six-Gene Recurrence Risk Score Assay for Gastric Cancer. Clin Cancer Res 22, 6228-6235 (2016).

[0140] Shin, M.K., et al. Long non-coding RNAs are significantly associated with prognosis and response to therapies in gastric cancer. Clin Transl Med 11, e421 (2021).

[0141] Simon, R., et al. Analysis of gene expression data using BRB-ArrayTools. Cancer Inform 3, 11-17 (2007).

[0142] Eisen, M.B., Spellman, P.T., Brown, P.O. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 95, 14863-14868 (1998).

[0143] Lee, J.S., et al. Application of comparative functional genomics to identify best-fit mouse models to study human cancer. Nature genetics 36, 1306-1311 (2004).

[0144] Kaposi-Novak, P., et al. Met-regulated expression signature defines a subset of human hepatocellular carcinomas with poor prognosis and aggressive phenotype. J Clin Invest 116, 1582-1595 (2006).

[0145] Hoadley, K.A., et al. Multiplatform analysis of 12 cancer types reveals molecular classification within and across tissues of origin. Cell 158, 929-944 (2014).

[0146] Kandoth C Fau - Schultz, N., et al. Integrated genomic characterization of endometrial carcinoma. Nature 497, 67-73 (2013).

[0147] TCGA. Comprehensive molecular portraits of human breast tumours. Nature 490, 61- 70 (2012).

[0148] Hoadley, K.A., et al. Cell-of-Origin Patterns Dominate the Molecular Classification of 10,000 Tumors. Cell 173, 291-304 (2018).

[0149] Wilkerson, M.D. & Hayes, D.N. ConsensusClusterPlus: a class discovery tool with confidence assessments and item tracking. Bioinformatics (Oxford, England) 26, 1572-1573 (2010).

[0150] Krzywinski, M., et al. Circos: an information aesthetic for comparative genomics. Genome Res 19, 1639-1645 (2009).

[0151] Tan, S.H., et al. AQP5 enriches for stem cells and cancer origins in the distal stomach. Nature 578, 437-443 (2020).

[0152] Barker, N., et al. Identification of stem cells in small intestine and colon by marker gene Lgr5. Nature 449, 1003-1007 (2007).

[0153] Tian, H., et al. A reserve stem cell population in small intestine renders Lgr5-positive cells dispensable. Nature 478, 255-259 (2011).

[0154] Lee, J.S., et al. Classification and prediction of survival in hepatocellular carcinoma by gene expression profiling. Hepatology 40, 667-676 (2004).

[0155] Lee, J.S., et al. A novel prognostic subtype of human hepatocellular carcinoma derived from hepatic progenitor cells. Nature Medicine 12, 410-416 (2006).

[0156] Kim, J.H., et al. Genomic predictors for recurrence patterns of hepatocellular carcinoma: model derivation and validation. PLoS Med 11, el001770 (2014).

[0157] Nagaoka, K., et al. Deep immunophenotyping at the single-cell level identifies a combination of anti-IL-17 and checkpoint blockade as an effective treatment in a preclinical model of data-guided personalized immunotherapy. Journal for immunotherapy of cancer 8, e001358 (2020).

[0158] van den Ende, T., et al. Neoadjuvant Chemoradiotherapy Combined with Atezolizumab for Resectable Esophageal Adenocarcinoma: A Single Arm Phase II Feasibility Trial (PERFECT). Clin Cancer Res 27, 3351-3359 (2021).

[0159] Park, S.H., et al. Phase III Trial to Compare Adjuvant Chemotherapy With Capecitabine and Cisplatin Versus Concurrent Chemoradiotherapy in Gastric Cancer: Final Report of the Adjuvant Chemoradiotherapy in Stomach Tumors Trial, Including Survival and Subset Analyses. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 33, 3130-3136 (2015).

[0160] Lee, J., et al. Phase III trial comparing capecitabine plus cisplatin versus capecitabine plus cisplatin with concurrent capecitabine radiotherapy in completely resected gastric cancer with D2 lymph node dissection: the ARTIST trial. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 30, 268-273 (2012).

[0161] Guo, S., et al. Molecular Pathology of Patient Tumors, Patient-Derived Xenografts, and Cancer Cell Lines. Cancer Res 76, 4619-4626 (2016).

[0162] Bladt, F., Friese-Hamim, M., Ihling, C., Wilm, C. & Blaukat, A. The c-Met Inhibitor MSC2156119J Effectively Inhibits Tumor Growth in Liver Cancer Models. Cancers 6, 1736- 1752 (2014).

[0163] Tibshirani, R., Hastie, T., Narasimhan, B. & Chu, G. Diagnosis of multiple cancer types by shrunken centroids of gene expression. Proc Natl Acad Sci U S A 99, 6567-6572 (2002).

[0164] Parker, J.S., et al. Supervised risk predictor of breast cancer based on intrinsic subtypes. Journal of clinical oncology : official journal of the American Society of Clinical Oncology 27, 1160-1167 (2009).

[0165] Nagafuchi, A., Shirayoshi, Y., Okazaki, K., Yasuda, K. & Takeichi, M. Transformation of cell adhesion properties by exogenously introduced E-cadherin cDNA. Nature 329, 341-343 (1987).

[0166] Yoshihara, K., et al. Inferring tumour purity and stromal and immune cell admixture from expression. Nat Commun 4, 2612 (2013).

[0167] Kang, Y.K., et al. Nivolumab in patients with advanced gastric or gastro-oesophageal junction cancer refractory to, or intolerant of, at least two previous chemotherapy regimens (ONO-4538-12, ATTRACTION-2): a randomised, double-blind, placebo-controlled, phase 3 trial. Lancet 390, 2461-2471 (2017).

[0168] Janjigian, Y.Y., et al. First-line nivolumab plus chemotherapy versus chemotherapy alone for advanced gastric, gastro-oesophageal junction, and oesophageal adenocarcinoma (CheckMate 649): a randomised, open-label, phase 3 trial. Lancet 398, 27-40 (2021).

[0169] Kelly, R.J., et al. Adjuvant Nivolumab in Resected Esophageal or Gastroesophageal Junction Cancer. The New England journal of medicine 384, 1191-1203 (2021).

[0170] Kim, S.T., et al. Comprehensive molecular characterization of clinical responses to PD-1 inhibition in metastatic gastric cancer. Nat Med 24, 1449-1458 (2018).

[0171] Kwon, M., et al. Determinants of Response and Intrinsic Resistance to PD-1 Blockade in Microsatellite Instability-High Gastric Cancer. Cancer discovery 11, 2168-2185 (2021).

[0172] Bang, Y.J., et al. Adjuvant capecitabine and oxaliplatin for gastric cancer after D2 gastrectomy (CLASSIC): a phase 3 open-label, randomised controlled trial. Lancet 379, 315- 321 (2012).

[0173] Cats, A., et al. Chemotherapy versus chemoradiotherapy after surgery and preoperative chemotherapy for resectable gastric cancer (CRITICS): an international, openlabel, randomised phase 3 trial. Lancet Oncol 19, 616-628 (2018).

[0174] Bang, Y.J., et al. Trastuzumab in combination with chemotherapy versus chemotherapy alone for treatment of HER2 -positive advanced gastric or gastro-oesophageal junction cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet 376, 687- 697 (2010).

[0175] Wong, G.S., et al. Targeting wild-type KRAS -amplified gastroesophageal cancer through combined MEK. Nat Med 24, 968-977 (2018).

[0176] Pollak, M. Insulin and insulin-like growth factor signalling in neoplasia. Nat Rev Cancer 8, 915-928 (2008).

[0177] Pollak, M. The insulin and insulin-like growth factor receptor family in neoplasia: an update. Nat Rev Cancer 12, 159-169 (2012).

[0178] Ghandi, M., et al. Next-generation characterization of the Cancer Cell Line Encyclopedia. Nature 569, 503-508 (2019).

[0179] Viswanathan, V.S., et al. Dependency of a therapy-resistant state of cancer cells on a lipid peroxidase pathway. Nature 547, 453-457 (2017).

[0180] Zhang, Q., et al. Atorvastatin Induces Mitochondria-Dependent Ferroptosis via the Modulation of Nrf2-xCT / GPx4 Axis. Front Cell Dev Biol 10, 806081 (2022).

[0181] Kim, B.M., Buchner, G., Miletich, I., Sharpe, P.T. & Shivdasani, R.A. The stomach mesenchymal transcription factor Barxl specifies gastric epithelial identity through inhibition of transient Wnt signaling. Dev. Cell 8, 611-622 (2005).

[0182] Perou, C.M., et al. Molecular portraits of human breast tumours. Nature 406, 747-752 (2000).

[0183] Davoli, T., Uno, H., Wooten, E.C. & Elledge, S.J. Tumor aneuploidy correlates with markers of immune evasion and with reduced response to immunotherapy. Science 355, eaaf8399 (2017).

[0184] Ock, C.Y., et al. Genomic landscape associated with potential response to anti-CTLA- 4 treatment in. Nat Commun 8, 1050 (2017).

[0185] Tatetsu, H., et al. SALL4, the missing link between stem cells, development and cancer. Gene 584, 111-119 (2016).

[0186] Donovan, K.A., et al. Thalidomide promotes degradation of SALL4, a transcription factor implicated in Duane Radial Ray syndrome. Elife 7, e38430 (2018).

[0187] Matyskiela, M.E., et al. SALL4 mediates teratogenicity as a thalidomide-dependent cereblon substrate. Nat Chem Biol 14, 981-987 (2018).Fan, C., et al. Cancer / testis antigens: from serology to mRNA cancer vaccine. Semin Cancer Biol 76, 218-231 (2021).

[0188] Although the disclosure has been described with reference to the presently preferred embodiment, it should be understood that various modifications can be made without departing from the spirit of the disclosure. Accordingly, the disclosure is limited only by the following claims.

Claims

What Is Claimed Is:

1. A method of classifying a genomic subtype of a gastric tumor comprising:(i) obtaining a digital gene expression profile of the gastric tumor;(ii) analyzing the digital gene expression profile using a first computer system for classifying the digital gene expression profile, wherein the first computer system classifies the genomic subtype of the gastric tumor as a CGS1 subtype tumor, a CGS2 subtype tumor, a CGS3 subtype tumor, a CGS4 subtype tumor, a CGS5 subtype tumor, or a CGS6 subtype tumor, thereby classifying the genomic subtype of the gastric tumor.

2. The method of claim 1, wherein the digital gene expression profile of the gastric tumor is compared to a reference digital gene expression profile of the genomic subtype of the gastric tumor or a reference database of digital gene expression profiles.

3. The method of claim 1, wherein the digital gene expression profile is obtained by normalizing gene expression data to one or more housekeeping genes using a second computer system.

4. The method of any one of claims 1-3, wherein the CGS1 subtype tumor comprises expression of DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and / or H19.

5. The method of any one of claims 1-3, wherein the CGS2 subtype tumor comprises expression of AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and / or SIOOP.

6. The method of any one of claims 1-3, wherein the CGS3 subtype tumor comprises expression of, OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 1 OB, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and / or CXCL9.

7. The method of any one of claims 1-3, wherein the CGS4 subtype tumor comprises expression of PC SKIN, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4,AP0A2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6 and / or SERPINH 1.

8. The method of any one of claims 1-3, wherein the CGS5 subtype tumor comprises expression of ZIC2, CXCL1, SERPINB5, METRN, H0XA9, CRYBG1, MYB, GSKIP, SPP1, N0X01, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and / or DOCK5.

9. The method of any one of claims 1-3, wherein the CGS6 subtype tumor comprises expression of GBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and / or CD247.

10. The method of any of claims 1-9, further comprising analyzing one or more of a digital proteomic expression profile, a copy number profile, a mutation profile, and / or methylation profile for the genomic subtype of the gastric tumor.

11. The method of claim 4, wherein the CGS 1 subtype tumor comprises a poor prognosis, a high stem cell characteristic, a high IGF1 expression, and / or a low level of genomic alteration.

12. The method of claim 5, wherein the CGS2 subtype tumor comprises canonical epithelial gene expression.

13. The method of claim 6, wherein the CGS3 subtype tumor comprises a high copy number alteration, a low immune reactivity, and / or HER2 activation.

14. The method of claim 7, wherein the CGS4 subtype tumor comprises a high copy number alteration, a low immune reactivity, SALL4 expression and / or KRAS activation.

15. The method of claim 8, wherein the CGS5 subtype tumor comprises a high mutation burden, moderately high immune reactivity and / or microsatellite instability.

16. The method of claim 9, wherein the CGS6 subtype tumor comprises a high level of methylation, a high immune reactivity and / or is positive for Epstein Barr virus.

17. The method of any one claims 1-16, wherein(i) the CGS1 subtype tumor comprises a stem score of about 0.8;(ii) the CGS2 subtype tumor comprises a stem score of about 0.4 to 0.5;(iii) the CGS3 subtype tumor comprises a stem score of about 0.2 to 0.3;(iv) the CGS4 subtype tumor comprises a stem score of about 0.4;(v) the CGS5 subtype tumor comprises a stem score of about 0.1 to 0.2; and / or(vi) the CGS6 subtype tumor comprises a stem score of about 0.2 to 0.4.

18. The method of any one of claims 1-16, wherein(i) the CGS1 subtype tumor comprises a CD8TRS score of about 0.4 to 0.6;(ii) the CGS2 subtype tumor comprises a CD8TRS score of about 0.2;(iii) the CGS3 subtype tumor comprises a CD8TRS score of about 0.2;(iv) the CGS4 subtype tumor comprises a CD8TRS score of about 0.3;(v) the CGS5 subtype tumor comprises a CD8TRS score of about 0.3 to 0.4; and / or(vi) the CGS6 subtype tumor comprises a CD8TRS score of about 0.6.

19. The method of any one of claims 1-16, wherein(i) the CGS1 subtype tumor comprises an anti-PDLl response score of about 0.4 to 0.5;(ii) the CGS2 subtype tumor comprises an anti-PDLl response score of about 0.3;(iii) the CGS3 subtype tumor comprises an anti-PDLl response score of about 0.3;(iv) the CGS4 subtype tumor comprises an anti-PDLl response score of about 0.4;(v) the CGS5 subtype tumor comprises an anti-PDLl response score of about 0.3 to 0.4; and / or(vi) the CGS6 subtype tumor comprises an anti-PDLl response score of about 0.3 to 0.4.

20. A method of selecting a therapeutic modality for treating a patient with a gastric tumor, comprising:(i) classifying the genomic subtype of the gastric tumor according to any one of claims 1-19; and(ii) selecting a therapeutic modality based on the genomic subtype of the gastric tumor.

21. The method of claim 20, comprising treating the patient with the CGS3 subtype tumor with a radiation therapy, a ferroptosis inducer, a HER2 targeting therapy or a combination thereof.

22. The method of claim 21 , wherein the HER2 targeting therapy is selected from neratinib, trastuzumab, fam-trastuzumab deruxtecan, hyaluronidase, dacomitinib, lapatinib, tucatinib, pertuzumab, margetuximab, ado-trastuzumab emtansine or a combination thereof.

23. The method of claim 21, wherein the ferroptosis inducer is selected from erastin, RSL5, DIDS, TMZ, MMRi62, Sorafenib, RSL3, ML162, DPI7, DPI10, FIN56, BSO, MEII, PE, AE, SAS, NDP4928, FIN56, Statin, Brequinar, Dexamethasone or a combination thereof.

24. The method of claim 20, comprising treating the patient with the CGS 1 subtype tumor, the CGS4 subtype tumor, the CGS5 subtype tumor, or the CGS6 subtype tumor with an immunotherapy.

25. The method of claim 24, wherein the immunotherapy is selected from a monoclonal antibody, a CAR T cell therapy, a checkpoint inhibitor, a cancer vaccine, a cytokine, an immune system modulator or a combination thereof.

26. The method of claim 25, wherein the checkpoint inhibitor is selected from atezolizumab, ipilimumab, pembrolizumab, nivolumab or a combination thereof.

27. The method of claim 25, wherein the cytokine is selected from interleukin-2 (IL-2), interferons-alpha (IFN-alpha), interleukin- 12 (IL-12), interleukin- 15 (IL-15), or a combination thereof.

28. The method of claim 20, comprising treating the patient with the CGS1 subtype tumor with an IGF1R targeting therapy.

29. The method of claim 28, wherein the IGF1R targeting therapy is selected from ganitumab, xentuzumab, teprotumumab, teprotumumab, Linsitinib (OSI-906), BMS- 754807, Picropodophyllin, NT157, an IGF-1R antisense oligonucleotide, an IGF1R small interfering RNA (siRNA) or a combination thereof.

30. The method of claim 20, comprising treating the patient with the CGS4 subtype tumor with a SALL4 or KRAS targeting therapy.

31. The method of claim 30, wherein the KRAS targeting therapy is selected from AMG510 (sotorasib), MRTX849 (adagrasib), ARS-3248, LY3499446, D-1553, JDQ443, GDC-6036 or a combination thereof.

32. The method of claim 30, wherein the SALL4 targeting therapy is selected from a SALL4 antisense oligonucleotide, a SALL4 small interfering RNA (siRNA), thalidomide, or a combination thereof.

33. The method of claim 20, comprising treating the patient with the CGS6 subtype tumor with an antiviral agent or a PI3K inhibitor.

34. The method of claim 33, wherein the antiviral agent is selected from acyclovir, desciclovir, ganciclovir, adenine arabinoside, phosphonoacetic acid or a combination thereof.

35. The method of claim 33, wherein the PI3K inhibitor is Copanlisib, Idelalisib, Umbralisib, Duvelisib, Alpelisib or a combination thereof.

36. A kit for classifying a genomic subtype of a gastric tumor comprising a probe or a primer for one or more genes associated with a gastric tumor subtype, wherein the one or more genes are selected from:DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and H19 associated with a CGS1 gastric tumor subtype;AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and SIOOP associated with a CGS2 subtype tumor;OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 10B, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and CXCL9 associated with a CGS3 subtype tumor;PC SKIN, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and SERPINH1 associated with a CGS4 subtype tumor;ZIC2, CXCL1, SERPINB5, METRN, H0XA9, CRYBG1, MYB, GSKIP, SPP1, N0X01, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and DOCK5 associated with a CGS5 subtype tumor; and / orGBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and CD247 associated with a CGS6 subtype tumor.

37. A kit for determining a therapeutic modality for treating a patient with a gastric tumor comprising a probe or a primer for one or more genes associated with a gastric tumor subtype, wherein the one or more genes are selected from:DES, MYH11, CNN1, ECRG4, TAGLN, MYLK, SPARCL1, SFRP1, BARX1, SCRG1, IGF1, FRZB, CXCL12, CAV1, FOXF1, VIM, PBX3, CYR61, CTGF, and H19 associated with a CGS1 gastric tumor subtype;AKR7A3, GKN1, PGC, MUC5AC, PSCA, TFF1, DPCR1, GKN2, LIPF, VSIG1, MUC6, MSMB, ANXA10, PGA3, CA2, CLDN18, REGIA, MT1G, SST, and SIOOP associated with a CGS2 subtype tumor;OLFM4, MUC17, CEACAM6, LCN2, LGALS4, GUCY2C, ATP 1 OB, SLC22A18, TMEM45B, HKDC1, GZMH, GBP5, CD8A, HCST, GZMB, CD52, CXCL10, CCL5, NKG7 and CXCL9 associated with a CGS3 subtype tumor;PCSK1N, QPRT, PEG10, MAGEA3, RBP4, CDKN2A, SALL4, APOA2, GGH, GLDC, MAGEA12, CHST13, ARID3A, APOH, DNMT3B, FGB, TF, SYNE4, HOXC6, and SERPINH1 associated with a CGS4 subtype tumor;ZIC2, CXCL1, SERPINB5, METRN, H0XA9, CRYBG1, MYB, GSKIP, SPP1, N0X01, PKP2, KRTCAP3, TFAP2A, SGPP2, MSLN, CLDN7, ETHE1, TNFRSF10A, IL20RA, and DOCK5 associated with a CGS5 subtype tumor; and / orGBP4, HLA-DOA, HLA-DPA1, GZMA, CX3CL1, CD40, CXCR6, CD3D, PTPRCAP, ADAMDEC1, APOCI, TGM2, RARRES3, CD2, CYBB, IGSF6, LAMP3, IL27RA, RAC2, and CD247 associated with a CGS6 subtype tumor; and selecting a therapeutic modality based on subtype.

Images