Method of predicting effectiveness of immune checkpoint inhibitor (ICI) treatment in subject, method of treating tumor in subject

Abstract

The present invention provides methods of predicting the effectiveness of an immune checkpoint inhibitor (ICI) treatment in a subject. The invention additionally provides methods of treating a tumor in a subject determined to be effective.

Description

TECHNICAL FIELD

[0001] The present application relates to a method of predicting effectiveness of an immune checkpoint inhibitor (ICI) treatment in a subject. The present application additionally relates to a method of treating a tumor in a subject determined to have effectiveness. BACKGROUND

[0002] Revolutionary progress has been made in tumor treatment by applying an immune checkpoint inhibitor to the tumor treatment. However, there still exists a patient group in which sufficient therapeutic effect is not observed even with the immune checkpoint inhibitor. The immune checkpoint inhibitor is expensive, and thus, if a patient who is expected to exert a therapeutic effect can be determined and the immune checkpoint inhibitor is administered mainly to the patient, it contributes to the medical economy. Patent Literature 1 discloses a method of predicting effectiveness of an immune checkpoint inhibitor based on a combination of evaluation items of PD-1 expression intensity and PD-1 expression ratio in T cells. Patent Literature 2 discloses a method of predicting effectiveness of an immune checkpoint inhibitor based on RNA-Seq of a tumor. + A method of predicting effectiveness of an immune checkpoint inhibitor based on a combination of evaluation items of PD-1 expression intensity and PD-1 expression ratio in T cells. Patent Literature 2 discloses a method of predicting effectiveness of an immune checkpoint inhibitor based on RNA-Seq of a tumor.

[0003] PRIOR ART DOCUMENTS

[0004] PATENT LITERATURE

[0005] Patent Literature 1: WO2019 / 230919

[0006] Patent Literature 2: WO2018 / 231772 SUMMARY

[0007] The present application provides a method of predicting effectiveness of an immune checkpoint inhibitor (ICI) treatment in a subject. In the present application, ICI treatment effectiveness on a subject can be predicted based on gene expression in cells contained in peripheral blood of the subject (e.g., peripheral blood mononuclear cells and mononuclear cells). The present application additionally provides a method of treating a tumor in a subject determined to have effectiveness.

[0008] According to the present application, for example, the following application is provided.

[0009] (1) A method of predicting tumor treatment effectiveness of an immune checkpoint inhibitor (ICI) in a subject, comprising:

[0010] determining an expression amount of a gene included in a gene group selected from one or more genes described in any one of Tables 1 to 7 (preferably, Table 1) in a sample (e.g., a sample containing mononuclear cells such as peripheral blood mononuclear cells) obtained from the above-mentioned subject; and

[0011] predicting the treatment effectiveness of the ICI from the determined expression amount using a prediction model constructed in a manner capable of predicting the above-mentioned treatment effectiveness or based on the determined expression amount,

[0012] The above-described prediction model is constructed based on the expression amount of the gene(s) including the above-described one or more genes and information on the therapeutic effectiveness of the ICI in a sample (e.g., a sample containing mononuclear cells such as peripheral blood mononuclear cells) obtained from a patient group in which the ICI is therapeutically effective and a patient group in which the ICI is not therapeutically effective, respectively.

[0013] (2) The method according to the above (1), wherein the one or more genes selected from the group of genes described in Table 1 include Figures 4A-4O any one or more genes described in Table 1.

[0014] (3) The method according to the above (1) or (2), wherein the one or more genes selected from the group of genes described in Table 1 include one or more or all of the genes selected from the group consisting of CLEC12A, TMEM176A, and TMEM176B.

[0015] (4) The method according to any one of the above (1) to (3), wherein the one or more genes selected from the group of genes described in Table 1 include five or more genes.

[0016] (5) The method according to any one of the above (1) to (4), wherein the one or more genes selected from the group of genes described in Table 1 include ten or more genes.

[0017] (6) The method according to any one of the above (1) to (5), wherein the one or more genes selected from the group of genes described in Table 1 include twenty or more genes.

[0018] (7) The method according to any one of the above (1) to (6), wherein the above-described prediction model is constructed by machine learning.

[0019] (8) The method according to any one of the above (1) to (6), wherein the above-described prediction model is constructed by logistic regression.

[0020] (9) A method of treating a tumor in a subject, comprising the step of administering a therapeutically effective amount of an immune checkpoint inhibitor (ICI) to the subject,

[0021] the subject being a subject predicted to be therapeutically effective for an immune checkpoint inhibitor (ICI) by the method according to any one of the above (1) to (8).

[0022] (10) The method according to any one of the above (1) to (9), wherein the immune checkpoint inhibitor (ICI) is an antibody or a PD-1 binding fragment of an antibody capable of binding to PD-1 and neutralizing the binding of PD-1 to PD-L1.

[0023] (11) The method according to any one of the above (1) to (9), wherein the immune checkpoint inhibitor (ICI) is an antibody or a PD-1 binding fragment of an antibody that is capable of binding to PD-L1 and neutralizing the binding of PD-1 to PD-L1.

[0024] (12) An immune checkpoint inhibitor (ICI) or a pharmaceutical composition containing an immune checkpoint inhibitor for use in the method according to any one of the above (9) to (11).

[0025] (13) The application according to any one of the above, wherein the one or more genes selected from the group of genes recited in Table 1 comprise the group of genes recited in any one of Figures 19-23

[0026] (14) A method of predicting an effect of a candidate substance on a therapeutic responsiveness of a patient to an immune checkpoint inhibitor therapy (ICI therapy), comprising:

[0027] determining an expression amount of a gene comprising one or more genes selected from the group of genes recited in any one of Tables 1 to 7 in each of a first sample obtained from the patient before administration of the candidate substance and a second sample obtained from the patient after administration of the candidate substance; and

[0028] predicting an effect of the ICI therapy based on the determined expression amounts using a prediction model constructed in a manner capable of predicting the effect of the ICI therapy (e.g., constructable by the above (1)),

[0029] the prediction model is constructed based on expression amounts of the gene comprising the one or more genes in samples obtained from a patient group in which the ICI therapy is effective and a patient group in which the ICI therapy is not effective, respectively, and information on the therapeutic effectiveness of the ICI,

[0030] an increase in a prediction result of the therapeutic effectiveness of the ICI for the first sample and the second sample indicates that the candidate substance is likely to improve the therapeutic effectiveness of the ICI, and a decrease in the prediction result indicates that the candidate substance is likely to reduce the therapeutic effectiveness of the ICI.

[0031] (15) The method according to the above (14), wherein the patient is a patient before the ICI therapy.

[0032] (16) The method according to the above (14), wherein the patient is a patient during the ICI therapy.

[0033] (17) The method according to any one of the above (14) to (16), wherein the candidate substance has an anti-tumor effect.

[0034] ​(18) The method according to any one of the above (14) to (16), wherein the candidate substance has no anti-tumor effect. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1 The relationship between the prognosis of each patient for immune checkpoint inhibitor (ICI) treatment and the proportion of classical monocytes (panel A) or the proportion of non-classical monocytes (panel B) in peripheral blood mononuclear cells is shown.

[0036] Figure 2 A cell map obtained by separating peripheral blood mononuclear cells based on the expression amount of CD14 and the expression amount of CD16 is shown.

[0037] Figure 3A The results of predicting the treatment effectiveness of ICI using a prediction model constructed by logistic regression analysis according to the expression amount of 1 gene are shown.

[0038] Figure 3B The same as above.

[0039] Figure 4A The results of predicting the treatment effectiveness of ICI using a prediction model constructed by logistic regression analysis according to the expression amount of 1 gene are shown.

[0040] Figure 4B The same as above.

[0041] Figure 4C The same as above.

[0042] Figure 4D The same as above.

[0043] Figure 4E The same as above.

[0044] Figure 4F The same as above.

[0045] Figure 4G The same as above.

[0046] Figure 4H The same as above.

[0047] Figure 4I The same as above.

[0048] Figure 4J The same as above.

[0049] Figure 4K The same as above.

[0050] Figure 4L The same as above.

[0051] Figure 4M The same as above.

[0052] Figure 4NThe same applies.

[0053] Figure 4O The same applies.

[0054] Figure 4P The same applies.

[0055] Figure 4Q The same applies.

[0056] Figure 4R The same applies.

[0057] Figure 4S The same applies.

[0058] Figure 4T The same applies.

[0059] Figure 4U The same applies.

[0060] Figure 4V The same applies.

[0061] Figure 4W The same applies.

[0062] Figure 4X The same applies.

[0063] Figure 4Y The same applies.

[0064] Figure 4Z The same applies.

[0065] Figure 4AA The same applies.

[0066] Figure 4BB The same applies.

[0067] Figure 4CC The same applies.

[0068] Figure 4DD The same applies.

[0069] Figure 4EE The same applies.

[0070] Figure 4FF The same applies.

[0071] Figure 4GG The same applies.

[0072] Figure 4HH The same applies.

[0073] Figure 4II The same applies.

[0074] Figure 4JJ The same applies.

[0075] Figure 4KK The same applies.

[0076] Figure 4LL As above.

[0077] Figure 4MM As above.

[0078] Figure 4NN As above.

[0079] Figure 5 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 10 genes shown are shown.

[0080] Figure 6 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 20 genes shown are shown.

[0081] Figure 7 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 160 genes shown are shown.

[0082] Figure 8 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 160 genes shown are shown.

[0083] Figure 9 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by decision tree analysis are shown.

[0084] Figure 10 Receiver operating characteristic curve (ROC), area under the curve (AUC), and validation results and others of the prediction model constructed by neural network are shown.

[0085] Figure 11 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by random forest method are shown.

[0086] Figure 12 Receiver operating characteristic curve (ROC), area under the curve (AUC), and validation results and others of the prediction model constructed by random forest method are shown.

[0087] Figure 13 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 30 genes shown are shown.

[0088] Figure 14 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 40 genes shown.

[0089] Figure 15 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 45 genes shown.

[0090] Figure 16 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 50 genes shown.

[0091] Figure 17 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 55 genes shown.

[0092] Figure 18 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 60 genes shown.

[0093] Figure 19 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 60 genes shown.

[0094] Figure 20 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 60 genes shown.

[0095] Figure 21 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 60 genes shown.

[0096] Figure 22 Receiver operating characteristic curve (ROC) and area under the curve (AUC) and others of the prediction model constructed by logistic regression analysis based on the expression amounts of the 60 genes shown.

[0097] Figure 23A list of 17 genes showing a p-value less than 0.0001 among 45 genes extracted by elastic net regression and a receiver operating characteristic curve (ROC) and an area under the curve (AUC) of a prediction model constructed using the 17 genes are shown below, along with others. DETAILED DESCRIPTION

[0098] In the present specification, the "subject" refers to a subject of a blood test. The subject can be an animal. The subject can be a fish, a bird, a reptile, a mammal, or an amphibian. The mammal can be a primate such as a dog, a cat, a cow, a horse, a sheep, a pig, a hamster, a mouse, a squirrel, a goat, a mule, a camel, an alpaca, a llama, a chicken, a gorilla, a chimpanzee, a bonobo, or a human. The subject is preferably a human.

[0099] In the present specification, the "antibody" refers to an immunoglobulin, and means a protein that takes a structure in which two heavy chains (H chains) and two light chains (L chains) are associated by a pair of disulfide bonds. The heavy chain is composed of a heavy chain variable region VH, a heavy chain constant region CH1, CH2, CH3, and a hinge region between CH1 and CH2, and the light chain is composed of a light chain variable region VL and a light chain constant region CL. Among them, a variable region fragment (Fv) composed of VH and VL directly participates in antigen binding, and is a region that contributes to the diversity of antibodies. In addition, an antigen binding region composed of VL, CL, VH, and CH1 is referred to as a Fab region, and a region composed of a hinge region, CH2, and CH3 is referred to as an Fc region.

[0100] The region in the variable region that directly contacts the antigen varies greatly, and is referred to as a complementarity-determining region (CDR). The portion other than the CDR, in which mutations are less frequent, is referred to as a framework region (FR). There are three CDRs in the variable region of the light chain and the heavy chain, respectively, and they are referred to as heavy chain CDR1 to 3 and light chain CDR1 to 3, respectively, in order from the N-terminal side. In the present specification, the "anti-A antibody" refers to an antibody that binds to A. The anti-A antibody can specifically bind to A. In the present specification, "specifically binds to A" means that the binding affinity to A is stronger than the binding affinity to other proteins other than A.

[0101] The antibody can be a monoclonal antibody, or a polyclonal antibody. In addition, the antibody can be any of IgG, IgM, IgA, IgD, IgE. It can be an antibody produced by immunization of a non-human animal such as a mouse, a rat, a hamster, a guinea pig, a rabbit, a goat, a sheep, a horse, a cow, a donkey, a camel, an American camel, a llama, a chicken, or the like, or a recombinant antibody, or a chimeric antibody, a humanized antibody, a fully human antibody, or the like. The chimeric antibody refers to an antibody in which fragments from antibodies of different species are linked.

[0102] A "humanized antibody" refers to an antibody in which the amino acid sequences characteristic of a non-human antibody are substituted for the corresponding positions of a human antibody, and examples include an antibody having heavy chain CDR1 to 3 and light chain CDR1 to 3 of an antibody prepared by immunizing a mouse or a rat, and other regions of the heavy chain and the light chain each of which 4 framework regions (FR) are derived from a human antibody. Such an antibody is also sometimes referred to as a CDR-grafted antibody. The term "humanized antibody" sometimes also includes a human chimeric antibody.

[0103] A "human chimeric antibody" refers to an antibody in which the constant region of a non-human antibody is substituted for the constant region of a human antibody in a non-human antibody. With respect to a human chimeric antibody, from the viewpoint of improving ADCC activity, for example, the subclass of a human antibody used for the constant region can be set to IgGl.

[0104] In the present specification, an "antigen-binding fragment" refers to a fragment of an antibody having binding affinity to an antigen. Specifically, in addition to Fab composed of a VL, a VH, a CL, and a CH1 region, F(ab')2 in which 2 Fabs are connected by a disulfide bond in a hinge region, Fv composed of a VL and a VH, and scFv which is a single-chain antibody in which a VL and a VH are connected by an artificial polypeptide linker, a bispecific antibody such as a diabody type, an scDb type, a tandem scFv type, a leucine zipper type, and the like can be exemplified, but is not limited to these.

[0105] In the present specification, an "immune checkpoint inhibitor" means an agent that activates immunity by relieving the suppression of immune cells caused by an immune checkpoint molecule. As the immune checkpoint molecule, Programmed cell death-1 (PD-1), CTLA-4, T-cell immunoglobulin domain and mucin domain-3 (TIM-3), lymphocyte activation gene 3 (LAG-3), and V-type immunoglobulin domain-containing suppressor of T-cell activation (VISTA) can be exemplified. The immune checkpoint for which each is responsible is referred to as a PD-1-based immune checkpoint, a CTLA-4-based immune checkpoint, a TIM-3-based immune checkpoint, a LAG-3-based immune checkpoint, and a VISTA-based immune checkpoint. An immune checkpoint inhibitor can, for example, bind to an immune checkpoint molecule or a ligand thereof and inhibit the function of the immune checkpoint. For example, by inhibiting the binding of PD-1 to PD-L1 or PD-L2, a PD-1-based immune checkpoint can be inhibited. In addition, by inhibiting the binding of CTLA-4 to CD80 or CD86, a CTLA-4-based immune checkpoint can be inhibited. In addition, by inhibiting the binding of TIM-3 to galectin-9, a TIM-3-based immune checkpoint can be inhibited. In addition, by inhibiting the binding of LAG-3 to MHC class II molecules, a LAG-3-based immune checkpoint can be inhibited. In addition, by inhibiting the binding of VISTA to VSIG-3 / IGSF11, a VISTA-based immune checkpoint can be inhibited. Thus, one or more immune checkpoints selected from the group consisting of a PD-1-based immune checkpoint, a CTLA-4-based immune checkpoint, a TIM-3-based immune checkpoint, a LAG-3-based immune checkpoint, and a VISTA-based immune checkpoint can be inhibited. An antibody that inhibits the binding of two proteins can bind to either the receptor or the ligand. For example, an antibody that inhibits a PD-1-based immune checkpoint can be an antibody selected from the group consisting of an anti-PD-1 antibody, an anti-PD-L1 antibody, and an anti-PD-L2 antibody (e.g., nivolumab, pembrolizumab, avelumab, atezolizumab, and durvalumab). In addition, an antibody that inhibits a CTLA-4-based immune checkpoint can be an antibody selected from the group consisting of an anti-CTLA-4 antibody, an anti-CD80 antibody, and an anti-CD86 antibody (e.g., ipilimumab and tremelimumab). In addition, an antibody that inhibits a TIM-3-based immune checkpoint can be an antibody selected from the group consisting of an anti-TIM-3 antibody and an anti-galectin-9 antibody (e.g., MGB453).In addition, the antibody that inhibits the VISTA system immune checkpoint can be an antibody selected from the group consisting of an anti-VISTA antibody and an anti-VSIG-3 / IGSF11 antibody (e.g., JNJ-61610588). As the immune checkpoint inhibitor, in addition to the antibody, an antigen-binding fragment of the antibody can also be used as such.

[0106] In the present specification, "cancer" means a malignant tumor. Tumors include benign tumors and malignant tumors. As the tumor, for example, a solid tumor (e.g., an epithelial tumor, a nonepithelial tumor), a tumor in hematopoietic tissue can be cited. In more detail, as the solid tumor that can be treated with the medicament of the present application, for example, a digestive system cancer (e.g., esophageal cancer, gastric cancer, colon cancer, rectal cancer), a lung cancer (e.g., small cell carcinoma, non-small cell carcinoma), a spleen cancer, a kidney cancer, a liver cancer, a pancreatic cancer, a bile duct cancer, a thymus cancer, a thyroid cancer, an adrenal gland cancer, a prostate cancer, a bladder cancer, a ureter cancer, an ovarian cancer, a uterine cancer (e.g., endometrial cancer, cervical cancer), a bone cancer, a skin cancer, a sarcoma (e.g., Kaposi's sarcoma), a melanoma, a neuroblastoma, a blastoma (e.g., neuroblastoma), a brain tumor, a cancer of unknown primary, and a cancer caused by recurrence and metastasis of these solid tumors can be cited. As the tumor that can be treated with the medicament of the present application, in addition, an adenocarcinoma, a squamous cell carcinoma, a non-squamous cell carcinoma can be cited. As the tumor in hematopoietic tissue that can be treated with the medicament of the present application, a leukemia (e.g., acute myeloid leukemia (AML), chronic myeloid leukemia (CML), acute lymphoblastic leukemia (ALL), chronic lymphoblastic leukemia (CLL), adult T-cell leukemia lymphoma (ATL), myelodysplastic syndrome (MDS)), a lymphoma (e.g., T lymphoma, B lymphoma, Hodgkin's lymphoma, non-Hodgkin's lymphoma), a myeloma (multiple myeloma), and a cancer caused by recurrence and metastasis of these tumors can be cited.

[0107] In the present specification, a "prediction model" is made based on statistical analysis of data, and predicts the result of an event to be predicted. A prediction model of treatment effectiveness is obtained from features in a sample obtained from a subject and information on treatment effectiveness. In constructing a prediction model, the relationship between features in a sample obtained from a subject and information on treatment effectiveness is usually formulated. In a prediction model using machine learning, treatment effectiveness in a subject can be predicted using a moving average method, an exponential smoothing method, or the like. When the gene expression amount used for prediction is one gene, the prediction model can be a model that simply determines effectiveness depending on whether or not a certain reference value is exceeded. When the gene expression amount used for prediction is two or more genes, the prediction model can be, for example, a prediction model constructed by logistic regression. In one mode, treatment effectiveness can mean a case where the progression-free survival (PFS) is 2 years or more. In one mode, treatment effectiveness can mean a case where a complete response (CR), a partial response (PR), and a stable disease (SD) are shown by treatment. This classification can be performed using RECIST (for example, RECIST 1.1). In one mode, treatment ineffectiveness can mean a case where a progression (PD) is shown for treatment. However, effectiveness or ineffectiveness can be defined otherwise as appropriate by those skilled in the art. For example, only CR can be evaluated as effective, or CR and PR can be evaluated as effective. In one embodiment, treatment ineffectiveness can mean a case where the progression-free survival (PFS) is less than 6 months. In one mode, treatment ineffectiveness can be a case other than the effective case. In the case of solid tumors and hematological tumors, the definition of treatment effect can be different.

[0108] In the present specification, "peripheral blood mononuclear cells" (PBMCs) are mononuclear cells isolated from the peripheral blood of a subject, and typically include monocytes, lymphocytes. PBMCs can usually be obtained by removing plasma components, red blood cells, platelets, granulocytes, and the like from collected peripheral blood.

[0109] In the present specification, the prediction can be performed using a prediction model in which the area under the curve (AUC) of a receiver operating characteristic curve (ROC) with respect to the therapeutic effectiveness of the ICI is greater than 0.5. The ROC can be plotted on a two-dimensional graph in which the vertical axis is set as the sensitivity and the horizontal axis is set as the 1-specificity (false positive rate). The prediction model is preferably constructed in such a manner that the sensitivity is > (1-specificity) at any false positive rate. As the prediction model, a prediction model based on a simple threshold, a prediction model based on a weighted score, a prediction model based on a logistic regression analysis, a prediction model based on machine learning (a prediction model based on a neural network, a prediction model based on a decision tree analysis, a prediction model based on a random forest method, a prediction model based on a logistic regression analysis, and the like) can be cited. By inputting the data of the gene expression amount in the peripheral blood or the PBMC into the obtained prediction model, a determination result with respect to the therapeutic effectiveness of the ICI can be obtained. The prediction model can be a model in which a reference value is set, and the prediction is performed by comparison with the reference value. The prediction model can be a model of a weighted score, that is, a model in which the expression amount of each gene (or a value obtained by standardizing the expression amount) is multiplied by a weighted coefficient and added to perform the scoring. The prediction model can be a model in which the scoring is compared with the reference value, and the effectiveness is predicted therefrom.

[0110] Examples of the scoring in the prediction model.

[0111]

[0112] In the above (Formula 1), Ex t is the expression amount of the tthgene among n kinds of genes to be measured or a value obtained by standardizing or normalizing the expression amount, A t is a weighted coefficient with respect to the expression amount of the tthgene, and S is a score. The weighted coefficient can be a value of + or -. The standardization indicates scaling in such a manner that the average value of the data is set to 0 and the variance is set to 1. Specifically, the standardized score is obtained by (expression amount - average value) / variance. The normalization indicates scaling in such a manner that the minimum value is set to 0 and the maximum value is set to 1. Specifically, the normalized score is obtained by (expression amount - minimum value) / (maximum value - minimum value).

[0113] The reference value can be determined by various methods. In a case where the prediction of the effectiveness is high when the expression amount or the score exceeds the reference value, the specificity increases but the sensitivity decreases as the reference value increases, and, on the contrary, the specificity decreases and the sensitivity increases as the reference value decreases. Therefore, the reference value can be appropriately determined depending on the purpose, the situation, and the like, and the preferable reference value can be changed from time to time. For example, the reference value can be increased if it is desired to limit the administration of the ICI to as few patients as possible, and the reference value can be decreased if it is desired to widely administer the ICI to as many patients as possible.

[0114] In one mode, in the case where effectiveness is predicted to be high when the expression amount or score exceeds the first reference value, the first reference value can be set to any one value selected from the group consisting of the average value, the top 40%, the top 33%, the top 25%, the top 20%, the top 15%, the top 10%, and the top 5% of the corresponding expression amount or score in the patient group in which ICI treatment is ineffective, and a value between two of these. In addition, the first reference value can be any one value selected from the group consisting of the average value, the top 60%, the top 70%, the top 80%, and the top 90% of the corresponding expression amount or score in the patient group in which ICI treatment is effective, and a value between two of these. In one mode, the first reference value can be a value between the average value in the patient group in which ICI treatment is effective and the average value in the patient group in which ICI treatment is ineffective.

[0115] In one mode, in the case where effectiveness is predicted to be high when the expression amount or score is lower than the second reference value, the second reference value can be set to any one value selected from the group consisting of the average value, the top 40%, the top 33%, the top 25%, the top 20%, the top 15%, the top 10%, and the top 5% of the corresponding expression or score in the patient group in which ICI treatment is effective. In addition, the second reference value can be any one value selected from the group consisting of the average value, the top 60%, the top 70%, the top 80%, and the top 90% of the corresponding expression amount or score in the patient group in which ICI treatment is ineffective, and a value between two of these. In one mode, the second reference value can be a value between the average value in the patient group in which ICI treatment is effective and the average value in the patient group in which ICI treatment is ineffective.

[0116] In one mode, in the case where effectiveness is predicted to be low when the expression amount or score exceeds the third reference value, the third reference value can be set to any one value selected from the group consisting of the average value, the top 40%, the top 33%, the top 25%, the top 20%, the top 15%, the top 10%, and the top 5% of the corresponding expression amount or score in the patient group in which ICI treatment is effective. In addition, the third reference value can be set to any one value selected from the group consisting of the average value, the top 60%, the top 70%, the top 80%, and the top 90% of the corresponding expression amount or score in the patient group in which ICI treatment is ineffective, and a value between two of these. In one mode, the third reference value can be a value between the average value in the patient group in which ICI treatment is effective and the average value in the patient group in which ICI treatment is ineffective.

[0117] In one aspect, the fourth reference value can be set to any value selected from the group consisting of the average value, the top 40%, the top 33%, the top 25%, the top 20%, the top 15%, the top 10%, and the top 5% of the corresponding expression amount or score in the patient group in which ICI treatment is ineffective, and a value between any two of these. In addition, the fourth reference value can be set to any value selected from the group consisting of the average value, the top 60%, the top 70%, the top 80%, and the top 90% of the corresponding expression amount or score in the patient group in which ICI treatment is effective, and a value between any two of these. In one aspect, the fourth reference value can be a value between the average value in the patient group in which ICI treatment is effective and the average value in the patient group in which ICI treatment is ineffective.

[0118] In this way, by comparing the expression amount or score with the reference value, the responsiveness of the subject to ICI treatment and the therapeutic effectiveness of ICI in the subject can be evaluated. A subject predicted to be effective for ICI treatment in one method can sometimes be predicted to be ineffective (low effectiveness) for ICI treatment in another method. In such a case, it can be determined by a doctor or the like whether or not to perform ICI treatment on the subject depending on the situation.

[0119] According to the present application, there is provided a method of predicting the tumor therapeutic effect of an immune checkpoint inhibitor (ICI) in a subject. According to the present application, there is provided a method for predicting the tumor therapeutic effect of an immune checkpoint inhibitor (ICI) in a subject. According to the present application, there is provided a method for predicting the tumor therapeutic effect of an immune checkpoint inhibitor (ICI) in a subject {wherein the method does not include medical treatment of a human}. In one aspect, the prediction method of the present application can predict the therapeutic effect using a prediction model on an electronic computer. The prediction method of the present application can be an industrially useful method. The prediction method of the present application belongs to in vitro diagnostics. The method of the present application can be performed by either or both of in vitro and in silico. The method of the present application can comprise determining the expression amount of a gene comprising one or more genes selected from the group of genes described in Table 1 in a test sample obtained from the above-mentioned subject. In one preferred aspect, the test sample contains monocytes.

[0120] The prediction method of the present application can include determining the expression amount of a gene comprising one or more genes selected from the group of genes described in Table 1 in a sample obtained from the above subject, particularly a sample containing monocytes, such as a sample containing blood cells, such as peripheral blood mononuclear cells, such as monocytes (e.g., classical monocytes and non-classical monocytes, and intermediate monocytes). Blood cells, such as peripheral blood mononuclear cells, such as monocytes (e.g., classical monocytes and non-classical monocytes, and intermediate monocytes) are contained in peripheral blood. Blood cells, such as peripheral blood mononuclear cells, such as monocytes (e.g., classical monocytes and non-classical monocytes, and intermediate monocytes) are preferably isolated from peripheral blood before the expression amount is determined. The expression amount of the gene is preferably the expression amount of messenger RNA (mRNA). The expression amount of mRNA can be quantified using conventional methods. The expression amount of mRNA can be determined, for example, using various methods such as quantitative PCR, quantitative RT-PCR, RNAseq based on next-generation sequencing (NGS), etc. Typically, mRNA can be determined as follows: after cell lysis or disruption and total RNA extraction, cDNA is synthesized using mRNA as a template, and the amount of mRNA is determined based on the amount of synthesized cDNA on the premise that the amount of synthesized cDNA reflects the amount of mRNA. When the expression amount of mRNA is correlated with the expression amount of protein, the amount of protein can also be determined as the expression amount of the gene. The expression amount of protein can be determined using various methods such as ELISA, Western blotting, dot blotting, etc. The expression of mRNA can be determined at the single cell level. For the determination of the expression amount of mRNA at the single cell level, cDNA can be synthesized after using primers having a sequence unique to each cell (barcode sequence). In addition, the primers can contain a sequence unique to each patient (index sequence). The barcode sequence and the index sequence can be designed and used as described in US9260753B.

[0121] The prediction method of the present application can include predicting the therapeutic effect of ICI based on the determined expression amount using a prediction model constructed in a manner that enables prediction of the therapeutic effect of ICI. The prediction method of the present application can include outputting a prediction result of the therapeutic effect of ICI based on the determined expression amount using a prediction model constructed in a manner that enables prediction of the therapeutic effect of ICI. The prediction model can be constructed based on the expression amount of a gene comprising one or more genes described above in peripheral blood mononuclear cells obtained from a patient group in which ICI treatment is effective and a patient group in which ICI treatment is not effective, and information on the therapeutic effectiveness of ICI. When training data and / or validation data are prepared, whether or not ICI treatment is effective in a particular patient can be determined by a medical practitioner such as a doctor. This can be performed on a display, on paper, in a recording medium (particularly a volatile memory or preferably a non-volatile memory), or by an online method to another device as a communication destination.

[0122] Thus, according to the present application, there is provided a method of predicting a tumor therapeutic effect of an immune checkpoint inhibitor (ICI) in a subject, comprising:

[0123] determining an expression amount of a gene comprising one or more genes selected from the group of genes described in any one of Tables 1 to 7 in blood cells, preferably peripheral blood mononuclear cells, more preferably mononuclear cells, obtained from the subject; and

[0124] predicting the therapeutic effect of the ICI from the determined expression amount using a prediction model constructed in a manner capable of predicting the therapeutic effect.

[0125] According to the present application, there is additionally provided a method of predicting a tumor therapeutic effect of an immune checkpoint inhibitor (ICI) in a subject, comprising:

[0126] inputting into an electronic computer an expression amount of a gene comprising one or more genes selected from the group of genes described in any one of Tables 1 to 7 in blood cells, preferably peripheral blood mononuclear cells, more preferably mononuclear cells, obtained from the subject; and

[0127] predicting the therapeutic effect of the ICI from the expression amount inputted into the electronic computer using a prediction model constructed in a manner capable of predicting the therapeutic effect, and outputting the result thereof.

[0128] The prediction model can be constructed based on the expression amount of the gene comprising the one or more genes in peripheral blood mononuclear cells (more preferably mononuclear cells) obtained from a patient group in which the ICI treatment is effective and information on the therapeutic effectiveness of the ICI. As the prediction model, various prediction models can be used as described above. The prediction model can function on an electronic computer. As the prediction model, a prediction model having an area under the curve (AUC) of a receiver operating characteristic curve (ROC) for the therapeutic effectiveness of the ICI of 0.6 or more, 0.65 or more, 0.7 or more, 0.75 or more, 0.8 or more, 0.9 or more, 0.95 or more, 0.96 or more, 0.97 or more, 0.98 or more, or 0.99 or more can be preferably used. The more the AUC approaches 1, the higher the sensitivity and the fewer the false positives. In addition, as the prediction model, a prediction model having an area under the curve (AUC) of a receiver operating characteristic curve (ROC) for the therapeutic ineffectiveness of the ICI of 0.6 or more, 0.65 or more, 0.7 or more, 0.75 or more, 0.8 or more, 0.9 or more, 0.95 or more, 0.96 or more, 0.97 or more, 0.98 or more, or 0.99 or more can be preferably used. The more the AUC approaches 1, the higher the sensitivity and the fewer the false positives.

[0129] The genes to be determined for the expression amount can include 1 or more, 2 or more, 3 or more, 4 or more, 5 or more, 10 or more, 15 or more, 20 or more, 25 or more, 30 or more, 35 or more, 40 or more, 45 or more, 50 or more, 55 or more, 60 or more, 65 or more, 70 or more, 75 or more, 80 or more, 85 or more, 90 or more, 95 or more, 100 or more, 110 or more, 120 or more, 130 or more, 140 or more, 150 or more, or 160 genes selected from the group of genes described in Table 1. In general, the more the number of genes selected from the group of genes described in Table 1, the more preferable. Figures 4A to 4NN The genes described in Table 1 are genes that can construct a prediction model by evaluation using any one of the genes. Therefore, it is possible to use only one of these genes for prediction, or it is possible to construct a prediction model by combining two or more of these genes. It is preferable to input the expression amounts of all genes used to construct the prediction model into the prediction model thus constructed. It is possible to extract a gene set with high prediction accuracy from the group of genes described in Table 1 using various methods. As such methods, there is no particular limitation, and methods such as stepwise method, LASSO regression, and elastic net regression can be cited.

[0130] The genes selected from the group of genes described in Table 1 preferably include CLEC12A.

[0131] The genes selected from the group of genes described in Table 1 preferably include TMEM176A, for example, CLEC12A and TMEM176A.

[0132] The genes selected from the group of genes described in Table 1 preferably include TMEM176B, for example, CLEC12A and TMEM176B, or CLEC12A, TMEM176A, and TMEM176B.

[0133] The genes selected from the group of genes described in Table 1 preferably include 1 or more, 2 or more, 3 or more, 4 or more, 5 or more, or all of the genes selected from the group consisting of TMEM176A, CLEC12A, S100B, CD52, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, and LST1. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.86 or more, 0.87 or more, 0.88 or more, or 0.89 or more.

[0134] The genes selected from the group of genes described in Table 1 preferably include one or more, two or more, three or more, four or more, five or more, ten or more, fifteen or more, or all of the genes selected from the group consisting of TMEM176A, CLEC12A, S100B, CD52, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, LST1, SH2B3, AIF1, LAIR2, RETN, LYPD2, RNF14, UBE2F, CTSS, FGL2, and HAVCR2. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.9 or more, 0.91 or more, 0.92 or more, 0.93 or more, or 0.94 or more.

[0135] The genes selected from the group of genes described in Table 1 preferably include one or more, two or more, three or more, four or more, five or more, ten or more, fifteen or more, twenty or more, twenty-five or more, or all of the genes selected from the group consisting of CLEC12A, TMEM176A, CD52, S100B, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, LST1, SH2B3, AIF1, LAIR2, RETN, LYPD2, DIP2A, RNF14, UBE2F, CTSS, FGL2, HAVCR2, LILRB2, TMEM176B, CLU, SELL, PHTF2, PSAP, SLA2, ID2, and HCK. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.9 or more, 0.91 or more, 0.92 or more, 0.93 or more, 0.94 or more, 0.95 or more, or 0.96 or more.

[0136] The genes selected from the group of genes described in Table 1 preferably include 1 or more, 2 or more, 3 or more, 4 or more, 5 or more, 10 or more, 15 or more, 20 or more, 30 or more, or all of the genes selected from the group consisting of CLEC12A, TMEM176A, CD52, S100B, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, LST1, SH2B3, AIF1, LAIR2, RETN, LYPD2, DIP2A, RNF14, UBE2F, CTSS, FGL2, HAVCR2, LILRB2, TMEM176B, CLU, SELL, PHTF2, PSAP, SLA2, ID2, HCK, EGR1, CD247, GNLY, SH2D1B, PRF1, IL1RN, PECAM1, KIR2DL3, UBE2G2, and C1QA. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.9 or more, 0.91 or more, 0.92 or more, 0.93 or more, 0.94 or more, 0.95 or more, 0.96 or more, or 0.97 or more.

[0137] The genes selected from the group of genes described in Table 1 preferably include 1 or more, 2 or more, 3 or more, 4 or more, 5 or more, 10 or more, 15 or more, 20 or more, 30 or more, 40 or more, or all of the genes selected from the group consisting of CLEC12A, TMEM176A, CD52, S100B, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, LST1, SH2B3, AIF1, LAIR2, RETN, LYPD2, DIP2A, RNF14, UBE2F, CTSS, FGL2, HAVCR2, LILRB2, TMEM176B, CLU, SELL, PHTF2, PSAP, SLA2, ID2, HCK, EGR1, CD247, GNLY, SH2D1B, PRF1, IL1RN, PECAM1, KIR2DL3, UBE2G2, C1QA, NR4A2, HK2, XRCC5, CXCR4, LYST, CSF1R, ANKRD28, WARS, HBEGF, and CAT. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.9 or more, 0.91 or more, 0.92 or more, 0.93 or more, 0.94 or more, 0.95 or more, 0.96 or more, or 0.97 or more.

[0138] The gene selected from the group of genes described in Table 1 preferably includes one or more, two or more, three or more, four or more, five or more, ten or more, fifteen or more, twenty or more, thirty or more, forty or more, fifty or more, or all of the genes selected from the group consisting of CLEC12A, TMEM176A, CD52, S100B, SH2D2A, SIDT2, GTF3C1, ZNF266, MS4A7, LST1, SH2B3, AIF1, LAIR2, RETN, LYPD2, DIP2A, RNF14, UBE2F, CTSS, FGL2, HAVCR2, LILRB2, TMEM176B, CLU, SELL, PHTF2, PSAP, SLA2, ID2, HCK, EGR1, CD247, GNLY, SH2D1B, PRF1, IL1RN, PECAM1, KIR2DL3, UBE2G2, C1QA, NR4A2, HK2, XRCC5, CXCR4, LYST, CSF1R, ANKRD28, WARS, HBEGF, CAT, CYBB, CTSL, PRMT2, SLC39A8, FCGR3A, GPR183, STAB1, CASP1, EGR2, and FOLR3. The AUC of the prediction model at this time can be, for example, 0.6 or more, 0.7 or more, 0.8 or more, 0.85 or more, 0.9 or more, 0.91 or more, 0.92 or more, 0.93 or more, 0.94 or more, 0.95 or more, 0.96 or more, 0.97 or more, or 0.98 or more.

[0139] Thus, by inputting the expression amount of the corresponding gene into the obtained prediction model, the therapeutic effect of the ICI can be predicted from the determined expression amount. Based on the description and technical knowledge of the present specification, a person skilled in the art can easily construct a prediction model in a manner that can predict the therapeutic effect of the ICI from the expression amount of the above-mentioned gene.

[0140] In one mode, the ICI can be, for example, an inhibitor of an immune checkpoint selected from the group consisting of a PD-1 system immune checkpoint, a CTLA-4 system immune checkpoint, a TIM-3 system immune checkpoint, a LAG-3 system immune checkpoint, and a VISTA system immune checkpoint.

[0141] The ICI, for example, can bind to an immune checkpoint molecule or a ligand thereof and inhibit the function of the immune checkpoint. For example, by inhibiting the binding of PD-1 to PD-L1 or PD-L2, a PD-1 system immune checkpoint can be inhibited. In addition, by inhibiting the binding of CTLA-4 to CD80 or CD86, a CTLA-4 system immune checkpoint can be inhibited. In addition, by inhibiting the binding of TIM-3 to galectin-9, a TIM-3 system immune checkpoint can be inhibited. In addition, by inhibiting the binding of LAG-3 to MHC class II molecules, a LAG-3 system immune checkpoint can be inhibited. In addition, by inhibiting the binding of VISTA to VSIG-3 / IGSF11, a VISTA system immune checkpoint can be inhibited. Thus, one or more immune checkpoints selected from the group consisting of a PD-1 system immune checkpoint, a CTLA-4 system immune checkpoint, a TIM-3 system immune checkpoint, a LAG-3 system immune checkpoint, and a VISTA system immune checkpoint can be inhibited. The antibody that inhibits the binding of two proteins can bind to a receptor or a ligand. For example, the antibody that inhibits a PD-1 system immune checkpoint can be an antibody selected from the group consisting of an anti-PD-1 antibody, an anti-PD-L1 antibody, and an anti-PD-L2 antibody (e.g., nivolumab, pembrolizumab, avelumab, atezolizumab, and durvalumab). In addition, the antibody that inhibits a CTLA-4 system immune checkpoint can be an antibody selected from the group consisting of an anti-CDLA-4 antibody, an anti-CD80 antibody, and an anti-CD86 antibody (e.g., ipilimumab and tremelimumab). In addition, the antibody that inhibits a TIM-3 system immune checkpoint can be an antibody selected from the group consisting of an anti-TIM-3 antibody and an anti-galectin-9 antibody (e.g., MGB453). In addition, the antibody that inhibits a VISTA system immune checkpoint can be an antibody selected from the group consisting of an anti-VISTA antibody and an anti-VSIG-3 / IGSF11 antibody (e.g., JNJ-61610588).

[0142] In one preferred embodiment, the blood cells are peripheral blood mononuclear cells. In one preferred embodiment, the blood cells are monocytes. In one preferred embodiment, the blood cells are classical monocytes. Classical monocytes are characterized by CD14 + CD16 低 In one preferred embodiment, the blood cells are non-classical monocytes. Non-classical monocytes are characterized by CD14 低 CD16 + In one preferred embodiment, the blood cells are intermediate monocytes. Intermediate monocytes are characterized by CD14 + CD16 + Expression of CD14 and CD16 of classical monocytes, non-classical monocytes, and intermediate monocytes is shown in Figure 2The blood cells are preferably non-classical monocytes, and the measured gene is any one or more of the genes listed in Table 3. The blood cells are preferably a mixture (pool) of non-classical monocytes and classical monocytes, and the measured gene is any one or more of the genes listed in Table 4. The blood cells are preferably intermediate monocytes, and the measured gene is any one or more of the genes listed in Table 5. The blood cells are preferably classical monocytes, and the measured gene is any one or more of the genes listed in Table 6. The monocytes are preferably monocytes of the monocyte types listed in Table 7, and the measured gene is the gene of the corresponding row. These blood cells are isolated, concentrated, or purified from blood.

[0143] Thereby, the ICI treatment effectiveness on the subject can be predicted.

[0144] The subject determined to be effective for ICI treatment can be a preferred treatment target of ICI. Therefore, according to the present application, there is provided a method of treating a tumor in a subject determined to be effective for ICI treatment, which includes the step of administering a therapeutically effective amount of ICI to the subject. In contrast, a subject determined to be ineffective for ICI treatment or a subject not determined to be effective can not be a preferred treatment target of ICI. Therefore, such a subject can be provided with a tumor treatment other than ICI.

[0145] In the present application, there is provided the use of a tool for checking gene expression in the manufacture of a kit for the above-mentioned prediction method. The checking of gene expression can be performed using a quantitative technique of mRNA such as a microarray, quantitative PCR, and RNASeq based on next-generation sequencing. As a tool for checking gene expression, for example, a reagent capable of measuring the expression amount of mRNA can be listed. As a reagent capable of measuring the expression amount of mRNA, for example, a primer for nucleic acid amplification and a probe (for example, a nucleic acid probe) that binds to mRNA can be listed. The tool for checking gene expression can be a tool suitable for each use of the quantitative technique of mRNA. Specifically, the probe can be provided in the form of a microarray by being immobilized on an array such as a microarray, or can be provided in the form of a conjugate with a fluorescent reagent. In addition, the checking of gene expression can be performed using an evaluation method known to those skilled in the art such as ELISA. As a tool for checking gene expression, for example, an antibody that binds to each protein, a labeled secondary antibody that recognizes the antibody, and a reagent that develops the label can be listed.

[0146] In the present application, there is provided the use of ICI in the manufacture of a medicament for the above-mentioned treatment method.

[0147] In the present application, there is provided ICI for the above-mentioned treatment method. In the present application, there is provided a pharmaceutical composition containing ICI for the above-mentioned treatment method.

[0148] The prediction model made based on the information on the effectiveness of the specific ICI can be preferably used to predict the effectiveness of the specific ICI. Thus, the specific ICI can be preferably administered to a patient determined to be effective for treatment with the specific ICI. However, the prediction model can also be made based on the information on the effectiveness of a plurality of ICIs, and the prediction model made can be used to predict the therapeutic effectiveness of the plurality of ICIs.

[0149] The prediction model made based on the expression amount of the specific gene preferably predicts the therapeutic effectiveness of the ICI based on the expression amount of the specific gene. In addition, the prediction model made based on the expression amounts of a plurality of specific genes preferably predicts the therapeutic effectiveness of the ICI based on the expression amounts of the plurality of specific genes.

[0150] In one mode, the peripheral blood mononuclear cells can be isolated mononuclear cells. In one mode, the mononuclear cells can be classical mononuclear cells. In one mode, the mononuclear cells can be non-classical mononuclear cells. In one mode, the mononuclear cells can be mononuclear cells having properties between classical mononuclear cells and non-classical mononuclear cells (e.g., CD14-positive CD16-positive).

[0151] In the present application, the method of predicting the therapeutic effect of the ICI can include the step of determining the proportion of classical mononuclear cells in peripheral blood mononuclear cells. In the present application, the method of predicting the therapeutic effect of the ICI can include the step of determining the proportion of non-classical mononuclear cells in peripheral blood mononuclear cells. This is because the smaller the proportion of classical mononuclear cells, the better the prognosis, and the larger the proportion of non-classical mononuclear cells, the better the prognosis.

[0152] According to the present application, there is provided a device for predicting the therapeutic effect of an ICI in a subject. The device for predicting the therapeutic effect of an ICI of the present application comprises one or more processors and one or more memories electrically connected to the one or more processors in an operable manner, the memories having commands stored therein,

[0153] The above commands are executed by at least one of the above one or more processors,

[0154] receiving information on the expression amount of one or more genes described in any one of Tables 1 to 7 in peripheral blood mononuclear cells obtained from the above subject, processing the received expression amount information using a prediction model, and outputting a prediction result of the therapeutic effect of the ICI,

[0155] The prediction model is constructed in a manner that can predict the treatment effectiveness based on information on the above treatment effectiveness of patients in each of a patient group in which ICI treatment is effective and a patient group in which ICI treatment is not effective and information on the expression amount of the above one or more genes of the patient associated with the information. According to the present application, there is further provided an application of the above device for predicting the treatment effect of ICI in a subject. According to the present application, there is further provided a method of predicting the treatment effect of ICI in a subject using the above device. The prediction model may, for example, have an AUC of 0.6 or more, 0.7 or more, 0.8 or more, 0.9 or more, 0.95 or more, 0.96 or more, 0.97 or more, or 0.98 or more. The prediction model is not particularly limited and may, for example, be a prediction model based on a simple threshold, a prediction model based on a weighted score, a prediction model based on logistic regression analysis, a prediction model based on machine learning (according to a prediction model based on a neural network, a prediction model based on decision tree analysis, a prediction model based on a random forest method, a prediction model based on logistic regression analysis, etc.).

[0156] Examples

[0157] Materials and methods

[0158] Details of experimental model and subjects

[0159] Peripheral blood samples were obtained from lung cancer patients (n = 8) before treatment using BD Vacutainer (trademark) CPT TM tubes (BD Biosciences). Cases that were pathologically diagnosed as lung cancer, received 4 or more cycles of anti-PD-1 antibody therapy (intravenous injection of nivolumab (240 mg) every 2 weeks or intravenous injection of pembrolizumab (200 mg) every 3 weeks) at Showa University Hospital from January 2017 to April 2019.

[0160] Progression-free survival (PFS) was defined as the period from anti-PD-1 antibody treatment until the disease worsened or the patient died for some reason. Responder (n = 4) was characterized as a patient who showed complete response or partial response by computed tomography (CT) and obtained PFS for 2 years or more. Non-responder (n = 4) refers to a patient who did not confirm objective tumor response and PFS was less than 6 months.

[0161] Isolation of peripheral blood mononuclear cells (PBMCs)

[0162] Whole blood was collected into 8 mL CPT vacutainer tubes with sodium heparin. The tubes were inverted several times to homogenize the heparin anticoagulant and blood. After blood collection, the tubes were spun to separate the sample into layers according to the following protocol. The upper layer was a suspension of plasma and PBMCs, the middle layer was polyester resin, and the lower layer was red blood cells and granulocytes. After centrifugation, the CPT was gently inverted several times to resuspend the PBMCs and plasma, and decanted into a sterile 50 mL conical tube. Phosphate buffered saline (PBS, Thermo Scientific) was added to adjust the final volume to 50 mL. The capped 50 mL conical tube was inverted and mixed, and centrifuged at 250 x g for 10 minutes at room temperature. The supernatant was carefully aspirated, and the cell pellet was gently resuspended in 10 mL of fresh PBS, followed by centrifugation at 250 x g for 10 minutes at room temperature. After this centrifugation step, the supernatant was again carefully aspirated without disturbing the cell pellet, and the pelleted PBMCs were resuspended in 3 mL of fetal bovine serum (FBS, HyClone). 500 μL of cells were aliquoted into 6 2 mL cryovials pre-filled with 500 μL of BAMBANKER (NIPPON GENETICS). The cryovials were stored at -80 °C for 24 hours in a BICELL (NIPPON FREEZER), and then transferred to long-term storage in liquid nitrogen. The remaining volume of PBMCs in FBS was retained for counting and viability evaluation.

[0163] Sorting of monocytes

[0164] PBMCs were stained with the following antibodies at 4 °C for 30 minutes. CD14-BV650 (#563419, BD Biosciences) and CD16-FITC (#555406, BD Biosciences) were used for staining at 4 °C for 30 minutes. FACS sorting of PBMCs was performed using a CytoFLEX SH800S (Sony). Then, the stained cells were gated for CD14 高 CD16 - (classical; classical) and CD14 低 CD16 + (non-classical; non-classical) subsets using conventional flow cytometry, and the gated cells were FACS sorted. The sorted cells were suspended in Bambanker and frozen at -80 °C.

[0165] Preparation of single cell suspension

[0166] After thawing the frozen cells, the viability and density of the cells were confirmed. The cells were diluted to a density of 4 x 10 4

[0167] Long single-cell RNA-seq with ICELL8 system

[0168] Takara IC ELL8 5,184 nanopore chips (350V chip) were used with the Long SMART-Seq ICELL8 kit. The cell suspension was fluorescently labeled by dead-live staining, Hoechst 33342, propidium iodide (NucBlue Cell Stain Reagent, Thermo Fisher Scientific) for 20 minutes at 37°C before dispensing into the 350V chip. The wells containing single cells and live cells were visualized and selected using the CellSelect software (Takara Bio). By this procedure, 302 cells containing single nuclei from all patients without omission were obtained, respectively. Then, the SMART-seq2 protocol 1 Preparation of single-cell cDNA and sequencing library. The long cDNA was synthesized by one-step RT-PCR reaction with barcoded primers (UMI) with poly(dT) ends. In addition to the Terra polymerase and reaction buffer, P5 index primers required for the subsequent library preparation were dispensed into all wells with different indexes. Transposase and reaction buffer (Tn5 mix) were dispensed into the selected wells. P7 index primers were dispensed into the wells. The Illumina library was amplified, and the material extracted from the chip was mixed. The mixed library was purified, and size selection was performed using Agencourt AMPure XP magnetic beads (Beckman Coulter) to achieve an average library size of 500 bp. The library was quality checked using the Agilent 2200 TapeStation. The final library was sequenced using the Novaseq 6000 system (Illumina) with 150 pairs of end cycles, and each cell obtained an average of about 20 million reads.

[0169] Preprocessing of data from IC ELL8

[0170] ​For raw sequencing files (bcl), Illumina bcl2fastq software (v2.20.0.422) was used to convert to a fastq file for each method. Each fastq file was de-duplicated and analyzed using Cogent NGS analysis pipeline (Cogent AP, version 1.5, Takara Bio inc.). Cogent demux assigned reads to cells based on cell barcodes recorded in the Well-list. Subsequently, preliminary analysis was performed by “cogent analyze” package function, which included: (1) read trimming with cutadapt 2 (version 3.4), (2) genome alignment with STAR 3 (version 2.7.8a) to human genome GRCh38. (3) read counting of exons, genome, mitochondrial regions of human genes from ENSEMBL gene annotation version 103 (https: / / www.ensembl.org / Homo_sapiens / Info / Index) using featureCounts 4 (version 2.0.0), (4) gene reads were summarized in gene matrix by the number of reads expressed in each cell per gene. For the raw gene matrix, the following parameters were used for quality control (QC) filtering of cells and genes: (a) for cells, only cells with at least 10,000 reads associated with at least 300 different genes; (b) for genes, only genes that were aligned by at least 100 reads from at least 3 different cells.

[0171] Gene marker detection

[0172] To determine gene markers expressed in the samples of Responder and Non-responder, Wilcoxon rank sum and signed rank test were used to determine differentially expressed genes between cells of Responder and Non-responder samples using the QC filtered read count matrix as input, and p-value was adjusted by Benjamini-Hochberg method. For each differential expression result, the rank score of each gene was calculated using the formula of rank = -log 10 (pval) * log2FC. Here, pval is the original p-value, and log2FC is log2 fold change.

[0173] Unsupervised cell clustering

[0174] Dimensionality reduction and visualization were performed by the Uniform Manifold Approximation and Projection (UMAP) algorithm using the best principal components (PCs) selected from elbow plots using the Cogent NGS discovery software v1.5 (CogentDS) R package by Subio. Further, unsupervised graph clustering to classify cells into specific clusters was performed using CogentDS. For each graph cluster, gene markers were detected. For the detected gene sets, plots were made in a heatmap.

[0175] Differential analysis of gene expression

[0176] Expression differences (DE) of each gene between various single cell groups specified herein were tested by non-parametric Wilcoxon rank-sum test implemented by CogentDS. When not specifically specified, each comparison was limited to genes showing an estimated log2 fold change (log_fc > +0.5 and detectable expression in more than 10% of cells of one of the two groups compared. The Benjamini-Hochberg method was used to calculate false discovery rate (FDR)-adjusted p-values (padj) for correction of multiple hypothesis testing. Differentially expressed genes were defined by FDR-adjusted p-value (padj) < 0.05 and |log2(fold change)| > 0.5. Genes detected with less than 30 cells and count < 10 were removed. Note that padj = P-value * N / i was used to calculate {here, N represents the number of genes, and i represents the rank of low p-value.}. Clustered heatmaps of gene expression were made using the Subio platform (Subio Inc., Aichi, Japan). Prediction models were constructed using a logistic regression (LR) classifier using JMP (version 16.0, SAS Institute Inc., Cary, NC, USA).

[0177] Statistical analysis

[0178] Statistical analysis was performed using JMP (version 16.0, SAS Institute Inc., Cary, NC, USA). The threshold for statistical significance was defined as p-value < 0.05.

[0179] Results

[0180] CD14 高 CD16 - (classic monocytes) and CD14 低 CD16 +(classical monocytes) and the proportion of non-classical monocytes in PBMCs were investigated, respectively. The relationship between each of the proportions thus obtained and overall survival (OS) of patients who received treatment with an immune checkpoint inhibitor was plotted in a graph (refer to Figure 1 A and B). As shown in Figure 1 , it was found that the fewer the classical monocytes, the better the OS, and the more the non-classical monocytes, the better the OS.

[0181] PBMCs were sorted by expression of CD14 and CD16, and a cell map was obtained with the expression intensity of CD16 on the vertical axis and the expression intensity of CD14 on the horizontal axis (refer to Figure 2 ). As shown in Figure 2 , monocytes can be roughly divided into CD14 高 CD16 - (classical monocytes) and CD14 低 CD16 + (non-classical monocytes), and intermediate monocytes showing intermediate properties therebetween. The intermediate monocytes can be obtained as a subgroup of a group of monocytes considered to be non-classical monocytes, and are clearly different from the classical monocytes.

[0182] PBMCs were recovered from 4 patients who responded to an immune checkpoint inhibitor (responders: patients with progression-free survival (PFS) of 2 years or more; CR) and 4 patients who did not respond (non-responders: patients with progression-free survival (PFS) of less than 6 months; PD), and classical monocytes and non-classical monocytes were obtained from each of the PBMCs for single-cell RNA-Seq analysis. There were 41 million reads (about 72%) that were uniquely mapped. It was thus considered that the quality of sequence interpretation was not a problem.

[0183] 160 genes that showed differential expression in responders and non-responders were selected. The selected genes are genes expressed in monocytes, and are summarized in Table 1. In Table 1, the gene description, gene name, NCBI accession number, and reference number of each gene in this specification are listed in the corresponding columns. In other tables, each gene is identified with reference to the reference number in Table 1. In addition, Tables 2 to 7 show genes that showed differential expression in each cell type in responders (CR) and non-responders (PD). Gene expression was normalized by the Count per million (CPM) method. When the median of the normalized genes is higher than the background, it is indicated as CR, and when it is lower than the background, it is indicated as PD.

[0184] It was confirmed whether each of the 160 genes could be used to distinguish responders and non-responders by logistic regression analysis. This is represented, for example, by the following equation:Figure 3A and 3B and 4A to 4OO. As Figure 3A and 3B and 4A to 4OO, these genes can be able to distinguish responders and non-responders using one gene. The logistic regression analysis was performed using JMP pro 16. Specifically, the univariate logistic regression analysis was performed using the efficacy / non-efficacy as the target variable, using the gene expression data of each gene using the data of all monocytes as the explanatory variable. For each gene, the significance difference and akaike’s information criterion (AIC) were calculated, the significance difference was confirmed, and the gene with a small AIC was confirmed as a gene more relevant to the therapeutic effect. When making a prediction model using the multivariate logistic regression analysis, the efficacy / non-efficacy was used as the target variable, and the gene expression data of the genes in the order of the small AIC for AIC in all monocytes was used as the explanatory variable. For all the prediction models, the Receiver Operating Characteristic curve (ROC) was made and the Area under the curve (AUC) was calculated. Monocytes migrate to tumor tissues and differentiate into macrophages. Thereby, the tumor environment is changed to inflammatory or immunosuppressive. It is considered that thereby the effectiveness of ICI to the tumor changes. Therefore, it is considered that, in principle, the method of the present application can be used to predict the effectiveness of all ICI.

[0185] The logistic regression analysis based on the expression data of a plurality of genes was performed. Figure 5 The results of the logistic regression analysis based on the expression of 3 genes shown in the table of the above figure are shown, Figure 6 The results of the logistic regression analysis based on the expression of 10 genes shown in the table of the above figure are shown, Figure 7 The results of the logistic regression analysis based on the expression of 20 genes shown in the table of the above figure are shown. In any case, the cells from responders and the cells from non-responders can be well determined. In addition, Figure 8 The results of the logistic regression analysis based on the expression of all 160 genes shown in Table 1 are shown. Figure 8 In this case, the AUC becomes 1, and the cells from responders and the cells from non-responders can be completely correctly distinguished. The logistic regression analysis based on 30 genes Figure 13 ), 40 genes Figure 14 ), 45 genes Figure 15 ), 50 genes Figure 16 ), 55 genes Figure 17 ), and 60 genes Figure 18) were able to well determine the cells from the responders and the cells from the non-responders in either case. It was seen that the cells from the responders and the cells from the non-responders could be distinguished based on the expression of each of the identified genes, and that the cells from the responders and the cells from the non-responders could be well distinguished when the combination of the expression of a plurality of genes was used. It is understood that the combination of the expression of several genes to be evaluated can be appropriately determined as needed.

[0186] Decision tree analysis was performed. The results are shown in Figure 9 Through the decision tree analysis, the cells from the responders and the cells from the non-responders could be well determined.

[0187] A machine learning model using a neural network was made. The learning model was made using a data set including the learning data, and the obtained learning model was used to analyze a data set including the validation data (data not included in the learning data). The results are shown in Figure 10 Through the machine learning model, the cells from the responders and the cells from the non-responders could be well determined. The neural network modeling was performed using JMP Pro 16. The efficacy / non-efficacy was used as the target variable, and the expression data of 160 genes in all mononuclear cells were used as the explanatory variable. In the modeling, a single layer of Tanh function was used as the activation function, and the number of nodes of the hidden layer was set to 3. The maximum number of learning and the number of iterations were set to 50 times each, and in the constructed model, the overfitting penalty value and the convergence criterion were set to 0.001 and 0.00001, respectively, in order to prevent overfitting in the learning model. As the validation method, the random K-fold method was used, and as the index for evaluating the prediction accuracy of the constructed model, the coefficient of determination R2 (R square) and the root of average squared error (RASE) were used.

[0188] A machine learning model using a random forest was made. The learning model was made using a data set including the learning data, and the obtained learning model was used to analyze the learning data. The results are shown in Figure 11 Through the machine learning model, the cells from the responders and the cells from the non-responders could be well determined. In addition, the learning model was made using a data set including the learning data, and the obtained learning model was used to analyze a data set including the validation data (data not included in the learning data). The results are shown in Figure 12As shown, it was also possible to well determine cells from a responder and cells from a non-responder using a machine learning model. The random forest modeling was implemented using JMP Pro 16. This was implemented using 160 gene expression data in all monocytes as explanatory variables using efficacy / non-efficacy as the target variable. The number of trees made with Bootstrap was set to 100, the number of extracted feature amounts for each branch was set to 110, the minimum size of the branch was set to 10, and the maximum size was set to 2000.

[0189] [Table 1]

[0190] Table 1: List of 160 genes and reference numbers of each gene in this specification

[0191]

[0192]

[0193]

[0194]

[0195]

[0196]

[0197]

[0198]

[0199]

[0200] [Table 2]

[0201] Table 2: Data based on total monocytes

[0202]

[0203]

[0204] [Table 3]

[0205] Table 3: Data based on non-classical monocytes

[0206]

[0207]

[0208] [Table 4]

[0209] Table 4: Data based on mixture of classical and non-classical monocytes

[0210]

[0211]

[0212] [Table 5]

[0213] Table 5: Data based on intermediate monocytes

[0214]

[0215]

[0216] [Table 6]

[0217] Table 6: Data based on classical monocytes

[0218] Reference number CR background log_fc pval padj 106 46.0357 25.88433219 0.57577825 0.022382 0.0223815 130 7.39273 4.361418589 0.5276993 0.00087 0.0017403

[0219] [Table 7]

[0220] Table 7: Gene groups other than the above

[0221] Gene name Monocyte type PD background log_fc pval padj ANKRD20 A11 P intermediate 4.02 1.60 0.92 2.25.E-06 5.12.E-05 ATP2B1-AS1 CD14+CD16 4.29 7.46 -0.55 6.87.E-08 3.74.E-07 CYP4F29 P intermediate 2.76 1.12 0.90 8.67.E-05 5.96.E-04 GPX1 CD14+CD16 5.54 9.69 -0.56 3.97.E-06 8.93.E-06 LINC01578 intermediate 3.41 1.78 0.65 6.79.E-03 1.76.E-02 NAPSB CD16 3.27 5.91 -0.59 1.33.E-02 3.18.E-02 SMIM25 total monocytes 3.68 2.01 0.61 3.16.E-07 4.97.E-06 TTTY15 intermediate 1.39 2.49 -0.58 1.02.E-02 2.45.E-02

[0222] Also, an attempt was made to screen genes considered useful for predicting therapeutic effects from 160 genes. Specifically, genes were screened by three different methods. A prediction model was constructed by logistic regression analysis based on the screened gene groups. The logistic regression analysis was performed using JMP pro 16.

[0223] First, genes associated with therapeutic effectiveness were screened by stepwise method according to the relationship between the expression amount of each of the 160 genes in mononuclear cells and therapeutic effects. As a result, as shown in Table 1, 10 genes were selected. When the expression of the 10 genes was used to predict whether the case from which each mononuclear cell was derived was PD or non-PD (non-PD), prediction shown by the ROC curve and AUC shown in Table 2 could be performed. Figure 19 Figure 19

[0224] Likewise, genes associated with therapeutic effectiveness were screened by LASSO regression according to the relationship between the expression amount of each of the 160 genes and therapeutic results. As a result, as shown in Table 3, 47 genes were selected. When the expression of the 47 genes was used to predict whether the case from which each mononuclear cell was derived was PD or non-PD (non-PD), prediction shown by the ROC curve and AUC shown in Table 4 could be performed. Further, 27 genes showing p < 0.0001 and prediction based on the expression of the 27 genes are shown in Table 5. Figure 20 Figure 20 Figure 21

[0225] ​​​​​Similarly, genes associated with the effectiveness of the treatment were selected by elastic net regression based on the relationship between the expression amount of each of the 160 genes and the treatment outcome. As shown in Table 8, 45 genes were selected. When the expression of the 45 genes was used to predict whether the case from which the mononuclear cells were derived was PD or non-PD, the prediction shown in the ROC curve and AUC shown in Table 9 was possible. Figure 22 Figure 22 Furthermore, 17 genes with p < 0.0001 and the prediction based on the expression of the 17 genes are shown in Table 10. Figure 23

[0226] In the above example, the treatment outcome (CR or PD) was classified based on the length of PFS, but in the following examples and Tables 8 to 10 and Figures 19-23 , the treatment outcome was classified based on the following criteria.

[0227] CR (complete response): a state in which the tumor completely disappeared

[0228] PR (partial response): partial response. A state in which the sum of the sizes of the tumors decreased by 30% or more.

[0229] SD (stable disease): a state in which the size of the tumor did not change.

[0230] PD (progressive disease): a state in which the sum of the sizes of the tumors increased by 20% or more and the absolute value also increased by 5 mm or more, or a state in which a new lesion appeared. Hereinafter, non-PD means other than PD, i.e., CR, PR, or SD.

[0231] [Table 8]

[0232] Table 8: 8 cases of lung cancer

[0233] Sample Age at treatment start Gender Stage Treatment Best effect Survival PFS OS 1 53 Male pTlcN2M0 IIIA stage Pembrolizumab non-PD Survival 334 2000 2 79 Male pT4N3Mlc IVB stage Pembrolizumab PD Death 77 226 3 81 Male pT2bN1M0 IIB stage Pembrolizumab non-PD Death 525 850 4 80 Male pT2aNOM0 IB stage Pembrolizumab PD Death 29 29 5 73 Male pT3NOM0 IIB stage Nivolumab PD Survival 66 1079 6 72 Male pTlbN2MO IIIA stage Pembrolizumab PD Death 168 196 7 51 Female cT3N2M0 IIIB stage cDDP+MTA+Pembrolizumab non-PD Survival 1372 1372 8 68 Female cT4N0M0 IIIA stage cDDP+MTA+Pembrolizumab non-PD Death 154 607

[0234] PBMCs from 8 lung cancer patients of Fukushima Medical University before treatment were newly obtained (see Table 8). As shown in Table 8, the best effect of the treatment of cases 1, 3, 7, and 8 was non-PD (i.e., any one of CR, PR, and SD), and the best effect of the treatment of cases 2 and 4 to 6 was PD. In the above example, the analysis of the gene expression in the mononuclear cells was performed, but in the present study, a study using the PBMCs themselves was performed. Specifically, PBMCs were obtained from the blood samples obtained from each case, and the PBMCs were used as the analysis target. Specifically, RNA was extracted from each of the obtained PBMCs, and RNA-Seq analysis was performed based on a conventional method, and the expression data of each gene were obtained.

[0235] In the above, 45 genes associated with the effectiveness of the treatment were selected based on the results of the stepwise selection. Figure 19 ​​The gene expression of the 10 genes shown was used to construct a prediction model. Using this prediction model, the treatment effectiveness of the patients of Table 8 was predicted based on the mRNA expression of the PBMC of each patient before treatment, and the results are shown in Table 9, with cases 1, 5, 7, and 8 predicted as non-PD, and the others predicted as PD. That is, for 6 of the 8 cases, it was possible to correctly predict whether it was non-PD or PD.

[0236] [Table 9]

[0237] Table 9: Prediction results using 10 genes

[0238] Patient Probability [PD]_1_2 Probability [non-PD]_1_2 Predicted result 1 0.07 0.93 non-PD 2 0.48 0.52 PD 3 0.74 0.26 PD 4 0.79 0.21 PD 5 0.26 0.74 non-PD 6 0.35 0.65 PD 7 0.22 0.78 non-PD 8 0.23 0.77 non-PD

[0239] Using 12 genes including RBBP4 and HSBP1 in addition to the above 10 genes, a prediction model based on logistic regression was constructed and evaluated again, as a result of which cases 1, 7, and 8 were predicted as non-PD, and the others were predicted as PD. That is, for 7 of the 8 cases, it was possible to correctly predict whether it was non-PD or PD. In this way, by appropriately using the screened gene group, it was possible to construct a prediction model of the treatment effect using the PBMC of a patient.

[0240] [Table 10]

[0241] Table 10: Prediction results using 12 genes

[0242] Patient Probability [PD]_1_2 Probability [non-PD]_1_2 Predicted result 1 0.93 0.07 non-PD 2 0.48 0.52 PD 3 0.30 0.70 PD 4 0.21 0.79 PD 5 0.70 0.30 PD 6 0.63 0.37 PD 7 0.77 0.23 non-PD 8 0.72 0.28 non-PD

[0243] As described above, when using lung cancer patients of the Fukushima Medical University in Japan as a validation dataset, it was also possible to predict the responsiveness of each patient to a PD-1-based immune checkpoint inhibitor. Furthermore, not only using isolated mononuclear cells, but also using a sample containing mononuclear cells such as PBMC, it was possible to predict the responsiveness. Moreover, the prediction can be performed based on a sample before treatment. As described above, it can be understood that by the RNA expression in the PBMC of a cancer patient before treatment, it is possible to predict the effectiveness of a PD-1-based immune checkpoint inhibitor. When the treatment effect increases or decreases in a patient who has been administered an immune checkpoint inhibitor in combination with another anticancer agent, by confirming the mRNA in the sample of that patient, it is also possible to predict the effect of the above-mentioned other anticancer agent on the treatment responsiveness of the patient to the immune checkpoint inhibitor treatment.

[0244] The immune checkpoint inhibitor releases the switch of the immune suppression of the immune cell. The present invention shows that the therapeutic effectiveness of ICI can be predicted by the gene expression on the immune cell side rather than the cancer cell side. That is, the present invention shows that, for the immune cell whose switch of the immune checkpoint is suppressed by ICI, the prediction is made as an index of the gene expression in the cell. The present invention provides a method of determining the degree of difficulty in releasing the immune checkpoint on the immune cell side by the gene expression on the immune cell side before treatment, and thus it is considered that the therapeutic effectiveness prediction method is useful for all cancer species for which ICI is considered to be effective regardless of the cancer species.

Claims

1. A method for predicting the therapeutic effect of immune checkpoint inhibitors (ICIs) on tumors in a subject, comprising: Determine the expression levels of genes in samples obtained from the subject that contain one or more genes selected from any of the gene populations listed in Tables 1-7; and Using a predictive model constructed in a manner capable of predicting the treatment effect, the therapeutic effect of ICI is predicted based on the determined expression level. The prediction model is constructed based on the expression levels of genes containing one or more of the genes in samples obtained from patient groups that responded to ICI treatment and patient groups that did not respond to ICI treatment, as well as information on the effectiveness of ICI treatment.

2. The method according to claim 1, wherein, One or more genes selected from the gene groups recorded in Table 1 include any one or more genes from the gene groups recorded in Figures 4A to 4NN.

3. The method according to claim 1 or 2, wherein, One or more genes selected from the gene groups listed in Table 1 include one or more or all of the genes selected from the group consisting of CLEC12A, TMEM176A and TMEM176B.

4. The method according to any one of claims 1 to 3, wherein, One or more genes selected from the gene groups recorded in Table 1, comprising more than 5 genes.

5. The method according to any one of claims 1 to 4, wherein, One or more genes selected from the gene groups recorded in Table 1 contain more than 10 genes.

6. The method according to any one of claims 1 to 5, wherein, The gene selected is one or more genes from the gene groups recorded in Table 1, which contain more than 20 genes.

7. The method according to any one of claims 1 to 6, wherein, The prediction model is built using machine learning.

8. The method according to any one of claims 1 to 6, wherein, The prediction model is constructed using logistic regression.

9. A method of treating a tumor in a subject, comprising the step of administering a therapeutically effective amount of an immune checkpoint inhibitor (ICI) to the subject. The subject is a subject predicted to be effective for immune checkpoint inhibitor (ICI) treatment by the method described in any one of claims 1 to 8.

10. The method according to any one of claims 1 to 9, wherein, Immune checkpoint inhibitors (ICIs) are antibodies or PD-1-binding fragments of antibodies that can bind to PD-1 and neutralize the binding of PD-1 to PD-L1.

11. The method according to any one of claims 1 to 9, wherein, Immune checkpoint inhibitors (ICIs) are antibodies or PD-1-binding fragments of antibodies that can bind to PD-L1 and neutralize the binding of PD-1 to PD-L1.

12. An immune checkpoint inhibitor (ICI) or a pharmaceutical composition containing an immune checkpoint inhibitor, used in the method of any one of claims 9 to 11.

13. A method for predicting the effect of a candidate substance on a patient's therapeutic response to immune checkpoint inhibitor therapy, comprising: Determine the expression levels of genes containing one or more genes selected from any of the gene populations listed in Tables 1-7 in both a first sample obtained from a patient before administration of the candidate substance and a second sample obtained from a patient after administration of the candidate substance; and A predictive model, constructed in a manner capable of predicting the efficacy of immune checkpoint inhibitor therapy, was used to predict the therapeutic effect of ICI based on the determined expression levels. The prediction model is constructed based on the expression levels of genes containing one or more of the aforementioned genes in samples obtained from patient groups that responded to ICI treatment and patient groups that did not respond to ICI treatment, as well as information on the effectiveness of ICI treatment. An increase in the predicted efficacy of ICI for the first and second samples indicates that the candidate substance may improve the efficacy of ICI, while a decrease in the predicted efficacy of ICI for the first and second samples indicates that the candidate substance may decrease the efficacy of ICI.

Images