Application of T-beta+CD8 + T cell as marker in preparation of medicine for treating gastrointestinal tumors

Abstract

The invention discloses application of T-bet + CD8 + T cells as markers in preparation of drugs for treating gastrointestinal tumors. Specifically, the content level of T-bet + CD8 + T cells in a sample is measured, and the treatment effectiveness of the digestive tract tumor PD-1 signal inhibitor (ICI) is judged according to whether T-bet + CD8 + T cell expression is positive or not. According to the method disclosed by the invention, microsatellite stable gastric cancer (MSS GC) patients are layered into MSS potential responders (MSS-PRs) and MSS non-responders (MSS-NRs) through positive-unmarked (PUL) semi-supervised machine learning, so that the long-standing clinical problem that the response of MSS GC to ICI is not accurately predicted is challenged. It is identified that infiltration of T-bet + CD8 + T cells in a tumor area is the key for determining that ICI treatment of MSS-PRs patients is effective, and it is proved that adoptive treatment of the T-bet + CD8 + T cells can enable tumors of immune deserts to be sensitive to ICI. The invention shows how advanced machine learning is effectively applied to medical research, and solves the clinical problem which originally has challenging property from two dimensions of accurately screening ICI treatment benefit people in MSS GC and improving immune desert MSS GC immune microenvironment and expanding ICI treatment benefit people.

Description

Technical Field

[0001] The present invention relates to the field of tumor immunotherapy, and in particular to the application of T-bet+CD8+T cells as markers in the preparation of drugs for treating digestive tract tumors. Background Art

[0002] Gastric cancer is one of the leading causes of cancer-related deaths worldwide. Chemotherapy remains the first-line treatment for most advanced GC, but the median overall survival (OS) has not exceeded 1 year. Immune checkpoint inhibitors (ICI, such as PD-1 / PD-L1 inhibitors) have shown great therapeutic promise in a variety of solid tumors, including gastric cancer. Unlike traditional chemotherapy that directly kills cancer cells, ICIs exert their anti-tumor effects by relieving T cell exhaustion and restoring T cell function. However, the efficacy of ICI in unscreened patient populations is only about 20%. There is an urgent need to identify efficacy predictive biomarkers to screen the superior treatment population and find effective strategies to enable more patients to benefit from ICI treatment.

[0003] Microsatellite instability-high / deficient mismatch repair (MSI-H / dMMR) has been approved as a biomarker for predicting the efficacy of ICI in gastric cancer. PD-1 inhibitors have shown good antitumor activity in MSI-H / dMMR GC, with an overall response rate (ORR) of 45% to 60%. However, MSI-H GC only accounts for 10% to 15% of all GC, and is reduced to 5% in advanced cases, which severely limits the clinical benefit of ICIs in GC. Recent clinical trials have shown that the ORR of microsatellite stable / complete mismatch repair (MSS / pMMR) GC is approximately 10%, but these patients are difficult to identify before treatment. Given the large number of MSS GC, we urgently need reliable biomarkers to preselect MSS patients who respond to ICI.

[0004] Researchers have been working to identify biomarkers predictive of immunotherapy efficacy in MSS GC. In the KEYNOTE-062 study, higher tumor mutation burden (TMB) was associated with better clinical response to ICI, but its predictive value decreased after excluding MSI-H patients. In KEYNOTE-059, the objective response rate (ORR) of unstratified MSS GC increased from 9% to 17.4% in patients with a PD-L1 composite positive score (CPS) ≥ 10 points. However, the calculation of CPS is affected by subjective factors of pathologists, which seriously limits the clinical application of CPS. A retrospective analysis of 61 patients with advanced gastric cancer treated with PD-1 antibodies showed that the ORR of Epstein-Barr virus (EBV) + MSS gastric cancer patients was 100%. This result has attracted global attention to the role of EBV status in ICI response. However, subsequent studies have shown that the ORR of EBV + MSS GC is only 25% to 33%, and its accuracy in predicting ICI response is still unsatisfactory. A fundamental shortcoming of current biomarkers is that they focus only on tumor cell-intrinsic factors of GC but largely ignore the complex nature of the tumor microenvironment (TME).

[0005] The TME is a complex ecosystem that is interwoven with numerous cellular and non-cellular components and interacts intricately. The rapid development of high-resolution technologies (multiple immunohistochemistry, spatial transcriptome sequencing, and single-cell transcriptome sequencing, etc.) and machine learning have promoted the study of the complexity of the TME. Tumor infiltrating lymphocytes (TILs) are the main anti-tumor cells in the TME and are also direct targets for ICI therapy. Understanding the functional heterogeneity of TILs can help us understand the determinants of various ICI responses in cancer, and most importantly, can provide us with potential strategies to transform immune cells to improve the efficacy of tumor immunotherapy.

[0006] In this study, we used semi-supervised machine learning to identify the TME of potential responders (MSS-PRs) to ICI therapy in MSS GC. Comprehensive evaluation of the TME of MSS-PRs showed that T-bet+CD8+T cells had superior anti-tumor ability. T-bet+CD8+T cells chemotaxis to the tumor core through CXCL9 / 10 / 11 secreted by tumor cells, forming a niche environment with high levels of interferon gamma (IFN-γ) and high expression of interferon-stimulated gene PD-L1; the positive feedback loop formed between CXCR3 ligands (CXCL9 / 10 / 11, etc.)+tumor cells and T-bet+CD8+T cells constitutes an immune microenvironment to drive the response of MSS GCs to ICI. Furthermore, adoptive therapy of T-bet+CD8+T cells can sensitize tumors in immune deserts to ICI. Therefore, this study solved the current bottleneck problem of immunotherapy for MSS GC from two dimensions: accurately screening the population that can benefit from ICI treatment in MSS GC and improving the immune microenvironment of immune-desert MSS GC to expand the population that can benefit from ICI treatment, which has great clinical significance. Summary of the invention

[0007] The present invention aims to provide the use of T-bet+CD8+T cells as a marker in the preparation of a drug for treating digestive tract tumors. In particular, T-bet+CD8+T cells are used as a biomarker to determine the efficacy of immunotherapy for digestive tract tumors.

[0008] In this study, we stratified MSS GC patients into MSS-PRs (MSS potential responders) and MSS-NRs (MSS non-responders) by semi-supervised machine learning of positive-unmarker (PUL), thus challenging the long-standing clinical problem of the lack of accurate prediction of MSS GC response to ICI. PUL predicted ICI response in MSS GC with an AUC of 0.940, which is the most accurate model currently. We identified 16.20% of MSS GC in the TCGA-STAD cohort and 10.25% of MSS-GC in the ACRG cohort as potential responders, which is consistent with published data that approximately 10% to 15% of MSS GC respond to ICI. In this study, we demonstrated how advanced machine learning can be effectively applied in medical research to solve otherwise challenging clinical problems.

[0009] The first purpose of the present invention is to provide a use of T-bet+CD8+T cells as a marker in the preparation of drugs for treating digestive tract tumors.

[0010] Furthermore, the level of T-bet+CD8+T cells in the sample was measured, and the therapeutic effectiveness of the PD-1 signal inhibitor for gastrointestinal tumors was determined based on whether the T-bet+CD8+T cell expression was positive.

[0011] Furthermore, the sample is selected from any one or more of peripheral blood, paracancerous tissue, and cancerous tissue.

[0012] Furthermore, the PD-1 signal inhibitor is selected from any one or more of anti-PD-1 antibodies, anti-PD-L1 antibodies and anti-PD-L2 antibodies.

[0013] Furthermore, the digestive tract tumor is selected from any one of gastric cancer, intestinal cancer, esophageal cancer, pancreatic cancer, liver cancer, and bile duct cancer.

[0014] Furthermore, the gastric cancer is microsatellite stable gastric cancer (MSS GC).

[0015] Furthermore, the T-bet+CD8+T cells were highly infiltrated in MSS GC.

[0016] Furthermore, the T-bet+CD8+T cells are chemotactic to the tumor core through the CXCL9 / 10 / 11-CXCR3 axis and exert anti-tumor activity by secreting IFN-γ, granzyme B and other cytotoxic molecules.

[0017] Furthermore, the abundance of T-bet+CD8+T cells was correlated with ICI response, and the AUC for predicting ICI response in MSS patients based on the cell fraction of tumor-infiltrating T-bet+CD8+T cells was 0.800.

[0018] Furthermore, the T-bet+CD8+T cells enhance anti-tumor immunity and confer sensitivity of immune-desert gastric cancer to ICI.

[0019] Furthermore, the T-bet+CD8+T cells are chemotactic to the tumor core through CXCL9 / 10 / 11 secreted by tumor cells, forming an ecological niche environment with high interferon γ (IFN-γ) and interferon-stimulated genes (ISG, such as CXCL9 / 10 / 11, PD-L1). There is a positive feedback loop between CXCR3 ligands (CXCL9 / 10 / 11, CCL3 / 5, etc.)+tumor cells and T-bet+CD8+T cells, which manifests as the T-bet+-PD-L1+ niche environment driving the sensitivity of ICI.

[0020] Furthermore, the niche environment is characterized by the coexistence of T-bet+T cells and PD-L1+ cells within a range of 50 μm.

[0021] Furthermore, the application specifically includes:

[0022] 1) Determine the level of T-bet+CD8+T cells in the sample to determine whether T-bet+CD8+T cell expression is positive;

[0023] 2) If the sample is positive for T-bet+CD8+T cell expression, proceed to step 3);

[0024] Or, if the expression of T-bet+CD8+T cells in the sample is negative, peripheral blood mononuclear cells (PBMCs) are cultured and activated in vitro to induce the expression of T-bet in CD8+T cells, and T-bet+CD8+T cells are obtained and then infused in vivo until the expression of T-bet+CD8+T cells is positive;

[0025] 3) detecting the abundance of the T-bet+CD8+T cells;

[0026] 4) When the abundance of the T-bet+CD8+T cells reaches a preset value, the T-bet+CD8+T cells are used in combination with a PD-1 inhibitor.

[0027] Furthermore, the step of inducing T-bet+CD8+T cells comprises:

[0028] (a) T cells were activated using culture plates coated with CD3 / CD28 antibodies for 48 hours;

[0029] (b) Add 100 IU / ml IL-2 to rapidly expand T cells;

[0030] (c) 72 hours after withdrawal of CD3 / CD28 antibodies, 50 ng / ml IL-12 was added to induce T-bet expression in CD8+ T cells.

[0031] Furthermore, the dosage of the T-bet+CD8+T cells is 1×10 6 -1×10 8 Cells / time.

[0032] Furthermore, the T-bet+CD8+T cells are used 1-3 times per week.

[0033] Furthermore, the T-bet+CD8+T cells are infused by intravenous injection or local injection.

[0034] Furthermore, the combined use of T-bet+CD8+T cells and PD-1 inhibitors is to add the PD-1 inhibitors within 24 hours after T-bet+CD8+T cell infusion.

[0035] Furthermore, the abundance of T-bet+CD8+T cells in the step 4) reaches a preset value, which means that the proportion of T-bet+CD8+T cells in all stromal cells is >0.1%; if the proportion of T-bet+CD8+T cells in all stromal cells is <0.1%, the process returns to step 2), and peripheral blood mononuclear cells PBMCs are cultured and activated in vitro to induce the expression of T-bet in CD8+T cells, and after obtaining T-bet+CD8+T cells, they are infused in vivo, and the abundance of T-bet+CD8+T cells is detected again.

[0036] Furthermore, when the abundance of the T-bet+CD8+T cells reaches a preset value and a T-bet+-PD-L1+ niche environment is generated, the T-bet+CD8+T cells are used in combination with a PD-1 inhibitor.

[0037] Furthermore, the niche environment is characterized by the coexistence of T-bet+T cells and PD-L1+ cells within a range of 50 μm.

[0038] Furthermore, the abundance detection method of T-bet+CD8+T cells and PD-L1+ cells includes immunohistochemistry (IHC) and multiplex immunohistochemistry (mIHC).

[0039] The second object of the present invention is to provide a pharmaceutical composition for treating gastrointestinal tumors, which pharmaceutical composition at least contains T-bet+CD8+T cells and a PD-1 signal inhibitor.

[0040] Furthermore, the content ratio of T-bet+CD8+T cells and PD-1 signal inhibitor is 5×10 6 Piece: 200μg.

[0041] We used the TCGA-STAD cohort and ACRG cohort to screen out a series of genes that were significantly elevated in MSS-PRs patients, including CXCR3 ligands (CXCL9, CXCL10, and CXCL11), IFNG, interferon-stimulated genes CD274 / PD-L1, and cytotoxic molecules GZMB. We were pleasantly surprised to find that genes such as CXCR3, IFN-γ, and granzyme B in CD8+T cells have a common transcription factor, T-bet. Since CXCL9 / 10 / 11 expressed by tumor cells can chemotactically attract CXCR3+T cells to infiltrate the tumor area, and CXCL9 / 10 / 11 can be upregulated in an interferon-γ (IFN-γ)-dependent manner; the lack of IFN-γ signaling pathways cannot respond to ICI, and IFN-γ is a key factor in upregulating PD-L1. Therefore, we speculate that the infiltration of T-bet+CD8+T cells in the tumor area and the expression of PD-L1 induced by IFN-γ may be the key to the sensitivity of MSS-PRs patients to ICI. Our data showed that T-bet+CD8+T cells can produce a large amount of cytotoxic granzymes and IFN-γ. The T-bet+-PD-L1+ niche can be seen in the tumor microenvironment of MSS-PRs patients and ICI-sensitive patients, which is consistent with the speculation. So far, we have proved that there is a potential relationship between the CXCR3 chemotactic pair and the interferon signaling pathway. These signaling pathways are coupled with T-bet+CD8+T cells, and finally form an ICI-sensitive TME: T-bet+CD8+T cells are chemotactic to the tumor core through chemokines CXCL9 / 10 / 11, forming a high IFN-γ and ISG (interferon-stimulated genes, such as CXCL9 / 10 / 11, PD-L1, etc.) niche environment. The T-bet+-PD-L1+ niche formed by the positive feedback loop between CXCR3 ligand+tumor cells and T-bet+T cells is the decisive factor for ICI treatment sensitivity.

[0042] Although the CPS score of PD-L1 is currently the most commonly used biomarker for predicting the efficacy of ICI, the accuracy of CPS is still relatively low. Our results strongly suggest that the predictive value of inducible PD-L1 for tumor cells is superior to that of constitutive PD-L1, emphasizing the importance of the IFN-γ signaling pathway. Therefore, our mechanism-based suggestion is to use the T-bet+-PD-L1+ niche as an evaluation indicator of ICI sensitivity. This means that the tumor itself has sufficient IFN-γ signaling pathways, the drug target of ICI PD-L1, and abundant tumor-infiltrating cytotoxic T cells. Therefore, this suggestion is more scientific and intuitive than the existing CPS for evaluating PD-L1, and may also be more accurate.

[0043] Given that more than 90% of advanced gastric cancer (GC) is microsatellite stable (MSS), the efficacy of ICIs in unstratified MSS GC is only about 10%. In this study, semi-supervised learning was applied to stratify potential responders to ICIs in MSS-GC, with an area under the curve of 0.924. Spatial analysis of the tumor microenvironment (TME) showed that T-bet+CD8+T cells were highly infiltrated in the tumor area of ​​GCs sensitive to ICIs; and the ability of T-bet+CD8+T cells to secrete cytotoxic molecules was enhanced. Therefore, T-bet+CD8+T cells infiltrating into the tumor area can exhibit stronger antitumor activity. Adoptive therapy of T-bet+CD8+T cells can enhance the antitumor immunity of immune-deserted MSS GC and confer susceptibility to ICIs. Spatial RNA sequencing showed that there is a positive feedback loop between T-bet+T cells and PD-L1+tumor cells, which ultimately drives T cell exhaustion and can therefore be treated with ICIs. Our study provides insights into the potential mechanisms of antitumor immunity.

[0044] The above stratification of MSS-PRs gastric cancer patients can significantly expand the population that can benefit from ICI treatment. However, since the immune microenvironment of most patients with MSS-type advanced gastric cancer is in an immune desert state, we proposed the adoptive T cell therapy (ACT) strategy to solve this problem. At present, it is difficult for adoptive T cells injected into the body to transport and infiltrate into the core of the tumor; even if they infiltrate into the core area of ​​the tumor, they become inhibitory TME due to rapid dysfunction. The above reasons greatly limit the application of ACT in solid tumors. Our previous results suggest that T-bet+CD8+T cells may be able to bypass these obstacles by enhancing tumor infiltration ability and forming a local immune active microenvironment. In a humanized gastric cancer mouse model, we demonstrated that the infusion of T-bet+CD8+T cells can enhance anti-tumor immunity and sensitize tumors in the immune desert to ICI treatment. In this study, we provide preclinical evidence that the combination of T-bet+CD8+T cell adoptive therapy and PD-1 inhibitors can improve the clinical outcomes of MSS GC in the immune desert.

[0045] Given the lack of immunotherapy response in MSS GC, it may be due to the lack of reliable biomarkers to preselect candidate ICIs or the immune desert nature of most GC cases. Therefore, we addressed this issue in two ways. First, we comprehensively characterized the immune characteristics of potential MSS GC responders. Spatial TME features such as the abundance of T-bet+CD8+ T cells and the presence of T-bet+-PD-L1+ niches can be used to stratify ICI treatment for MSS GC. Second, we demonstrated that adoptive T-bet+CD8+ T cell therapy can enhance the antitumor activity of immune desert MSS GC and sensitize them to ICIs. In summary, our study provides new insights into the fundamental determinants of ICI response in MSS GC and provides a remedy for optimizing immunotherapy of immune desert tumors. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 The immune landscape of different gastric cancer (GC) subtypes;

[0047] a. Violin plots of immune infiltration in EBV+MSS (n=26), MSl-H (n=45), and EBV-MSS (n=190) patients in the TCGA-STAD cohort estimated by cibersortX. The statistical significance between subtypes was evaluated by two-sided Mann-Whitney U test, *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001;

[0048] b. Multiple immunohistochemistry images (mIHC) of MSI-H GC patients (GC-44), immune-deserted MSS GC patients (GC-60), and immune-infiltrated MSS GC patients (GC-66). Red represents CD4+T cells, yellow represents CD8+T cells, cyan represents tumor cells, and DAPI staining represents cell nuclei;

[0049] c. UMAP image of 118,887 cells from 24 gastric cancer tissues (GSE183904). Endo is the abbreviation of endothelial cells, DC is dendritic cells, and epi is epithelial cells;

[0050] d. Stacked bar chart of cell components of MS1-H GC (n=6) and MSS GC (n=18). Each color represents a cell type, such as Figure 1 As shown in c;

[0051] e. Volcano plot of differentially expressed genes in CD8+T cells between MSl-H GCs and MSS GCs. The y-axis represents the adjusted P value, and the x-axis represents the log2-fold change. Genes with |log2-fold change|≥1.5 and P value ≤1×e-5 are indicated in red. P values ​​were calculated using DESeq2, and multiple comparisons were adjusted using the Benjamini-Hochberg method;

[0052] f. The receiver operating characteristic (ROC) curve of the PRJEB25780 cohort (validation dataset) shows the ratio of the false positive rate to the true positive rate for predicting ICI responders and non-responders in MSS patients based on the PUL score (AUC = 0.924);

[0053] g. T-SNE plots of patients with MSl-H & TMB-H & TILs-H (n = 40), MSS-PR (n = 46), and MSS-NR (n = 238) in the TCGA-STAD cohort. The positional proximity between the data points in the left figure indicates their similarity in immune characteristics. The proximity of the MSI-H & TMB-H & TILs-H subtype to the MSS-PR subtype was observed. The right figure is a heat map of PUL scores in patients with MSl-H & TMB-H & TILs-H (n = 40), MSS-PR (n = 46), and MSS-NR (n = 238) in the TCGA-STAD cohort.

[0054] Figure 2 is the comprehensive feature of MSS-PRs; among them,

[0055] a. Violin plots of immune infiltration in patients with MSI-H & TMB-H & TILs-H (n = 31), MSl-PR (n = 49), and MSS-NR (n = 167) in the TCGA-STAD cohort, estimated by cibersortX. The statistical significance between subtypes was evaluated by two-sided Mann-Whitnev U test, *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001;

[0056] b. Differentially expressed genes of MSS-PRs and MSS-NRs in TCGA-STAD and ACRG cohorts. The y-axis represents the fold change after merging the TCGA-STAD group and the ACRG group;

[0057] c. ROC curve of the top 8 upregulated genes in MSS-PRs predicting ICI outcomes in MSS patients in the PRJEB25780 cohort (average AUC = 0.85 ± 0.08).

[0058] Figure 3T-bet+T cells showed superior anti-tumor immunity; among them,

[0059] a. Flow cytometry analysis of the expression of IFN-γ, GZMB, and CXCR3 in T-bet+CD8+T cells (n=1675) and T-bet-CD8+T cells (n=12852);

[0060] b. Flow cytometry analysis of the expression of IFN-γ, GZMB, and CXCR3 in T-bet+CD4+T cells (n=1,778) and T-bet-CD4+T cells (n=19,704);

[0061] c. T-bet+CD4+T cells, T-bet+CD8+T cells, and T-bet+CXCR3+ cells exist in the GC microenvironment. mlHC images were taken from GC patients in the GC cohort;

[0062] d. Schematic diagram of downstream genes of transcription factor T-bet. In T cells, GZMB, PRF1, IFNG and CXCR3 are genes directly transcribed by T-bet;

[0063] e. Line graphs of the average percentages of CD4+T cells, CD8+T cells, T-bet+CD4+T cells, and T-bet+CD8+T cells in normal tissue (N), normal tumor margin (M), tumor margin (E), and tumor core (C) (N=19, 1 MSI-H GC, 7 EBV+MSS GC, 11 EBV-MSS GC). Gray edges indicate standard errors. The original line graphs were drawn using the interp1d function in SciPy (v1.8.1);

[0064] f. Schematic diagram of constructing a tumor-bearing humanized mouse model;

[0065] g. Line graph of tumor volume changes over time (P = 2.197e-04, two-way two-way ANOVA, with time and group as two independent variables). Tumor volume was calculated as 1 / 2×(Length×Width 2 ). The right side shows the infusion of T-bet high T cells or T-bet low Box plot of tumor weight of mice in the T cell group at the endpoint (P = 0.0292, two-sided paired t-test, *P < 0.05, ***P < 0.001);

[0066] h. Infusion of T-bet high T cells or T-bet lowmIHC of tumor-infiltrating T-bet+CD4+T cells and T-bet+CD8+T cells in humanized mice with T cells.

[0067] i. Compare T-bet infusion high T cells or T-bet low Box plot of the percentage of T-bet+CD8+T cells (P=2.434e-03, two-sided paired t-test) and T-bet+CD4+T cells (P=9.848e-03, two-sided paired t-test) in tumor-infiltrating immune cells of mice with T cells. Original data generated by flow cytometry analysis, **P<0.01;

[0068] j. Box plots of geometric mean fluorophore intensities (gMFIs) of IFN-γ and GZMB in GC tissue-infiltrating T-bet+CD8+T cells and T-bet-CD8+T cells from 10 gastric cancer CDX models (IFN-γ, P = 7.685e-04; GZMB, P = 2.461e-04), ***P < 0.001;

[0069] k. Kaplan-Meier curves of the relationship between overall survival and T-bet+CD8+T cell abundance (left) and T-bet+CD4+T cell abundance (right) in the GC cohort. The percentage was calculated as the number of T-bet+CD8+ or CD4+T cells divided by the number of all cells in the respective patient. 0.2% was used as the cutoff value for distinguishing high and low infiltration of T-bet+CD8+ and T-bet+CD4+T cells.

[0070] Figure 4 Tumor-infiltrating T-bet+CD8+T cells can predict the response of GC patients to ICIs; among them,

[0071] a. CD8 / CD4 ratio before PBMC infusion (n=5), the proportion of CD8+T cells is 0.36 times that of CD4+T cells (P=0.008, bilateral Mann-Whitney U test). The right figure shows the CD8 / CD4 ratio in GC TME after infusion (n=5), the proportion of T-bet+CD8+T cells is 1.67 times that of T-bet+CD4+T cells (P=0.016, bilateral Mann-Whitney U test). Data were obtained by flow cytometry, *P<0.05, **P<0.01;

[0072] b. UMAP diagram of TBX21 encoding T-bet expressed in different cell types in GC tumors;

[0073] c. Bar graph of the number of different types of T-bet+ cells in GC TME. In the TME of 24 GC patients, CD8+ T cells were the majority of T-bet+ cells;

[0074] d. The process of tissue segmentation using inFrom software (v 2.3.0). The red area represents the tumor area, and the green area represents the stromal area. Tissue segmentation was performed based on the expression pattern of pan-CK;

[0075] e. Scoring diagram of tumor and stromal regions. Cells with green nuclei and yellow membranes are T-bet+CD8+T cells; cells with green nuclei and red membranes are T-bet+CD4+T cells;

[0076] f. The proportion of T-bet+CD8+T cells and T-bet+CD4+T cells in the interstitial area and tumor area of ​​the GC cohort (n=120). The percentage was calculated by dividing the non-tumor cells in each area respectively. In the interstitial area, there was no significant difference in the proportion of T-bet+CD8+T cells and T-bet+CD4+T cells (0.21% vs 0.19%, P=0.068, bilateral Mann-Whitney U test). In the tumor area, T-bet+CD8+T cells were more abundant than T-bet+CD4+T cells (0.25% vs. 0.14%, P=0.0018, bilateral Mann-Whitney U test, **P<0.01);

[0077] g. mIHC of TME of ICI responders (ICl-12), ICI non-responders (ICl-7), MSS-PR patients (GC-97), and MSS-NR patients (GC-24). As shown by the white arrows, T-bet+CD8+T cells infiltrated into the tumor areas of ICI responders and MSS-PR patients in large numbers;

[0078] h. ROC curve for predicting ICI response in MSS patients based on the cell fraction of tumor-infiltrating T-bet+CD8+T cells, estimated by CibersortX (AUC=0.800). On the right is a box plot of tumor-infiltrating T-bet+CD8+T cells in MSS responders and MSS non-responders in the PRJEB25870 cohort (P=0.010, two-sided Mann-Whitney U test, *P<0.05). The mean T-bet+CD8+T fraction in MSS responders was 0.086.

[0079] Figure 5 T-bet+CD8+T cell infusion confers ICls susceptibility in immune desert GCs;

[0080] a. Schematic diagram of constructing the immune desert gastric cancer tumor-bearing model in humanized mice. The detailed information is described in the methods section;

[0081] b. Line graph of tumor size over time (T-bet low +IgG4 vs T-bet high +IgG4, P = 0.012; T-bet low +anti-PD-1 vs T-bet high +anti-PD-1,P=1.191e-07;T-bet low +lgG4 vs T-bet low +anti-PD-1,P=3.099e-08;T-bet high +lgG4 vs T-bet high +anti-PD-1, P = 2.201e-06. All statistical significances were tested by two-sided two-way ANOVA). *P < 0.05, ****P < 0.0001;

[0082] c. mIHC of TME in humanized mouse models of different groups. From left to right are T-bet low +IgG4、T-bet low +anti-PD-1, T-bet high +IgG4、T-bet high +anti-PD-1 group;

[0083] Box plot of the percentage of dT-bet+CD8+T cells (P=0.040, two-sided paired t-test) or PD-1+CD8+T cells (P=0.026, two-sided paired t-test) in tumor-infiltrating immune cells. Data were generated by flow cytometry analysis, *P<0.05;

[0084] eT-bet high mIHC images of mice treated with CD8+T cells and anti-PD-1 monoclonal antibodies. On the left are DAPI, CD4 CD8, PD-1, PD-

[0085] Composite image of L1 and T-bet. Only DAPI, PD-L1 and T-bet are shown on the right. T-bet+-PD-L1+ niche can be observed.

[0086] Figure 6 The positive feedback loop between T-bet+T cells and PD-L1+ tumor cells drives the sensitivity of ICIs;

[0087] aT-bet+-PD-L1+ niche exists in patients with abundant T-bet+CD8+T cell infiltration (GC-44);

[0088] bSchematic diagram of neighboring cells of T-bet-CD8+T cells and T-bet+CD8+T cells. Neighboring cells are defined as cells that are less than 50 μm away from the original cells. Orange is PD-L1+ cells, and blue is PD-L1- cells;

[0089] c. Bar graph of the percentage of PD-L1+ cells in cells adjacent to T-bet+CD8+ T cells, T-bet-CD8+ T cells, T-bet+CD4+ T cells, and T-bet-CD4+ T cells in the GC cohort (T-bet+CD8+ T cells vs. T-bet-CD8+ T cells, P=8.26e-24; T-bet+CD4+ T cells vs. T-bet-CD4+ T cells, P=6.45e-51, two-sided Mann-Whitney U test, ****P<0.0001).

[0090] d. Region of interest selection in the TME of ICl responders (ICl-12). T-bet+-PD-L1+ niches were identified using mlHC images of adjacent sections (T-bet, CD4, CD8, PD-1, PD-L1, pan-CK, and DAPI staining);

[0091] e. Expression of MSS-PRs-related genes (CXCL9, CXCL10, CCL3, CCL5, IFNG, GZMB, CXCR3, IDO1) from the center to the edge of the T-bet+-PD-L1+ niche. Four niches were taken from ICl responders (ICl-12) and another four niches were taken from MSS-PR patients (GC-97);

[0092] Schematic diagram of the positive feedback loop between fT-bet+T cells and PD-L1+ tumor cells.

[0093] Figure 7 Immune infiltration of different GC subtypes; among them,

[0094] a. Violin plots of immune infiltration in patients with EBV+MSS (n=18), MSl-H (n=29), and EBV-MSS (n=202) in the ACRG cohort, estimated by cibersortX, *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001;

[0095] b. Box plot of the percentage of CD8+T cells in GC TME. Box plot of the percentage of CD4+T cells in GC TME, *P<0.05;

[0096] c. Bar graph of CD8+ T cell percentage in GC TME. The percentage was calculated by dividing the number of CD8+ cells by the number of all cells. The blue bar represents an MSl-H GC patient, the red bar represents an EBV+ MSS patient, and the yellow bar represents an EBV-MSS GC patient;

[0097] d. Bar graph of CD4+ T cell percentages in GC TME. Percentages were calculated as the number of CD4+ cells divided by the number of all cells. Each blue bar represents an MS1-H GC patient, and each yellow bar represents an MSS GC patient. All statistical significances were assessed using a two-sided Mann-Whitney U test. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

[0098] Figure 8 Single-cell sequencing analysis of gastric cancer;

[0099] a. Heatmap of cell marker gene expression in different cell types. The detailed method of cell type annotation is described in the methods section.

[0100] b. UMAP images of MSl-H GC (n=30,846) and MSS GC (n=88,041) cells.

[0101] c. Box plots of CD8+T cell fractions in MSl-H and MSS GC patients (P=0.224, two-sided Mann-Whitney U test) and box plots of CD4+T cell fractions in MSl-H and MSS GC patients (P=0.494, two-sided Mann-Whitney U test). *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

[0102] Fig. 9 For the construction of positive-unlabeled learning model;

[0103] a. The process of constructing positive-unlabeled learning models for microarray data (ACRG). MSl-H GC patients with a large number of T cell infiltration were classified as positive samples (due to the lack of TMB in the ACRG cohort), while MSS GC patients with uncertain responses were classified as unlabeled samples. The specific methods are described in the methods section.

[0104] b. Bar graph of the association between PUL scores and their response to ICIs in MSS patients in the PRJEB25780 cohort. Red bars indicate responders, and gray bars indicate non-responders. The gray dashed line indicates the cutoff value of 0.97 for distinguishing MSS GC responders.

[0105] c. T-SNE plot of patients with MSl-H & TILs-H (n = 24), MSS-PR (n = 27), and MSS-NR (n = 217) in the ACRG cohort. The geographic proximity between data points indicates their similarity in immune characteristics. The proximity of the MSI-H & TILs-H subtype to the MSS-PR subtype was observed. The right figure is a heat map of PUL scores in patients with MSl-H & TILs-H (n = 24), MSS-PR (n = 27), and MSS-NR (n = 217) in the ACRG cohort. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

[0106] Fig.10 Violin plots of immune infiltration in patients with MSI-H, MSl-PR (n=27), and MSS-NR (n=193) in the ACRG cohort;

[0107] Statistical significance among subtypes was evaluated by two-sided Mann-Whitney U test, estimated by cibersortX, *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001.

[0108] Fig.11 Tumor-infiltrating T-bet+T cells express higher levels of cytotoxic molecules; among them,

[0109] a. Box plots of geometric mean fluorophore intensities (gMFIs) of IFN-γ, GZMB, and CXCR3 in T-bet+CD8+ T cells and T-bet-CD8+ T cells of 6 gastric cancer cases (IFN-γ, P = 4.329e-03; GZMB, P = 2.165e 03; CXCR3, P = 4.113e-02, two-sided Mann-Whitney U test).

[0110] b. Box plots of geometric mean fluorophore intensities (gMFIs) of IFN-γ, GZMB, and CXCR3 in T-bet+CD4+ T cells and T-bet-CD4+ T cells from 6 gastric cancer patients (IFN-γ, P = 2.165e-03; GZMB, P = 2.165e 03; CXCR3, P = 4.113e-02, two-sided Mann-Whitney U test).

[0111] Flow cytometry-like plots of single-cell transcriptome expression of cT-bet downstream genes in T-bet+CD8+T cells and T-bet-CD8+T cells (TBX21, IFNG, GZMB, PRF1, NKG7, GZMH, and GNLY). The x-axis represents the normalized expression of each downstream gene. The y-axis represents the normalized density of cells at a specific expression level. The red density plot is the expression distribution of T-bet+CD8+T cells, and the gray density plot is the expression distribution of T-bet-CD8+T cells. The percentage of double-positive cells is indicated in red, and the percentage of single-positive cells of downstream genes (T-bet negative) is indicated in black. Normalized expression > 0 was used as the cutoff value for each gene.

[0112] Flow cytometry-like plots of single-cell transcriptome expression of dT-bet downstream genes in T-bet+CD4+T cells and T-bet-CD4+T cells (TBX21, IFNG, GZMB, PRF1, NKG7, GZMH, and GNLY). The x-axis represents the normalized expression of each downstream gene. The y-axis represents the normalized density of cells at a specific expression level. Normalized expression > 0 was used as the cutoff value for each gene. *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001.

[0113] Fig.12 For the preparation of T-bethigh T cells; wherein,

[0114] a. Gating strategy for CD8+CD3+ or CD4+CD3+ T cells after negative separation of CD3+ T cells in PBMC by flow cytometry. The percentages of CD8+ T cells and CD4+ T cells used for transfusion were approximately 32.5% and 62.5%, respectively.

[0115] b. Gating strategy for CD8+CD45+ cells after negative separation of CD8+T cells in PBMC. low ) and IL-12 group (T-bet high )Percentage of T-bet+CD8+T cells.

[0116] c. Flow cytometry was used to analyze the effects of IL-12 or PBS on the expression of T-bet and downstream cytokines in CD8+T cells.

[0117] d. Box plot of geometric mean fluorescence intensity (MFl) of each marker in CD8+ T cells from 6-8 different healthy donors. P values ​​were calculated using a two-sided paired t-test. high ) group and control (T-bet low) group. *P<0.05, **P<0.01, ***P<0.001, ****P<0.0001.

[0118] Fig.13 Adoptive infusion of T-bet+T cells can improve the anti-tumor immunity of humanized mice;

[0119] a. Infusion of T-bet high T cells and T-bet low Tumors at the endpoint of a humanized mouse model of T cells.

[0120] b. Box plots of geometric mean fluorophore intensities (gMFIs) of IFN-γ and GZMB in T-bet+CD4+ T cells and T-bet-CD4+ T cells of 10 CDX GC tumors detected by flow cytometry (IFN-γ, P = 5.828e-04; GZMB, P = 1.827e-04). *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

[0121] Fig.14 T-bet+CD8 T cells infiltrate into the tumor area of ​​gastric cancer (GC) and colorectal cancer (CRC);

[0122] a. mlHC images of MS1-H / dMMR colorectal cancer (CRC) tissues in ICI response (left) and ICI non-response (right). T-bet+CD8+T cells are highly infiltrated in the tumor area of ​​ICI responders, as indicated by white arrows.

[0123] b. Bar graph of T-bet+CD8+T cells in GCs tumor area (n=120). Each blue bar represents an MS1-H GC patient, the red bar represents an EBV+MSS patient, and the yellow bar represents an EBV-MSS GC patient.

[0124] Fig.15 Adoptive infusion of T-bet+CD8+T cells combined with PD-1 antibodies can improve anti-tumor immunity in immune desert tumors;

[0125] Among them, T-bet low CD8+T cells+IgG4 control antibody, T-bet low CD8+T cells+anti-PD-1 antibodies, T-bet low CD8+T cells+IgG4 isotype control antibody and T-bet high Humanized gastric cancer mouse model at the endpoint of CD8+T cell+anti-PD-1 antibody treatment.

[0126] Fig.16 The T-bet+-PD-L1+ niche exists in MSS-PR GC; among them,

[0127] Selection of regions of interest (ROI) in the T-bet+-PD-L1+ niche TME of MSS-PR(GC-97) was identified using mIHC images of adjacent sections (T-bet, CD4, CD8, PD-1, PD-L1, pan-CK and DAPI staining). DETAILED DESCRIPTION

[0128] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0129] The drugs and reagents used in the following examples and experiments can be obtained through commercial channels.

[0130] Example

[0131] 1. Methods and Materials

[0132] 1. Patient characteristics and clinical outcomes

[0133] Clinical information of patients with advanced gastric cancer (GC-ICI cohort) (n=25, Table 1) who received PD-1 inhibitor treatment at the First Affiliated Hospital of China Medical University from May 2019 to December 2021 was retrospectively collected. Patients in the GC-ICI cohort had completed at least 2 cycles of standard treatment with PD-1 inhibitors and were evaluated for response according to RECIST 1.1, with a median follow-up of 135 days. No patient achieved complete remission (CR) in this study. There were 7 patients with partial remission (PR), 15 with stable disease (SD), and 3 with progressive disease (PD). Responders were defined as RECIST CR or PR, and nonresponders were defined as SD or PD. 120 patients with stage I to III gastric adenocarcinoma who underwent surgical treatment at the First Affiliated Hospital of China Medical University were included in the GC cohort, with a median follow-up of 117 months (Table 2). Human PBMCs were isolated from the peripheral blood of healthy donors. The study protocol was approved by the Research Ethics Committee of the First Hospital of China Medical University (#2017-207) and was implemented in accordance with the Helsinki Declaration on Human Research.

[0134] Table 1. Clinical characteristics of patients in the GC-ICI cohort

[0135]

[0136] Table 2. Clinical characteristics of GC patients

[0137]

[0138] 2. MMR status assessment

[0139] Immunohistochemical staining was used to detect the expression of four MMR proteins, MLH1 (MXB Biotechnologies, MAB-0789), MSH2 (MXBBiotechnologies, MAB-0789), MSH6 (MXB Biotechnologies, MAB-0831) and PMS2 (MXBBiotechnologies, MAB-0859), to evaluate the MMR status of patients. Normal tissue adjacent to the cancer was used as a positive internal control. When any MMR protein was missing in the nucleus of tumor cells, it was determined to be MMR-deficient.

[0140] 3. Batch RNA-seq Data Processing

[0141] The raw fasta files of PRJEB25780 and PRJEB40416 were downloaded from https: / / www.ebi.ac.uk / . Alignment to the human reference genome (GRCh38) was performed using STAR (version 2.6.1) according to the recommendations of the GTEx project (https: / / github.com / broadinstitute / gtex-pipeline / blob / master / TOPMed_RNAseq_pipeline.md), followed by standard quality control and gene-level read count and transcript (TPM) quantification using RSEM (version 1.3.1). Raw counts of the TCGA-STAD cohort were downloaded from the GDC data portal (https: / / portal.gdc.cancer.gov / ), and clinical data were retrieved from the original publications. Read counts of PRJEB25780, PRJEB40416, and TCGA-STAD datasets were normalized using ComBat-seq to remove batch effects from different experiments.

[0142] 4. Microarray Data Preprocessing

[0143] The original expression matrix of the ACRG cohort was downloaded from https: / / www.ncbi.nlm.nih.gov / geo / (accession number: GSE62254). Probes were annotated using Affymetrix Human Genome U133plus 2.0Array. RMA normalization was used to remove background noise. For genes that matched multiple probes, the probe with the highest average value in the sample was selected for downstream analysis.

[0144] 5. Single-cell RNA-seq Data Analysis

[0145] We downloaded the raw count files of scRNA sequencing of 26 primary tumor samples from https: / / www.ncbi.nlm.nih.gov / geo / (accession number: GSE183904). After filtering out genes shared by less than 3 cells, cells with less than 500 or more than 6,000 genes, and cells with a mitochondrial RNA percentage ≥ 20, we further filtered samples with less than 1,000 cells. Finally, 26,571 genes from 119,878 cells were retained for the following analysis. The analysis was performed using scanpy 1.8.1 (Wolf, FA, et al. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol 19, 15 (2018)). Gene expression was normalized by dividing by the cell size factor estimated by the calculateSumFactors function and then log-transformed. The Leiden algorithm (resolution = 1) was used to perform unsupervised clustering of the 10-dimensional features of the 4000 highly variable genes after pc reduction. The unified manifold approximation and projection (UMAP) was used to visualize the data.Cell types were manually annotated based on the expression patterns of the following markers: CD4+ T cells (PTPRC+, CD3D+, IL7R+, CD8A-, FOXP-); CD8+ T cells (PTPRC+, CD3D+, IL7R-, CD8A+); natural killer cells (PTPRC+, CD3D-, KLRD1+, NCAM1+); Treg cells (PTPRC+, CD3D+, IL7R+, CD8A-, FOXP+); B cells (PTPRC+, CD19+, CD79A+, SSR3 low, SSR4 low, CD83+); plasma cells (PTPRC1 low, CD19 low, CD79A+, SSR3high, SSR4 high,SDC1+); monocytes / macrophages (PTPRC+,CD14+,CD68+,CD83+,CD86+,CD163+); dendritic cells (PTPRC+,CD14-,CD16-,CD163-,CD74+,HLA-DR+); mast cells (PTPRC+,TPSB2+); cancerous epithelial cells (EPCAM+,KRT8+,KRT18+,KRT19+,CLDN4 high,CLDN7 high,TFF2 low,MUC5AClow); normal epithelial cells (EPCAM+,KRT8+,KRT18+,KRT19+,CLDN4 low,CLDN7 low,TFF2 high,MUC5AC high); fibroblasts (COL1A1+,COL1A2+,COL3A1+); vascular endothelial cells (PECAM1+,CD34+).

[0146] We downloaded raw counts of scRNA sequencing of tumor-infiltrating T cells from https: / / www.ncbi.nlm.nih.gov / geo / (accession number; GSE156728). Only tumor-infiltrating T cells in solid tumors were included in the next analysis. As described in the original paper, we filtered out cells with less than 200 genes detected or >10% mitochondrial UMI counts and genes detected in ≤3 cells. Gene expression was normalized by dividing by the cell size factor estimated by the calculateSumFactors function and then log-transformed. T-bet+CD4+T cells or T-bet+CD8+T cells were defined as CD4+T cells or CD8+T cells with TBX21>0 normalized expression.

[0147] 6. Whole exome sequencing data preprocessing

[0148] The original fasta files of the PRJEB25780 cohort were downloaded from https: / / www.ebi.ac.uk / . The sequences of the WES reads were aligned to the reference human genome GRCh38 using BWA-MEM, and then the Genome Analysis Toolkit (GATK, version 4.2.4.1) was used to perform insert re-alignment and base realignment to generate BAM files that can be used for analysis.

[0149] 7. Immune Infiltration Estimation

[0150] For RNA-seq and microarray data, we estimated the proportions of 22 immune cells using CibersortX (absolute mode) running 1,000 permutations. Samples with P < 0.05 were included in the analysis. QuanTiseq was applied to calculate the proportion of tumor-infiltrating T cells using gene sets for CD4+ T cells, CD8+ T cells, and Treg cells. CibersortX inferred the proportion of T-bet+CD8+ T cells in the PRJEB25780 and PRJEB40416 cohorts using a GC-specific tag matrix constructed based on annotated scRNA-seq data (Table 3).

[0151] Table 3. Differentially expressed genes between MSS-PRs and MSS-NRs

[0152]

[0153]

[0154]

[0155]

[0156]

[0157]

[0158]

[0159]

[0160] 8. Positive-Unlabeled (PUL) Machine Learning

[0161] Due to the significant differences in the structure of RNA-seq and microarray data, we constructed two independent positive-label-free semi-supervised machine learning models.

[0162] 1) Identification of positively labeled samples and unlabeled samples

[0163] Positive samples in the TCGA-STAD cohort were defined as MSI-H patients whose TMB was in the top 20% of all patients and whose proportion of tumor-infiltrating T cells estimated by QuanTiseq was also in the top 20%. Due to the lack of TMB information in the ACRG cohort, positive samples were defined as MSI-H patients whose proportion of tumor-infiltrating T cells was in the top 20% of all patients after QuanTiseq evaluation. All MSS patients in TCGA-STAD, ACRG, and PRJEB25780 were defined as unlabeled samples, regardless of their TMB, T cell infiltration, or clinical outcome.

[0164] 2) Construction of immunological characteristics

[0165] 4,872 immune signature gene sets (ImmuneSigDB subset of C7) were downloaded from https: / / www.gsea-msigdb.org / gsea / msigdb / . Signature gene sets of 28 immune cell types were retrieved from published data (Charoentong, P. et al. Pan-cancer Immunogenomic Analyses Reveal GenotypeImmunophenotype Relationships and Predictors of Response to CheckpointBlockade. Cell Rep 18, 248-262 (2017).). The original immunological features were constructed by running a single sample GSEA on the merged 5000 gene sets (Hanzelmann, S., et al. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14, 7 (2013)). High variance features were selected using sklearn. Feature_selection function, with a variance threshold set at 5e-04. The screened high-variance features are subjected to principal component analysis to form the final 5-dimensional immune features.

[0166] 3) Build training and validation PUL model

[0167] The training dataset (TCGA or ACRG data) and validation dataset (MSS patients in PRJEB25780) were included. The PUL model was trained to predict the PUL scores of MSS patients in the TCGA-STAD data (RNA-seq data) and ACRG data (microarray data) of MSS patients, respectively. Then, the positive and unlabeled samples in the TCGA-STAD data, plus 1 sample from the validation dataset, were used to construct the PUL model of the validation dataset, and the PUL scores of each validation dataset sample were predicted. After iterating all samples of the validation / extended validation set, their PUL scores were used to evaluate their specificity and sensitivity in predicting ICI efficacy.

[0168] 4) Implementation of positive-unlabeled learning

[0169] The bagging support vector machine positive-unlabeled machine learning model was constructed as originally described by Mordelet et al. and the Python package pulearn ( https: / / pulearn.github.io / pulearn / doc / pulearn / ) was implemented (Mordelet, F. et al. PA bagging SVM to learn from positive and unlabeled examples. Pattern Recognition Letters 37:201-209, 2014). The PUL score of unlabeled samples was calculated as the average of 1,000 permutation predictions (positive is 1, negative is 0). Patients with a PUL score ≥ 0.953 were considered potential responders (MSS-PRs) in MSS patients ( Fig. 9 a,9b).

[0170] 9. Analysis of Differentially Expressed Genes

[0171] Differential expression analysis of RNA-seq data was performed using “DESeq2” on the raw counts of the respective datasets (Love, MI, et al. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15, 550 (2014)). Differential expression analysis of microarray data was performed using “Limma” on Log2-normalized data (Ritchie, ME et al. limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res43, e47 (2015)). P values ​​were corrected using the Benjamini-Hochberg method; adjusted P values ​​< 0.05 were considered significant. Upregulated genes in MSS-PR were defined as DEGs with significant P values ​​and positive log fold changes in the TCGA-STAD and ACRG cohorts. Downregulated genes in MSS-PR were defined as DEGs with significant P values ​​and negative log fold changes in the TCGA-STAD and ACRG cohorts.

[0172] 10. Multiplex immunohistochemistry (mIHC)

[0173] Opal 7-Color Manual IHC Kit (Akoya Biosciences, NEL811001KT) was used for mIHC. 4-μm-thick paraffin-embedded tissue (FFPE) sections were dewaxed and rehydrated. For the first round of antigen retrieval, slides were immersed in preheated EDTA solution (pH=8.0, 95°C), heated in a microwave oven at low power for 15 min, then cooled to room temperature (RT) and washed with TBST / 0.5% Tween (3 times, 5 min). Slides were washed and blocked with blocking buffer for 10 min. Primary antibody CD4 (MXB Biotechnologies, MAB-0251, pre-diluted) was incubated at room temperature (RT) for 1 h. Slides were washed and incubated with HRP-labeled secondary antibody (1:1000) for 10 min at RT. After washing, Opal 650 fluorophore (1:200) was used for incubation for 10 min. The above procedure was repeated 5 more times using the following antibody-fluorophore combinations: PD-1 (Cell Signaling Technology, 86163, 1:100) with opal 620 fluorophore (1:100), CD8 (MXB Biotechnologies, MAB-0021, pre-diluted) with opal 540 fluorophore (1:300), pan-CK (MXB Biotechnologies, kit-0009, pre-diluted) with opal 690 fluorophore (1:60), PD-L1 (Cell Signaling Technology, 13684, 1:200) with opal 570 fluorophore (1:150) and T-bet (Abcam, ab154200, 1:100) with opal 520 fluorophore (1:100). The PD-L1 primary antibody was replaced with the CXCR3 primary antibody (R&D, MAB160, 10 μg / ml), and other conditions remained unchanged to observe the co-localization of CXCR3 and T-bet.

[0174] 11.mIHC Image Analysis

[0175] Images of more than nine 20× fields of view were obtained from each section using the Mantra 2 quantitative pathology workstation (Akoya Biosciences). For sections with a clear tumor-normal boundary, images of adjacent normal tissue (N), tumor-normal interface edge (M), tumor edge close to the interface (E), and tumor core (C) were also obtained. The acquired images were analyzed using inFrom 2.3.0 (Akoya Biosciences), followed by tissue segmentation, cell segmentation, and cell phenotyping. Based on the pan-CK expression pattern, a supervised machine learning model was trained in pre-defined tumor areas, stromal areas, and blank areas. Due to the significant differences in cell morphology, cell segmentation in tumor and stromal areas was trained separately based on nuclear staining DAPI. The phenotype of each cell in the tumor and stroma was determined using the following cutoff intensities: T-bet (opal520) ≥2, CD8 (opal540) ≥2, PD-L1 (opal 570) ≥2, PD-1 (opal 620) ≥2, CD4 (opal 650) ≥2, and pan-CK (opal690) ≥2.

[0176] 12. Quantification of Neighboring Cells of T-bet+ / T-bet- T Cells

[0177] To evaluate whether T-bet+ T cells can upregulate PD-L1 on neighboring cells, we retrieved the coordination information of 132,241 T cells in the gastric cancer cohort. Neighboring cells of a specific T cell were defined as any cell within a radius of 50 μm. The percentage of PD-L1+ neighboring cells was calculated for each T cell.

[0178] 13. Spatial transcriptome analysis

[0179] First, FFPE samples of ICI-12 (partial response, MSI-H) and GC-97 (potential response, MSS) were stained using the above mIHC method to determine the location of the T-bet+-PD-L1+ niche. Four niches were identified from each section. DSP Slide Preparation User Manual Process adjacent FFPE slides. Bake the slides in a 60°C oven for at least 3 hours to expose the antigens. After dehydration, add proteinase K for digestion. Mix the slides with WTA probes and incubate overnight. On the second day, stain with PD-L1 (OriGene, CF809809), CD45 (CST, 13917), PanCK (Novus, NBP2-33200AF532) and Cyto13 (ThermoFisher, S7575) for 2 hours. Draw three concentric circles as regions of interest (ROIs) on each T-bet+-PD-L1+ niche microenvironment. For each ROI, the area of ​​interest (AOIs) was further divided into immune cells and cancer cells based on the expression of CD45 and PanCK proteins. Oligoes in the ROI were collected by UV lysis. PCR reaction purification. The library was subjected to end-pair sequencing. Using DSP data analysis software performed raw counts for all ROIs. Data were normalized using the background subtraction Q3 normalization method.

[0180] 14. T cell culture and expansion

[0181] Healthy human PBMCs were separated by centrifugation at 400g for 30 min using Ficoll solution (Tbd science, LTS1077). Non-CD3+ cells or non-CD8+ cells in PBMCs were removed by the following method: first incubation with biotin antibody cocktail, then incubation with magnetic streptavidin nanobeads (Biolegend, 480022, 480012). Untreated CD3+ or CD8+ cells were collected using a magnetic separator (StemCell Technologies, 18000). CD3+ cells or CD8+ cells were then washed and resuspended in human lymphocyte culture medium (Takara, GT-T551) + 10% FBS (Gibco, 10099) at a concentration of 1×10 6 / ml. 1ng / ml CD3 antibody (BD Biosicences, 566685) and 5ng / ml CD28 antibody (BDBiosicences, 555725) were coated on a 12-well plate (Nest, 712011) that had not been treated with TC overnight at 4°C. After activating T cells on the culture plate coated with CD3 / CD28 antibodies for 48h, IL-2 (100IU / ml, Peprotech, 200-02) was added to rapidly expand T cells. Fresh IL-2 was added every day, and the cell density was maintained at 0.5-2×10 during in vitro culture. 6T cells were activated once every 7 days using CD3 / CD28 antibody-coated plates, and dead cells were removed using a dead cell removal kit according to the manufacturer's instructions (Miltenyi Biotec, 130-090-101).

[0182] 15. Induce T-bet expression in T cells

[0183] T-bet expression is upregulated by IFN-γ and IL-12 stimulation in T cells. 72 h after CD3 / CD28 antibody withdrawal, IL-12 (50 ng / ml, Peprotech, 200-12 h) was added to the culture medium to induce T-bet high CD3+T cells (or T-bet high CD8+T cells), and the control group was added with an equal amount of PBS to maintain low T-bet expression. Fresh IL-12 was added every day for at least two consecutive days. Intracellular flow cytometry confirmed that T-bet and its downstream cytokines were upregulated in T cells.

[0184] 16. Humanized tumor-bearing mouse model

[0185] Female NYG (Prkdc- / -il2rg- / -nod) mice aged 4 to 6 weeks were purchased from Liaoning Changsheng Biotechnology and raised under pathogen-free conditions. On day 0, 5×10 6 PBMCs were injected into mice to reconstitute the mouse immune system. On day 5, 5×10 6 HGC-27 cells were implanted subcutaneously on the right side of the mice. On day 8, the mice were randomly divided into T-bet high Group and T-bet low On the 8th and 10th days, 1×10 7 T-bet high or T-bet low CD3+ T cells. Tumor length and width were measured every 2 days by an independent investigator who was unaware of the treatment conditions. Tumor volume was calculated as 1 / 2×(Length×width 2 All animal experimental protocols were approved by the Research Ethics Committee of China Medical University (No. CMUXN2021712).

[0186] 17. Humanized tumor-bearing mouse model of immune desert

[0187] HGC-27 cells were inoculated subcutaneously on the right side of multiple NYG mice. When the tumor length reached 1 cm, the immune desert tumors were collected. The tumors were evenly cut into 20 mm 3The grafts were subcutaneously transplanted into the right side of the humanized mice. On the 5th and 7th days after transplantation, 5×10 PBMCs from the same healthy donor were injected into the tail vein. 6 T-bet high or T-bet low CD8+T cells. Anti-PD-1 (200 μg / injection, Teplizumab, Junshi Pharmaceuticals) or human IgG4 isotype control (50 μg / injection, Sigma-Aldrich, 14640) was administered via the tail vein every 2 days after transplantation. Tumor length and width were measured every 2 days by an independent researcher who was unaware of the treatment conditions.

[0188] 18. Isolation of Tumor-Infiltrating Immune Cells

[0189] Tumors from gastric cancer patients and CDX mouse models were cut into small pieces with sterile scissors in HBBS containing calcium and magnesium (Solarbio, H1025). The tissue was dissociated in 200U / ml collagenase type IV (Gibco, 17104019), 3mM CaCl2 (Solarbio, C7250), 0.01mg / ml DNAse I (Solarbio, D8071) and 0.025mg / ml hyaluronidase (Solarbio, H8030), using GentleMACS C tubes on GentleMACS dissociator (Miltenyi Biotec, 130-093-235), and then incubated at 37°C for 1h. The resulting cell suspension was passed through a 70-μm filter (MACS Smart Strainers, 5201110409) and prepared into single cells by centrifugation at 350g for 5min. Incubate in RBC lysis buffer (Biolegend, 420301) at room temperature for 10 min to remove red blood cells. Wash cells twice and resuspend in calcium / magnesium-free HBBS (Solarbio, H1040). CD45+ cells are then isolated using CD45+ positive magnetic beads (StemCell Technologies, 100-0107) and a magnetic separator.

[0190] 19. Multicolor flow cytometry

[0191] Single cell suspensions were stained with 5 μg / ml human Fc receptor blocking solution (biolegene, 422302) at 4°C for 5 min. Cells were stained with a fluorescein-labeled antibody cocktail for 40 min at 4°C. Flow cytometry was performed using a multicolor flow cytometer (Sony, SA3800), and data were analyzed using FlowJo V10.4 software. For intracellular staining of T-bet, surface-stained cells were fixed and permeabilized using transcription factor buffer (BD Biosciences, 562574) according to the manufacturer's instructions. For intracellular cytokine staining, PMA (Sigma-Aldrich, P8139) and ionomycin (Sigma-Aldrich, I3909) were added for 12 h, followed by brefeldin A (Biolegend, 420601) for 4 h to enhance the intracellular staining signal before cell surface staining.

[0192] 20. Statistical Analysis

[0193] The Mann-Whitney U test was used for comparison between the two groups. Kaplan-Meier analysis of the overall survival of the GC cohort was performed using the R package 'survival' (version 3.5). Two-way analysis of variance was performed on the data collected at different time points, using time and interference as the two independent variables, to compare the tumor size between the two groups. The Benjamini-Hochberg correction method was applied to reduce the false discovery rate of multiple comparisons. P < 0.05 was considered statistically significant (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001). All statistical analyses were performed using R 4.1.1 or SciPy 1.8.1 (Python 3.9.7).

[0194] 2. Conclusion

[0195] 1. Patient characteristics and clinical outcomes

[0196] Clinical information of patients with advanced gastric cancer (GC-ICI cohort) (n=25, Table 1) who received PD-1 inhibitors at the First Affiliated Hospital of China Medical University from May 2019 to December 2021 was retrospectively collected. Patients in the GC-ICI cohort had completed at least 2 cycles of standard treatment with PD-1 inhibitors and were evaluated for response according to RECIST 1.1, with a median follow-up of 135 days. No patient achieved complete remission (CR), 7 patients achieved partial remission (PR), 15 patients had stable disease (SD), and 3 patients had progressive disease (PD).

[0197] A total of 120 patients with stage I-III gastric adenocarcinoma who underwent surgical treatment at the First Affiliated Hospital of China Medical University were included, with a median follow-up of 117 months. In the gastric cancer cohort, 5.83% (7 / 120) of patients were MSI-H / dMMR. The clinical characteristics of GC patients are shown in Table 2.

[0198] 2. Immune heterogeneity of different GC subtypes

[0199] In multiple clinical trials, MMR status and EBV status have been associated with different sensitivities to ICI treatment. In our GC-ICI cohort, the remission rates of MSI-H, EBV+MSS, and EBV-MSS subtypes were 57.14% (4 / 7), 25% (1 / 4), and 14.29% (2 / 14), respectively (Table 1). To understand the determinants of different responses, we performed immune microenvironment analysis of different GC subtypes based on bulk RNA sequencing, single-cell RNA sequencing (scRNA-seq), and multiplex immunohistochemistry (mIHC). In the TCGA-STAD and ACRG cohorts, CD8+ T cells estimated by CibersortX were the most abundant in EBV+MSS GC ( Figure 1 a, Figure 7 a). In the ACRG cohort, the proportion of CD8+T cells in MSI-H GC was slightly higher than that in EBV-MSS GC (P=0.05, Figure 7 a), but not in the TCGA-STAD cohort (P = 0.39, Figure 1 a). CD4+ T cells in EBV+MSS and MSI-H GC tend to be polarized toward a more activated phenotype, as shown by a higher activation / resting ratio of memory CD4+ T cells ( Figure 1 a, Figure 7 a). In EBV+MSS and MSI-H patients, macrophages (TAMs) tend to exhibit an M1 phenotype with anti-tumor activity ( Figure 1 a, Figure 7 a). To compare the abundance of T cell infiltration in real-world patients, we performed mIHC staining on tumor tissues from patients in our GC cohort ( Figure 1 b). Although there was no significant difference in the abundance of CD8+T cells (P=0.44) or CD4+T cells (P=0.98) between MSI-H and MSS patients ( Figure 7 b), but we observed significant immune infiltration heterogeneity in the TME of MSS GC. Compared with the immune desert state exhibited by most MSS GC, a small number of MSS GCs had abundant tumor-infiltrating T cells in the TME ( Figure 7 c,7d).

[0200] To investigate the functional diversity of TILs in the TME of GC, we performed unbiased single-cell clustering of 6 MSI-H and 18 MSS GCs and ultimately annotated 14 cell types ( Figure 1 c, Figure 8 As expected, the proportion of CD8+ T cells (P=0.22) or CD4+ T cells (P=0.49) among immune cells was not significantly increased in MSI-H GC ( Figure 1 d, Figure 8 c). The levels of cytotoxic molecules (GZMB, GZMH, GZMA, GNLY, NKG7) and immune checkpoint molecules (LAG3, CTLA-4, HAVCR2 / TIM-3, TIGIT) in CD8+ T cells of MSI-H GCs were increased ( Figure 1 e). The above results indicate that the heterogeneity of the immune microenvironment of different GC subtypes exists not only in the composition of immune cells but also in the diversity of functions and cell-cell interactions.

[0201] 3. Comprehensive characteristics of MSS-PRs

[0202] Using positive-unmarked (PUL) machine learning as described in Methods and Materials 8, MSS GCs in TCGA-STAD and ACRG were divided into MSS-PRs and MSS-NRs. The proportion of MSS-PRs in MSS GCs was 19.37% in the TCGA-STAD cohort and 11.07% in the ACRG cohort, which is comparable to the ORR reported in various clinical trials ( Figure 1 g, Fig. 9 c). To independently verify the accuracy of the PUL model, we included MSS GC treated with PD-1 inhibitors in the PRJEB25780 cohort into the validation dataset. Surprisingly, the PUL model performed very well in identifying ICI responders from MSS GC, with an AUC of 0.924, making it the most accurate model to date ( Figure 1 f).

[0203] To comprehensively characterize the characteristics of MSS-PRs, we performed multidimensional differential analysis of immune infiltration and transcriptome expression between MSS-PRs and MSS-NRs. As expected, the immune characteristics of MSS-PRs patients were similar to those of MSI-H, TMB-H, and TILs-H patients (positive samples), with higher infiltration of CD8+ T cells, activated CD4+ memory T cells, and M1-polarized macrophages ( Figure 2 a, Fig.10). The differentially expressed genes (DEGs) between MSS-PRs and MSS-NRs suggested potential determinants of ICI response (Table 3). We noted that CXCR3-ligands (CXCL9, CXCL10, CXCL11) were among the most significantly upregulated genes in MSS-PRs ( Figure 2 b). CXCL9 / 10 / 11 are known to be upregulated in an interferon-γ (IFN-γ)-dependent manner. Interestingly, MSS-PRs also showed elevated levels of IFN-γ (gene named IFNG) and many interferon-stimulated genes (ISGs, including IDO1, CD274 / PD-L1, WARS, GBP1 / 4 / 5) ( Figure 2 b). In addition, multiple cytotoxic molecules (GZMB, GNLY, GZMA, GZMH, NKG7, PRF1) and immune checkpoints (KLRC1, FASLG, LAG3, KLRD1) were also upregulated in MSS-PRs ( Figure 2 b). The most highly upregulated DEGs showed significant accuracy in predicting ICI response in MSS patients, with an average AUC of 0.85 ( Figure 2 c).

[0204] 4. T-bet+ T cells show stronger anti-tumor immunity

[0205] The upregulated genes in MSS-PRs highlight the importance of the IFN-γ signaling pathway and the CXCR3-CXCL9 / 10 / 11 chemoattractant receptor-ligand pair for ICI outcomes. CXCR3 and IFNG are target genes directly transcribed by the immune cell-specific T-box transcription factor T-bet. Surprisingly, in addition to the classical target genes of T-bet transcription, multiple cytotoxic molecules, such as GZMB, PRF1, NKG7, GZMH, and GNLY, were also upregulated in tumor-infiltrating T-bet+ T cells, which was also verified by flow cytometry and ScRNA-Seq ( Figure 3 a,3b, Fig.11 a-11d). The large overlap between upregulated genes and T-bet-regulated genes in MSS-PRs suggests that T-bet+ T cells have a potential key role in ICI outcomes and antitumor immunity.

[0206] Multiple immunohistochemistry confirmed that CXCR3 and T-bet were co-expressed in tumor-infiltrating CD4+ and CD8+ T cells ( Figure 3 c). We speculate that T-bet+ T cells may be chemotactic to the tumor core through CXCL9 / 10 / 11-CXCR3 chemotaxis and exhibit antitumor activity by secreting IFN-γ, granzyme B, and other cytotoxic molecules ( Figure 3d). To test this hypothesis, we used mIHC to quantitatively analyze the proportion of T-bet+ and T-bet- T cells in 19 gastric cancers (18 MSS gastric cancers and 1 MSI-H gastric cancer) from adjacent normal tissues (N), normal tumor margins (M), tumor margins (E), and tumor cores (C). From N to C, the proportion of CD8+ T cells in the stroma gradually decreased. The proportion of CD4+ T cells peaked at the tumor margin and dropped sharply before entering the tumor core ( Figure 3 e). Interestingly, the proportion of T-bet+CD8+ and T-bet+CD4+ T cells increased from N to C, indicating that T-bet+ T cells have a stronger ability to infiltrate into the tumor core, which may be due to the chemotactic effect of the CXCL9 / 10 / 11-CXCR3 chemoattractant pair ( Figure 3 e).

[0207] After we induced T-bet expression with IL-12, the expression of IFN-γ and granzyme B in CD8+T cells increased significantly ( Fig.12 a-12d). In order to evaluate the anti-tumor ability of T-bet+T cells in vivo, we isolated T-bet high or T-bet low T cells were infused into the peripheral circulation of humanized gastric cancer mice ( Figure 3 f) As expected, compared to accepting T-bet low Compared with control mice treated with T cells, mice that received T-bet high The tumor volume of mice treated with T cells was significantly reduced ( Figure 3 g, Fig.13 a) As demonstrated by mIHC and intratumoral flow cytometry, infusion of T-bet high There are a large number of T-bet+CD8+ and T-bet+CD4+ T cells infiltrating the TME of mice with T cells, while T-bet low T cell-infused mice remain in an immune desert state ( Figure 3 h, 3i). Tumor-infiltrating T-bet+CD4+ or CD8+ T cells also have higher levels of granzyme B and IFN-γ, showing stronger cytotoxicity ( Figure 3 j, Fig.13 b).

[0208] To validate the anti-tumor activity of tumor-infiltrating T-bet+T cells in real-world gastric cancer patients, we calculated the percentage of T-bet+CD8+T cells and T-bet+CD4+T cells in a gastric cancer patient cohort using mIHC. We found that patients with a higher abundance of T-bet+CD8+T cells in the TME had a longer median OS (77.0 months vs. 24.0 months), indicating that T-bet+CD8+T cells have a stronger anti-tumor activity ( Figure 3 k). Our results suggest that T-bet+ T cells, especially T-bet+CD8+ T cells, are a unique T cell subset with stronger infiltration capacity and exert antitumor activity by secreting IFN-γ and granzyme B.

[0209] 5. T-bet+CD8+T cells are highly infiltrated in ICI-sensitive tumors

[0210] Surprisingly, we found that despite the presence of T-bet high CD4+T cells / T-bet high The ratio of CD8+T cells was 2.75 times, but in the TME, the ratio of tumor-infiltrating T-bet+CD4+T cells / T-bet+CD8+T cells dropped to 0.60 ( Figure 4 a). Similarly, single-cell analysis of tumor-infiltrating immune cells in GCs revealed that the majority of tumor-infiltrating T-bet+ cells were CD8+ T cells, suggesting that T-bet+CD8+ T cells play a more important role in the TME of GCs ( Figure 4 To further understand the determinants of ICI response at the spatial level, we quantified the proportion of T-bet+CD8+ T cells in different regions of GC ( Figure 4 d, 4e). The percentage of T-bet+CD8+T cells continued to increase from 0.21% in the interstitial area to 0.25% in the tumor area, while the percentage of T-bet+CD4+T cells decreased to about 50% of the T-bet+CD8+T cells in the tumor area, indicating that T-bet+CD8+T cells were closer to tumor cells ( Figure 4 f).

[0211] We observed similar spatial distribution of T-bet+CD8+ T cells in tumor regions of ICI responders and MSS-PRs, whereas non-responders showed little infiltration despite being MSI-H ( Figure 4 g, Fig.14a). In the GC cohort, MSI-H GC had a trend of increased T-bet+CD8+T cell infiltration, and 14.16% (16 / 113) of MSS GC had more T-bet+CD8+T cell infiltration than most MSI-H GC ( Fig.14 b). To test whether tumor-infiltrating T-bet+CD8+ T cells can predict ICI response in a wider range of GC patients, we deconvoluted the proportion of tumor-infiltrating T-bet+CD8+ T cells in patients treated with ICIs using the immune cell score generated from scRNA-seq data (Table 4). We found that in MSS GC, higher abundance of T-bet+CD8+ T cells was associated with better ICI response, with an AUC of 0.800 ( Figure 4 h).

[0212] Table 4. Deconvolution of the proportion of tumor-infiltrating T-bet+CD8+ T cells in patients treated with ICI

[0213]

[0214]

[0215]

[0216]

[0217] 6. T-bet+CD8+T cells can increase the sensitivity of immune neglected GC to ICI treatment

[0218] Tumor-infiltrating T-bet+CD8+T cells exhibit superior antitumor activity and predict ICI response in MSS GC. We sought to evaluate the role of T-bet high Can adoptive infusion of CD8+T cells improve the outcome of ICI treatment of immune-desert MSS GC? To mimic the TME of immunodeficient patients, we injected human MSS GC cells HGC-27 subcutaneously into immunodeficient NYG mice to generate immune-desert tumors, and then minced these tumors and transplanted them subcutaneously into humanized mice ( Figure 5 a). Continuous injection of T-bet via tail vein high CD8+ or T-bet low CD8+T cells, PD-1 inhibitors or isotype IgG4 were administered every 2 days ( Figure 5 a) At the treatment end point, at T-bet low In CD8+ T cell-infused mice, anti-PD-1 mAb reduced tumor size by 45.2% compared with administration of isotype IgG4; high In mice with CD8+ T cells transferred, this number increased to 68.3% ( Fig.15 , Figure 5 b) As you can see, T-bet high CD8+T cell infusion significantly enhanced the efficacy of ICI. mIHC and flow cytometry in tumor tissue confirmed that T-bet high Combination therapy of CD8+ T cell infusion and anti-PD-1 antibody significantly increased immune infiltration in the tumor core of mice ( Figure 5 c,5d). Based on the above research results, we conclude that T-bet high CD8+T cells enhance antitumor immunity and confer sensitivity of immune-desert GC to ICI. This combination regimen is expected to revolutionize the treatment of immune-desert GC, which currently accounts for the majority of advanced GC.

[0219] 7. Positive feedback loop between T-bet+ T cells and PD-L1+ tumor cells drives ICI sensitivity

[0220] We further studied the TME of ICI-sensitive GC and found that there was a T-bet+-PD-L1+ niche in patients and humanized mice with a large infiltration of T-bet+CD8+T cells ( Figure 5 e, Figure 6 a). We speculated that the high expression of PD-L1 (and other ISGs) might be induced by T-bet+ T cells secreting IFN-γ. We calculated the percentage of PD-L1+ cells in the neighboring cells (within a radius of 50 μm) of T-bet+ or T-bet- T cells ( Figure 6 b), the results showed that the proportion of PD-L1+ cells in the neighboring cells of T-bet+T cells increased significantly: the proportion of PD-L1 positive cells in the neighboring cells of T-bet-CD8+ cells was 1.19%, and the proportion of PD-L1 positive cells in the neighboring cells of T-bet+CD8+ cells was 4.82%; the proportion of PD-L1 positive cells in the neighboring cells of T-bet-CD4+T cells was 0.86%, and the proportion of PD-L1 positive cells in the neighboring cells of T-bet+CD4+T cells was 2.93% ( Figure 6 c). To further analyze the transcriptional signature of the T-bet+-PD-L1+ microenvironment, we performed spatial RNA-seq analysis in one ICI responder and one MSS-PR patient ( Figure 6 d, Fig.16 From the edge to the center of the T-bet+-PD-L1+ microenvironment, the levels of a large number of genes (such as CXCL9 / 10 / 11, CCL3 / 4 / 5, IFNG, GZMB, CXCR3, IDO1) continued to increase, forming a chemotactic gradient that facilitates the migration of T-bet+ T cells ( Figure 6 e).

[0221] Therefore, we proposed a positive feedback loop between T-bet+ T cells and PD-L1+ tumor cells (ISGs+) mediated by IFN-γ signaling and the CXCL9 / 10 / 11-CXCR3 chemotactic pair, which ultimately drives ICI sensitivity in gastric cancer patients ( Figure 6 f). T-bet+CD8+T cells are chemotactic to the tumor core through the CXCL9 / 10 / 11-CXCR3 ligand-receptor pair, forming a microenvironment with high concentrations of IFN-γ. IFN-γ upregulates the expression of ISGs such as CXCL9 / 10 / 11 and PD-L1 in neighboring cells, resulting in more T-bet+T cells being chemotactic to the tumor core to exert cytotoxic effects, and IFN-γ further upregulates T-bet expression in other T cells. The PD-L1-induced expression of neighboring cells and antigen-driven CD8+T cell exhaustion provide the best prerequisites for ICIs to exert their therapeutic effects. In mIHC, this positive feedback loop is regarded as a unique T-bet+-PD-L1+ niche, which is not only a milestone for predicting the anti-tumor immune efficacy of ICI, but also clarifies the complex interaction between its tumor immunosuppression and ICI susceptibility.

[0222] The technical scheme of the present invention is described herein in conjunction with some specific embodiments and examples, and many details are stated and illustrated. However, it should be understood that the specific embodiments and examples provided herein are merely exemplary and do not limit the scope of protection of the present invention. In fact, for those skilled in the art, the present invention can also be implemented by adopting other specific modes, and the corresponding modifications, changes or adjustments do not depart from the spirit and purpose of the present invention, and thus should be deemed to be included within the scope of the present invention.

Claims

1. Application of T-bet+CD8+T cells as markers in the preparation of drugs for the treatment of gastrointestinal tumors.

2. The use according to claim 1, characterized in that: The level of T-bet+CD8+T cells in the sample was measured, and the therapeutic effectiveness of the PD-1 signal inhibitor for gastrointestinal tumors was determined based on whether the T-bet+CD8+T cell expression was positive.

3. The use according to claim 2, characterized in that: The sample is selected from any one or more of peripheral blood, paracancerous tissue, and cancerous tissue.

4. The use according to claim 2, characterized in that: The PD-1 signal inhibitor is selected from any one or more of anti-PD-1 antibodies, anti-PD-L1 antibodies and anti-PD-L2 antibodies.

5. The use according to claim 1, characterized in that: The digestive tract tumor is selected from any one of gastric cancer, intestinal cancer, esophageal cancer, pancreatic cancer, liver cancer, and bile duct cancer.

6. The use according to claim 5, characterized in that: The gastric cancer is microsatellite stable gastric cancer MSS GC.

7. The use according to claim 6, characterized in that: The T-bet+CD8+ T cells were highly infiltrated in MSS GC.

8. The use according to claim 6, characterized in that: The T-bet+CD8+T cells are chemotactic to the tumor core through the CXCL9 / 10 / 11-CXCR3 axis and exert anti-tumor activity by secreting IFN-γ, granzyme B and other cytotoxic molecules.

9. The use according to claim 6, characterized in that: The abundance of T-bet+CD8+T cells was correlated with ICI response, and the AUC for predicting ICI response in MSS patients based on the cell fraction of tumor-infiltrating T-bet+CD8+T cells was 0.

800.

10. The use according to claim 6, characterized in that: The T-bet+CD8+T cells enhance anti-tumor immunity and confer sensitivity to ICI in immune-desert gastric cancer.

11. The use according to claim 2, characterized in that: The applications specifically include: 1) Determine the level of T-bet+CD8+T cells in the sample to determine whether T-bet+CD8+T cell expression is positive; 2) If the sample is positive for T-bet+CD8+T cell expression, proceed to step 3); Or, if the expression of T-bet+CD8+T cells in the sample is negative, the peripheral blood mononuclear cells PBMCs are cultured and activated in vitro to induce the expression of T-bet in CD8+T cells, and the T-bet+CD8+T cells are obtained and then infused in vivo until the expression of T-bet+CD8+T cells is positive; 3) detecting the abundance of the T-bet+CD8+T cells; 4) When the abundance of the T-bet+CD8+T cells reaches a preset value, the T-bet+CD8+T cells are used in combination with a PD-1 inhibitor.

12. The use according to claim 11, characterized in that: The induction step of T-bet+CD8+T cells comprises: (a) T cells were activated using culture plates coated with CD3 / CD28 antibodies for 48 hours; (b) Add 100 IU / ml IL-2 to rapidly expand T cells; (c) 72 hours after withdrawal of CD3 / CD28 antibodies, 50 ng / ml IL-12 was added to induce T-bet expression in CD8+ T cells.

13. The use according to claim 11, characterized in that: The dosage of T-bet+CD8+T cells is 1×10 6 -1×10 8 Cells / time.

14. The use according to claim 11, characterized in that: The T-bet+CD8+T cells are used 1-3 times per week.

15. The use according to claim 11, characterized in that: The T-bet+CD8+T cells are infused via intravenous injection or local injection.

16. The use according to claim 11, characterized in that: The combined use of T-bet+CD8+T cells and PD-1 inhibitors is to add the PD-1 inhibitor within 24 hours after T-bet+CD8+T cell infusion.

17. The use according to claim 11, characterized in that: The abundance of T-bet+CD8+T cells in the step 4) reaches the preset value, which means that the proportion of T-bet+CD8+T cells in all stromal cells is >0.1%; if the proportion of T-bet+CD8+T cells in all stromal cells is <0.1%, return to step 2), and the peripheral blood mononuclear cells PBMCs are cultured and activated in vitro to induce the expression of T-bet in CD8+T cells, and after obtaining T-bet+CD8+T cells, they are infused in vivo, and the abundance of T-bet+CD8+T cells is detected again.

18. A pharmaceutical composition for treating digestive tract tumors, characterized in that: Contains at least T-bet+CD8+T cells and PD-1 signaling inhibitors.

19. The pharmaceutical composition according to claim 18, characterized in that: The content ratio of T-bet+CD8+T cells and PD-1 signal inhibitor is 5×10 6 Piece: 200μg.

20. The pharmaceutical composition according to claim 18, characterized in that: The T-bet+CD8+T cells were infused twice, and 200 μg of PD-1 monoclonal antibody was infused once every 2 days for a total of 11 times.

Images