A method for constructing a prediction model of immune-related adverse reaction events, and a marker and a kit

Abstract

The application discloses a method for constructing a prediction model of immune-related adverse reaction events, and uses external data to construct a model for predicting the occurrence of adverse reactions based on the abundance of intestinal flora, and proposes a RF score for the occurrence of immune-related adverse reactions, and uses the Youden index to calculate the cutoff value of the RF score, if the patient score RF score is less than 0.499, it indicates that the patient belongs to the high-risk group of immune therapy adverse reactions, otherwise belongs to the low-risk group of immune therapy adverse reactions. The application also provides application of intestinal flora markers in preparation of a kit for predicting immune therapy adverse reactions, and the intestinal flora is Enterococcus faecalis, Massilimicrobiota timonensis and the like. The application can predict the immune therapy adverse reactions based on the baseline intestinal microbial abundance and divide the high-risk and low-risk groups.

Description

Technical Field

[0001] This invention belongs to the field of biological detection and relates to a biomarker and a reagent kit. Specifically, it relates to a method for constructing a predictive model for the occurrence of immune-related adverse events, as well as a biomarker and a reagent kit. Background Technology

[0002] With the rapid development of tumor immunology, tumor immunotherapy, especially the application of immune checkpoint inhibitors (ICIs), has reshaped the paradigm of cancer treatment, bringing new treatment options and disease remission to many patients with refractory advanced tumors. Traditional ICIs, including anti-cytotoxic T-lymphocyte-associated antigen 4 (CTLA-4) and checkpoint inhibitors targeting programmed death receptor 1 (PD-1) / programmed death receptor-ligand 1 (PD-L1), have been approved for the treatment of various cancers, such as melanoma and non-small cell lung cancer. However, while effectively activating the body's anti-tumor immune response, ICIs may also cause immune-related adverse reactions, collectively known as immune-related adverse events (irAEs). Statistics show that nearly half of cancer patients using ICIs experience irAEs.

[0003] The occurrence of irreversible adverse events (irAEs) is mainly due to the overactivation of the body's autoimmune response, leading to damage to normal tissues. Severe irAEs can cause discontinuation of intracellular immunoglobulins (ICIs) and can also damage multiple target organs throughout the body, even endangering life. Common clinical manifestations of irAEs can range from mild (such as vomiting or rash) to severe (such as pneumonia or myocarditis). Notably, compared with patients who do not experience irAEs, statistics show that patients who do experience irAEs are generally associated with better clinical outcomes. Patients with irAEs tend to have better progression-free survival (PFS) and overall survival (OS), and many predictive models developed for irAEs have also shown good predictive value for treatment efficacy.

[0004] Therefore, effective prevention and treatment of irreversible adverse events (irAEs) are of great clinical significance for maximizing the efficacy of immunotherapy. Clinically, oncologists must weigh the risk of irAEs in patients against the clinical benefits of ICIs before prescribing them. This has driven the development of inventions related to potential biomarkers for identifying the development and severity of irAEs. The goal is to develop efficient biomarkers for identifying and predicting the occurrence of irAEs, thereby guiding the rational prescribing of these ICIs and developing monitoring strategies for high-risk patients to enable earlier detection and intervention of irAEs. Although an increasing number of inventions aim to explore the potential mechanisms and corresponding strategies for identifying patients prone to immune adverse reactions, the lack of scientific rigor and reproducibility, coupled with the complexity of clinical irAE diagnosis criteria and the limitations of the scale of clinical inventions, makes comprehensive methods for identifying efficient biomarkers of irAEs challenging.

[0005] A growing body of research indicates that gut microbiota can synergistically enhance the efficacy of immunotherapy (ICI) or influence the development of irreversible adverse events (irAEs). Targeting the gut microbiome has been shown to enhance ICI efficacy while reducing toxicity. Supplementation with specific probiotics can alleviate immune checkpoint inhibitor-associated colitis in mice, and fecal microbiota transplantation (FMT) has been clinically proven to be effective in treating refractory immune checkpoint inhibitor-associated colitis. Therefore, using the gut microbiome as a novel biomarker for the prevention and treatment of irAEs holds great promise. Recently, several inventions have also proposed specific disease prediction models and ICI efficacy based on the gut microbiome. A set of biomarkers constituting a baseline gut microbiome can predict which candidate patients will benefit more from ICI treatment. However, highly effective microbial biomarkers for predicting fewer adverse events during immunotherapy have been proposed. Furthermore, developing potential microbial biomarkers to predict the probability of irAEs and uncovering the underlying mechanisms driving checkpoint blockade toxicity will have significant clinical implications. More targeted treatment strategies can identify high-risk patients before treatment, thereby providing preventative measures to mitigate adverse reactions.

[0006] Quantitative techniques for fecal gut microbiota include high-throughput sequencing (including 16S rDNA sequencing and shotgun metagenomic sequencing) and real-time quantitative PCR, which can specifically detect the relative abundance of gut microbiota in feces. With the widespread application of high-throughput technologies, their advantages such as high efficiency, broad applicability, and comprehensive coverage of the detected microbiota allow for rapid understanding of the abundance characteristics of gut microbiota in multiple samples. This has broad application prospects in scientific invention and clinical practice, contributing to a better understanding of the link between gut microbiota and disease. However, due to the diversity of gut microbiota, there are significant differences in the characteristic microbiota associated with diseases screened in different invention cohorts. This is partly limited by factors such as the sample size, patient origin, sequencing platform, and sequencing strategy of different invention cohorts. Integrating multiple invention cohorts for specific microbiota screening and analysis can assist in the development of more efficient microbial prediction models and help to better explore the link between gut microbiota and host diseases. Summary of the Invention

[0007] To address the aforementioned technical problems in the prior art, this invention provides a method for constructing a predictive model for immune-related adverse events, as well as biomarkers and kits. This method, biomarkers, and kits aim to solve the technical problem of poor efficacy in preventing and treating immune-related adverse events in the prior art.

[0008] This invention provides a method for constructing a predictive model for the occurrence of immune-related adverse events, comprising the following steps:

[0009] 1) Microbial biomarker screening and model construction:

[0010] Using published studies containing data on immune-related adverse events (irAEs) and gut microbiota as the training set, iterative feature screening was performed on the included microbial markers related to immune-related adverse events. A machine learning algorithm (random forest algorithm to construct a decision tree model) was used to extract microbial features for model construction, resulting in fourteen gut microbial markers: *Enterococcus faecalis*, *Massilimicrobiota timonensis*, *Enterocloster aldenensis*, *Ligilactobacillus salivarius*, *Ruminococcus bromii*, *Porphyromonas somerae*, *Agathobacter rectalis*, *Actinomyces oris*, *Subdoligranulum variabile*, *Prevotellabuccalis*, *Parabacteroides merdae*, *Beduinibacterium massiliense*, *Parabacteroides goldsteinii*, and *Ruminococcus*. champanellensis;

[0011] 2) Cross-validation between cohorts and leave-one-out validation are performed in all studies on the training set to evaluate the stability and reproducibility of the model;

[0012] 3) Using the training set data, for the 14 selected gut microbial biomarkers, input the microbial feature abundance matrix, use the Random Forest function in Python-Scikit-Learn, use the predict() function to predict the probability of occurrence, and calculate the adverse reaction score RF score. Use the Youden index to calculate the cutoff value of the adverse reaction score, and obtain the adverse reaction score RF score = 0.499. Use this cutoff value as the threshold for subsequent validation of the model evaluation.

[0013] 4) A validation process: collect patient stool samples and use quantitative PCR sequencing or microbial 16S rDNA sequencing to obtain the relative abundance characteristics of the gut microbiota markers obtained in step 1); calculate the adverse reaction score (RF score) using the method in step 3). If the patient's adverse reaction score is less than 0.499, it indicates that the patient belongs to the high-risk group for adverse reactions to immunotherapy; if the patient's adverse reaction score is greater than 0.499, it indicates that the patient belongs to the low-risk group for adverse reactions to immunotherapy.

[0014] This invention also provides the application of gut microbiota markers in the preparation of kits for predicting adverse reactions to immunotherapy. The gut microbiota markers are Enterococcus faecalis, Massilimicrobiota timonensis, Enterococcus aldenensis, Ligilactobacillus salivarius, Ruminococcus bromii, Porphyromonas somerae, Agathobacter rectalis, Actinomyces oris, Subdoligranulum variabile, Prevotella buccalis, Parabacteroides merdae, Beduinibacterium massiliense, Parabacteroides goldsteinii, and Ruminococcus champanellensis.

[0015] Furthermore, patient stool samples were collected, and the relative abundance of the aforementioned gut microbiota markers in the stool was obtained using real-time quantitative PCR or fecal microbial 16S rDNA sequencing. The relative abundance characteristics of the gut microbiota markers were input into a microbial feature abundance matrix, and the probability of occurrence was predicted using the Random Forest function predict() in Python-Scikit-Learn. An adverse reaction occurrence score (RF score) was calculated. If the patient's score was less than 0.499, it indicated that the patient belonged to a high-risk group for adverse reactions to immunotherapy; if the patient's score was greater than 0.499, it indicated that the patient belonged to a low-risk group for adverse reactions to immunotherapy.

[0016] The present invention also provides a kit for predicting adverse reactions to immunotherapy, comprising reagents for detecting the following gut microbiota markers: Enterococcus faecalis, Massilimicrobiotatimonensis, Enterococcus aldenensis, Ligilactobacillus salivarius, Ruminococcus bromii, Porphyromonas somerae, Agathobacter rectalis, Actinomyces oris, Subdoligranulum variabile, Prevotella buccalis, Parabacteroides merdae, Beduinibacterium massiliense, Parabacteroides goldsteinii, and Ruminococcus champanellensis.

[0017] Furthermore, patient stool samples were collected, and the relative abundance of the aforementioned gut microbiota markers in the stool was obtained using real-time quantitative PCR or fecal microbial 16S rDNA sequencing. The microbial feature abundance matrix was input, and the probability of occurrence was predicted using the `predict()` function in Python-Scikit-Learn. An adverse reaction occurrence score (RF score) of 0.499 was calculated. If the patient's RF score was less than 0.499, it indicated that the patient belonged to a high-risk group for adverse reactions to immunotherapy; if the patient's RF score was greater than 0.499, it indicated that the patient belonged to a low-risk group for adverse reactions to immunotherapy.

[0018] Furthermore, the kit can detect microbial abundance using real-time quantitative PCR technology with primers containing markers for detecting gut microbiota, as shown below:

[0019] The upstream primer sequence for detecting Enterococcus faecalis is: gttggtgaggtaacggctca;

[0020] The downstream primer sequence for detecting Enterococcus faecalis is: tgctcggtcagactttcgtc;

[0021] The upstream primer sequence for detecting Massilimicrobiota timonensis is: atacatgcaagtcggacgca;

[0022] The downstream primer sequence for detecting Massilimicrobiota timonensis is: ggggcaggttgcttatgtct;

[0023] The upstream primer sequence for detecting Enterocloster aldenensis is: gacgatcagtagccgacctg;

[0024] The downstream primer sequence for detecting Enterocloster aldenensis is: cgggctttcactccagactt;

[0025] The upstream primer sequence for detecting Ligilactobacillus salivarius is: agtcacggctaactacgtgc;

[0026] The downstream primer sequence for detecting Ligilactobacillus salivarius is: accggctttgggtgttacaa;

[0027] The upstream primer sequence for detecting Ruminococcus bromii is: gtgaggtaacggctcaccaa;

[0028] The downstream primer sequence for detecting Ruminococcus bromii is: tgtctcagtcccaatgtggc;

[0029] The upstream primer sequence for detecting Porphyromonas somerae is: atgatgtaggcggaatgcgt;

[0030] The downstream primer sequence for detecting Porphyromonas somerae is: gtaagctgccttcgcaatcg;

[0031] The upstream primer sequence for detecting Agathobacter rectalis is: aacgtgctacaatggcgta;

[0032] The downstream primer sequence for detecting Agathobacter rectalis is: caccttccgatacggctacc;

[0033] The upstream primer sequence for detecting Actinomyces oris is: ggaccggtttttgctggttc;

[0034] The downstream primer sequence for detecting Actinomyces oris is: aaaacacccaaaggcgcatc;

[0035] The upstream primer sequence for detecting Subdoligranulum variabile is: actcctgtcgttagggacga;

[0036] The downstream primer sequence for detecting Subdoligranulum variabile is: tctacgcattccaccgctac;

[0037] The upstream primer sequence for detecting Prevotella buccalis is: gcacggtaaacgatggatgc;

[0038] The downstream primer sequence for detecting Prevotella buccalis is: tgtaacacgtgtgtagcccc;

[0039] The upstream primer sequence for detecting Parabacteroides merdae is: gaggaaggtcccccacattg;

[0040] The downstream primer sequence for detecting Parabacteroides merdae is: gtaagctgccttcgcaatcg;

[0041] The upstream primer sequence for detecting *Beduinibacterium massiliense* is: agagatcgggaggaacacca;

[0042] The downstream primer sequence for detecting Beduinibacterium massiliense is: gtttgctacccacgctttcg;

[0043] The upstream primer sequence for detecting Parabacteroides goldsteinii is: gtgaggtaacggctcaccaa;

[0044] The downstream primer sequence for detecting Parabacteroides goldsteinii is: ccttcatccttcacgcgact;

[0045] The upstream primer sequence for detecting Ruminococcus champanellensis is: gacgatcagtagccggactg;

[0046] The downstream primer sequence for detecting Ruminococcus champanellensis is: caatattccgcactgctgcc.

[0047] This invention utilizes published research containing data on "adverse reactions of immune checkpoint inhibitors" and "gut microbiota" as a training set. Through machine learning methods, a random forest model is constructed. Finally, a microbial prediction model is built using a total of 14 gut microbiota biomarkers, as detailed below.

[0048] As shown in the table below.

[0049]

[0050]

[0051] Using the Random Forest function in Python-Scikit-Learn, the predict() function is used to predict the probability of occurrence and calculate the adverse reaction score (RF score). The Youden index is used to calculate the cutoff value of the adverse reaction score (RF score = 0.499).

[0052] Furthermore, in addition to using 16S rDNA sequencing in the microbiome, real-time quantitative PCR can also be used to quantitatively detect the abundance of microbial communities 1-14 in the microbiome that predicts adverse reaction events.

[0053] Furthermore, in the real-time quantitative PCR detection method, 1-14 microbial community-specific primers are used to predict the occurrence of adverse reaction events, and standard 16S microbial primers are required as standards for quantification.

[0054] This invention discloses gut microbiota biomarkers for predicting the occurrence and severity of immune-related adverse events in cancer patients receiving anti-PD-1 / PD-L1 immunotherapy. This invention systematically reviews currently published inventions (data containing clinical information on immune-related adverse events in PD1 immunotherapy patients and baseline fecal microbial 16S rDNA sequencing information, N>50). All raw 16S sequencing data were quality controlled and merged using QIIME2. Bacterial genomes were annotated using the Ribosomal Database Project (RDP), and bacterial information was statistically analyzed at the genus level. Finally, 14 bacterial genera associated with the occurrence of immune-related adverse events were selected from three cohorts (N=190), and a predictive model for the occurrence of immune-related adverse events was constructed. The risk score calculated by the model can effectively predict the occurrence of immune-related adverse events in patients receiving immunotherapy.

[0055] Compared with existing technologies, the technical effects of this invention are positive and significant. This invention provides a set of gut microbiota biomarkers, kits, and a method for constructing a model for predicting the probability of immune-related adverse events (irAEs) induced by ICI drugs. The correlation between gut microbiota characteristics and the occurrence of immune-related adverse events induced by PD-1 / PD-L1 blockade therapy may have commonalities across different types of tumors. This invention can predict and classify high- and low-risk groups for irAEs in patients using immune checkpoint inhibitor drugs (anti-PD1 / PD-L1 drugs) based on baseline gut microbiota diversity levels. The gut microbiota-based immune-related adverse event prediction risk model of this invention has the potential to be applied in various types of tumors, such as colorectal cancer, lung cancer, melanoma, and gastric cancer, and may benefit patients. Attached Figure Description

[0056] Figure 1 This is a flowchart of the model creation process.

[0057] Figure 2 After iterative feature screening, the optimal microbial community is used for cross-validation between inventions and leave-one-out validation of the results.

[0058] Figure 3 It is the ROC curve and the area under the curve (AUC) of the microbial community features that best fit the model AUC after iterative feature screening.

[0059] Figure 4 The model prediction scores from the training set were validated using one external dataset (the Shanghai cohort was a fecal 16S rDNA sequencing cohort (N=65)) (chi-square test). Detailed Implementation

[0060] The following detailed description, in conjunction with the accompanying drawings and specific embodiments, of the gut microbiota model for predicting adverse events related to immunotherapy and its construction method, aims to provide a clearer understanding of its structural type and usage, but should not be construed as limiting the scope of protection of this invention.

[0061] Example 1

[0062] like Figure 1 As shown, this invention reviews existing published inventions by collecting existing published inventions related to programmed death receptor-1 / programmed death ligand-1 (PD1 / PD-L1) immune checkpoint inhibitor therapy.

[0063] The included inventions must specifically meet the following requirements:

[0064] 1. Patients receiving anti-PD1 / PDL1 drugs alone;

[0065] 2. The invention can provide detailed clinical information (especially including prognostic and irAEs information for patients using immunotherapy);

[0066] 3. The invention provides downloadable and analyzeable raw 16S rDNA sequencing data.

[0067] The following inventions were ultimately included and their SRA download serial numbers: Chau et al. (PRJNA687361), Hakozaki et al. (PRJNA606061), Zhang et al. (PRJEB48780), McCulloch et al. (PRJNA762360).

[0068] Raw sequencing data download and processing:

[0069] Based on the SRA sequence number, the raw 16S rDNA sequencing data were downloaded using QIIME2 software. The sequencing reading frame quality control score was Q>25. All raw 16S rDNA sequencing data were subjected to quality control and data merging. The bacterial genome was annotated using the Ribosomal Database Project (RDP) database.

[0070] Handling of confounding factors:

[0071] Given that the population characteristics (including tumor type, gender ratio, age and BMI, etc.) of the subjects included in different inventions may differ, and that the sequencing time, sequencing platform, sequencing depth, etc. of different invention cohorts may also differ, the inventions treat the many factors that may affect the purpose of the invention as confounding factors, and block the confounding factors with a greater impact by dividing them into intervals (the final number of inventions included (N>50) is used as the training set (N=190).

[0072] Screening of important microbial biomarkers:

[0073] Considering the significant confounding factors involved in the "invention," the inventors used the Unified Pipeline Meta-analysis (MMUPHin) method for heterogeneity in microbiome inventions to perform batch correction of the microbial abundance matrix based on the "invention." Furthermore, the significance of differential abundance was calculated for each individual microbial marker using the two-sided blocking Wilcoxon rank-sum test implemented in the "coin" package of R software (V.4.1.2). To identify more potential microbial biomarkers, 61 microbial features with p-values ​​<0.1 were included for further analysis.

[0074] Many inventions have pointed out that patients who benefit from immune checkpoint inhibitors may experience an increased probability of adverse events due to prolonged drug use, suggesting a correlation between efficacy and adverse reactions. Therefore, the inventors used the same method to screen for efficacy-related microbial markers and identified 34 efficacy-related microbial markers (P<0.1). Subsequently, efficacy-related microbiota (N=11) were further removed from the screened adverse reaction-related microbial markers, ultimately including 51 gut microbiota specifically related to adverse reactions for subsequent feature screening and model construction.

[0075] Microbial feature screening and model building:

[0076] The inventors integrated and analyzed the microbial characteristics and adverse immune reaction outcomes of patients collected in the training set. For the training set data (N=190), they iteratively screened 51 microbial markers related to adverse reactions and constructed a decision tree model using a machine learning algorithm—random forest—with a root node number of N=501, where each child node occupies 10% of the previous node. Figure 3 As shown, using iterative feature extraction, 14 optimal microbial features for model construction were extracted, which resulted in the optimal area under the receiver operating curve (ROC) (AUC = 0.88).

[0077] The Permutation Importance function in the ELI5 package (https: / / eli5.readthedocs.io / ) is used to calculate the feature importance of the model. The specific microbial features and their importance in the model are shown in the table below.

[0078]

[0079]

[0080] like Figure 2 As shown, to further verify that the 14 important microbial characteristics selected in the model can be stably applied in different research cohorts, the stability and reproducibility of the model in different research cohorts were further evaluated by performing study-by-study cross-validation and leave-one-subject-out (LOSO) cross-validation across all studies in the training set. The results show that the 14 important characteristic microbial communities selected by the inventors can be stably applied in different research cohorts.

[0081] Microbial predictive adverse reaction scoring and application

[0082] The inventors used training set data and, for 14 selected key features, input a microbial feature abundance matrix. They then used the `predict()` function in `RandomFroest` to predict the probability of adverse reactions and calculated an adverse reaction score (called the RF score). The Youden index was then used to calculate a cutoff value for the adverse reaction score in the training set. This cutoff value (RF score = 0.499) was used as a threshold for subsequent model evaluation.

[0083] Example 2

[0084] Experimental methods

[0085] External validation was performed using a pan-cancer cohort.

[0086] Patient recruitment and clinical sample collection

[0087] We are recruiting 65 patients (pan-cancer types) who wish to undergo immunotherapy (PD-1 / PD-L1 inhibitors) in clinical practice. Details of the patient recruitment are as follows:

[0088] Inclusion Criteria

[0089] 1) Patients with tumors that have been definitively diagnosed by histopathology;

[0090] 2) There are clearly measurable lesions;

[0091] 3) Outpatient and inpatient clinical data are clear and available, and patients are regularly admitted for efficacy evaluation (including imaging data such as MRI, enhanced CT of the chest and abdomen, and bone scintigraphy);

[0092] 4) Cancer patients receiving treatment with tumor immune checkpoint inhibitor drugs such as anti-PD1 / PD-L1 inhibitors.

[0093] Exclusion Criteria

[0094] 1) Having suffered from severe systemic infection, nasopharyngeal and oral inflammation (such as periodontitis, gingivitis, tonsillitis, etc.), respiratory tract infection, soft tissue or skin infection, abscess, or endocarditis within the past 3 months;

[0095] 2) The following medications have been used within the past 3 months: antibiotics, nonsteroidal anti-inflammatory drugs, probiotics, hormones, or immunosuppressants for more than 1 week;

[0096] 3) History of severe constipation or diarrhea, or significant changes in bowel habits within the past 3 months;

[0097] 4) A personal history of cancer (oral cancer, pharyngeal cancer, esophageal cancer, etc.), organ transplantation, or severe parasitic disease, or other digestive system diseases (such as inflammatory bowel disease, cirrhosis, etc.);

[0098] 5) History of trauma, surgery, etc. within the past 3 months;

[0099] 6) History of gastrointestinal bleeding, obstruction, or perforation within the past 3 months;

[0100] 7) Uncontrolled chronic metabolic, infectious, or endocrine disorders (such as hypertension, diabetes, hyperlipidemia, hyperuricemia, hyperpurine, thyroid dysfunction, etc.);

[0101] 8) Vegetarians or those whose eating habits have changed significantly within the past 3 months.

[0102] Baseline sample data (including plasma and stool samples) and relevant clinical information (including gender, age, body fat percentage, underlying diseases, and histopathological characteristics) were collected from each patient before receiving immunotherapy. Stool samples were collected longitudinally at the end of each treatment cycle to establish an immunotherapy sample bank. Tumor tissue samples (including biopsy and surgical samples) were collected from some patients at the time of initial diagnosis.

[0103] The method for collecting fecal specimens is as follows: To prevent the collected feces from being contaminated by urine, the inventor should be instructed to urinate completely before collecting the feces. The fecal specimen should weigh at least 5-10g. After collection, the fecal specimen should be stored in a disposable sterile container in a cool place (avoiding light or high temperature) and immediately frozen at -80℃.

[0104] In total, stool samples were collected from 65 cancer patients at Renji Hospital, affiliated with Shanghai Jiao Tong University School of Medicine, who received PD-1 monoclonal antibody / PD-L1 monoclonal antibody immunotherapy. The inventors performed 16S rDNA sequencing on the stool samples from Renji Hospital patients, as follows:

[0105] A total of 65 fecal samples were collected from cancer patients receiving PD-1 / PD-L1 monoclonal antibody immunotherapy at Renji Hospital, affiliated with Shanghai Jiao Tong University School of Medicine. Researchers performed 16S rDNA sequencing on the fecal samples from Renji Hospital patients. Details are as follows:

[0106] Real-time quantitative PCR technology for microorganisms

[0107] Primers were designed based on model microbial synthesis, as follows:

[0108] The upstream primer sequence for detecting Enterococcus faecalis is: gttggtgaggtaacggctca;

[0109] The downstream primer sequence for detecting Enterococcus faecalis is: tgctcggtcagactttcgtc;

[0110] The upstream primer sequence for detecting Massilimicrobiota timonensis is: atacatgcaagtcggacgca;

[0111] The downstream primer sequence for detecting Massilimicrobiota timonensis is: ggggcaggttgcttatgtct;

[0112] The upstream primer sequence for detecting Enterocloster aldenensis is: gacgatcagtagccgacctg;

[0113] The downstream primer sequence for detecting Enterocloster aldenensis is: cgggctttcactccagactt;

[0114] The upstream primer sequence for detecting Ligilactobacillus salivarius is: agtcacggctaactacgtgc;

[0115] The downstream primer sequence for detecting Ligilactobacillus salivarius is: accggctttgggtgttacaa;

[0116] The upstream primer sequence for detecting Ruminococcus bromii is: gtgaggtaacggctcaccaa;

[0117] The downstream primer sequence for detecting Ruminococcus bromii is: tgtctcagtcccaatgtggc;

[0118] The upstream primer sequence for detecting Porphyromonas somerae is: atgatgtaggcggaatgcgt;

[0119] The downstream primer sequence for detecting Porphyromonas somerae is: gtaagctgccttcgcaatcg;

[0120] The upstream primer sequence for detecting Agathobacter rectalis is: aacgtgctacaatggcgta;

[0121] The downstream primer sequence for detecting Agathobacter rectalis is: caccttccgatacggctacc;

[0122] The upstream primer sequence for detecting Actinomyces oris is: ggaccggtttttgctggttc;

[0123] The downstream primer sequence for detecting Actinomyces oris is: aaaacacccaaaggcgcatc;

[0124] The upstream primer sequence for detecting Subdoligranulum variabile is: actcctgtcgttagggacga;

[0125] The downstream primer sequence for detecting Subdoligranulum variabile is: tctacgcattccaccgctac;

[0126] The upstream primer sequence for detecting Prevotella buccalis is: gcacggtaaacgatggatgc;

[0127] The downstream primer sequence for detecting Prevotella buccalis is: tgtaacacgtgtgtagcccc;

[0128] The upstream primer sequence for detecting Parabacteroides merdae is: gaggaaggtcccccacattg;

[0129] The downstream primer sequence for detecting Parabacteroides merdae is: gtaagctgccttcgcaatcg;

[0130] The upstream primer sequence for detecting *Beduinibacterium massiliense* is: agagatcgggaggaacacca;

[0131] The downstream primer sequence for detecting Beduinibacterium massiliense is: gtttgctacccacgctttcg;

[0132] The upstream primer sequence for detecting Parabacteroides goldsteinii is: gtgaggtaacggctcaccaa;

[0133] The downstream primer sequence for detecting Parabacteroides goldsteinii is: ccttcatccttcacgcgact;

[0134] The upstream primer sequence for detecting Ruminococcus champanellensis is: gacgatcagtagccggactg;

[0135] The downstream primer sequence for detecting Ruminococcus champanellensis is: caatattccgcactgctgcc. Microbial 16S rDNA sequencing technology.

[0136] The specific experimental steps are as follows:

[0137] 1. Fecal DNA extraction: Total microbial genomic DNA was extracted from 100-200 mg of fecal samples using the HiPure Stool DNA Mini Kit (China). The DNA extract was stored at -80°C.

[0138] 1) Transfer 100–200 mg of fecal sample to a 2 ml centrifuge tube, immediately add 1.2 ml of Buffer SSL to the sample, and vortex for up to 1 minute to fully disperse the sample. If the sample is liquid, pipette 0.15–0.2 ml of the sample. Processing clumped samples may require longer vortexing time to fully disperse the sample.

[0139] 2) Incubate in a 70℃ water bath for 10 minutes. If you need to extract recalcitrant bacterial DNA, increase the water bath temperature to 90℃. If you only need to extract human DNA, omit the 70℃ or 90℃ setting.

[0140] 3) Vortex for 15 seconds. Centrifuge at ≥14,000 x g for 10 minutes at room temperature.

[0141] 4) Transfer 250 μl of supernatant to a new 1.5 ml centrifuge tube. If RNA needs to be removed, add 2 μl of RNase A to the lysis buffer and let stand at room temperature for 15 minutes.

[0142] 5) Add 20 μl Proteinase K and 250 μl Buffer AL supernatant. Invert and mix 10 times. Incubate at 70°C for 10 minutes.

[0143] 6) Add 250 μl of anhydrous ethanol to the sample and mix by inverting 10 times.

[0144] 7) Load HiPure DNA Mini Column I into a 2ml collection tube. Transfer the mixture to the column. Centrifuge at 10,000x g for 30–60 seconds.

[0145] 8) Discard the eluent and reattach the column to the collection tube. Add 500 μl of Buffer GW1 (diluted with anhydrous ethanol) to the column. Centrifuge at 10,000 x g for 30–60 seconds. Buffer GW1 must be diluted with anhydrous ethanol. Follow the dilution instructions on the bottle label or in the instruction manual.

[0146] 9) Discard the filtrate and reassemble the column into the collection tube. Add 600 μl of Buffer GW2 (diluted with ethanol) to the column.

[0147] Centrifuge at 10,000 x g for 30–60 seconds. Buffer GW2 must be diluted with anhydrous ethanol. Dilute according to the bottle label or instructions.

[0148] 10) Discard the filtrate and reassemble the column into the collection tube. Add 600 μl of Buffer GW2 (diluted with ethanol) to the column. Centrifuge at 10,000 x g for 30–60 seconds.

[0149] 11) Discard the filtrate and reassemble the column into the collection tube. Centrifuge at 13,000 × g for 2 minutes to dry the column.

[0150] 12) Load the column into a 1.5 ml centrifuge tube. Add 30–100 μl of preheated ddH2O (to 70 °C) to the center of the membrane on the column and incubate at room temperature for 2 minutes. Centrifuge at 13,000 × g for 1 minute.

[0151] 13) Discard the DNA binding column and store the DNA at -20°C.

[0152] 2. DNA quality inspection

[0153] Take 1 μL of DNA product and perform DNA quantification using a Qubit fluorometer (Thermo Scientific, USA). The quality standard is (DNA > 50 ng, and a clear main band can be seen on DNA electrophoresis).

[0154] 3.16S DNA V4 region PCR amplification and purification

[0155] The V3-V4 region of the bacterial genome 16S region was selected as the amplification region.

[0156] The primer sequences are:

[0157] 341F:CCTACGGGNGGCWGCAG;

[0158] 785R:GACTACHVGGGTATCTAATCC.

[0159] Seven nucleotide bases were randomly added before the upstream primer sequence of the universal primer to serve as barcode sequences to distinguish different samples.

[0160] The PCR reaction conditions were as follows: pre-denaturation, 94℃, 3 min; denaturation, 94℃, 10 s; annealing, 50℃, 20 s; extension, 72℃, 30 s; 20 cycles; final extension, 72℃, 10 min.

[0161] The number of PCR cycles should be adjusted appropriately based on the microbial content of different samples (maximum not exceeding 35 cycles). The amplified products were purified by agarose gel electrophoresis using the AXygen DNA gel extraction kit (AxyPrep DNAGel Extraction Kit, Axygen, USA). The specific steps are as follows:

[0162] 1) After PCR, take 3 μl for agarose gel electrophoresis;

[0163] 2) Mix the PCR clean beads and Votex capture magnetic beads (Vazyme, China) thoroughly and let stand at room temperature for 30 minutes before use;

[0164] 3) Add 20 μL of PCR clean beads, mix thoroughly, and incubate at room temperature for 10 minutes;

[0165] 4) Then place it on the magnetic rack, and remove the supernatant after the beads have completely separated;

[0166] 5) Add 200 μL of 80% ethanol, remove the supernatant after 30 seconds, and repeat the washing once;

[0167] 6) Allow the sample to air dry at room temperature until no liquid droplets remain;

[0168] 7) Add 22 μL of ddH2O to fully suspend the substance, and let it stand at room temperature for 5 minutes;

[0169] 8) Then place it on a magnetic rack. After the beads are completely separated, transfer 20 μL of supernatant to a new 1.5 mL centrifuge tube.

[0170] 9) Qubit quantification, pending sequencing.

[0171] 4. Illumina high-throughput sequencing and raw data processing

[0172] Sequencing was performed using the Miseq high-throughput sequencer (illumnia, USA).

[0173] Data preprocessing

[0174] Paired-end sequencing of the library was performed using an Illumina Miseq sequencer. Raw sequencing data was filtered according to the following criteria: ① Non-low-quality sequences were defined as those with an average quality greater than 25 for 50 consecutive bases and a sequence length greater than 50. Low-quality sequences that did not meet these criteria were discarded. ② The corresponding end sequences were ligated using Flash software, and sequences that could not be ligated were discarded. ③ Ligated sequences with a base length between 200 and 1000, containing ambiguous bases, containing mismatched bases, or with a maximum of 6 consecutive identical bases were discarded. The filtered sequence data were then assigned to specific individuals based on barcode labels.

[0175] 16S rDNA sequencing sequence classification

[0176] The obtained high-quality sequences are grouped into multiple OTUs (operational taxonomic units) in QIIME based on sequence similarity. uclust (http: / / www.drive5.com / uclust) is then used to cluster the sequences, and the longest sequence in each cluster is selected as the representative sequence.

[0177] The sequences from the RDP-classifier database (http: / / rdp.cme.msu.edu) were used as the training set to annotate the representative sequences of OTUs, thereby obtaining the taxonomic information of each OTU.

[0178] Verification results:

[0179] like Figure 4 As shown, the relative abundance of gut microbiota was calculated using the Shanghai cohort (16S rDNA sequencing cohort, N=65), and 14 characteristic microbiota were screened in the model. The model classifier from the training set was used to predict the probability of adverse events in each patient. Subsequently, a model cutoff threshold (RF score = 0.499, as mentioned above) was used for population prediction and classification. The chi-square test was used to evaluate the model's discriminative power (P<0.05), indicating that the model developed by the inventors based on 14 gut microbiota characteristics for predicting adverse immune events in patients receiving immune checkpoint inhibitors has certain feasibility and reproducibility. Simultaneously, model evaluation parameters based on the training set (such as sensitivity / specificity / accuracy / true positive rate (PPV) / true negative rate (NPV) / area under the precise recall curve (PRAUC)) were also evaluated.

Claims

1. A method for constructing a predictive model for the occurrence of immune-related adverse events in patients receiving anti-PD-1 / PD-L1 immunotherapy, characterized in that, Includes the following steps: 1) Microbial biomarker screening and model construction: Using published studies on "immune-related adverse reactions" and "gut microbiota" as a training set, feature screening was performed on the included microbial markers related to "immune-related adverse reactions" based on the training set data. Machine learning algorithms were used to iteratively extract microbial features for model construction, resulting in fourteen gut microbial markers, as shown below: Enterococcus faecalis, Massilimicrobiota timonensis, Enterococcus aldenensis, Ligilactobacillus salivarius, Ruminococcus bromii, Porphyromonas somerae, Agathobacter rectalis, Actinomyces oris, Subdoligranulum variabile, Prevotella buccalis, Parabacteroides merdae, Beduinibacterium massiliense, Parabacteroides goldsteinii, and Ruminococcus champanellensis. 2) Cross-validation between cohorts and leave-one-out validation are performed in all studies on the training set to evaluate the stability and reproducibility of the model; 3) Using the training set data, for the 14 selected gut microbial biomarkers, input the microbial feature abundance matrix, use the Random Forest function in Python-Scikit-Learn, use the predict() function to predict the probability of occurrence, and calculate the adverse reaction score RF score. Use the Youden index to calculate the cutoff value of the adverse reaction score RFscore = 0.499, and use this cutoff value as a threshold for subsequent model validation evaluation.

2. The application of a reagent for detecting gut microbiota markers in the preparation of a kit for predicting immune-related adverse events in patients receiving anti-PD-1 / PD-L1 immunotherapy, characterized in that, Intestinal flora markers are Enterococcusfaecalis, Massilimicrobiota timonensis, Enterocloster aldenensis, Ligilactobacillus salivarius, Ruminococcus bromii, Porphyromonas somerae, Agathobacter rectalis, Actinomyces oris, Subdoligranulum variabile, Prevotellabuccalis, Parabacteroides merdae, Beduinibacterium massiliense, Parabacteroidesgoldsteinii, Ruminococcus champanellensis.

3. The application according to claim 2, characterized in that, The kit is used to collect patient stool samples. The relative abundance of the aforementioned gut microbiota markers in the stool is obtained using real-time quantitative PCR or fecal microbial 16S rDNA sequencing. The relative abundance characteristics of the gut microbiota markers are input into a microbial feature abundance matrix. The Random Forest function in Python-Scikit-Learn is used to predict the probability of occurrence using the predict() function, and the adverse reaction occurrence score (RFscore) is calculated. If the patient's score is less than 0.499, it indicates that the patient belongs to the high-risk group for adverse reactions to immunotherapy. If the patient's score is greater than 0.499, it indicates that the patient belongs to the low-risk group for adverse reactions to immunotherapy.

4. A kit for predicting the occurrence of immune-related adverse events in patients receiving anti-PD-1 / PD-L1 immunotherapy, characterized in that, The reagent contains a reagent for detecting the following intestinal flora markers: Enterococcus faecalis, Massilimicrobiota timonensis, Enterococcus aldenensis, Ligilactobacillus salivarius, Ruminococcus bromii, Porphyromonas somerae, Agathobacter rectalis, Actinomyces oris, Subdoligranulum variabile, Prevotellabuccalis, Parabacteroides merdae, Beduinibacterium massiliense, Parabacteroides goldsteinii, and Ruminococcus champanellensis.

5. A kit for predicting the occurrence of immune-related adverse events in patients receiving anti-PD-1 / PD-L1 immunotherapy according to claim 4, characterized in that, The kit is used to collect patient stool samples. The relative abundance of the above-mentioned intestinal flora markers in the stool is obtained by real-time quantitative PCR or fecal microbial 16S rDNA sequencing. The microbial feature abundance matrix is ​​input, and the probability of occurrence is predicted using the Random Forest function in Python-Scikit-Learn and the predict() function. The adverse reaction occurrence score RFscore is calculated, and the adverse reaction score is 0.

499. If the patient's adverse reaction score is less than 0.499, it indicates that the patient belongs to the high-risk group for adverse reactions to immunotherapy. If the patient's adverse reaction score is greater than 0.499, it indicates that the patient belongs to the low-risk group for adverse reactions to immunotherapy.

6. A kit for predicting the occurrence of immune-related adverse events in patients receiving anti-PD-1 / PD-L1 immunotherapy according to claim 4, characterized in that, The kit contains primers for detecting gut microbiota markers, and uses real-time quantitative PCR technology to detect microbial abundance. The primers are shown below: The upstream primer sequence for detecting Enterococcus faecalis is: gttggtgaggtaacggctca; The downstream primer sequence for detecting Enterococcus faecalis is: tgctcggtcagactttcgtc; The upstream primer sequence for detecting Massilimicrobiota timonensis is: atacatgcaagtcggacgca; The downstream primer sequence for detecting Massilimicrobiota timonensis is: ggggcaggttgcttatgtct; The upstream primer sequence for detecting Enterocloster aldenensis is: gacgatcagtagccgacctg; The downstream primer sequence for detecting Enterocloster aldenensis is: cgggctttcactccagactt; The upstream primer sequence for detecting Ligilactobacillus salivarius is: agtcacggctaactacgtgc; The downstream primer sequence for detecting Ligilactobacillus salivarius is: accggctttgggtgttacaa; The upstream primer sequence for detecting Ruminococcus bromii is: gtgaggtaacggctcaccaa; The downstream primer sequence for detecting Ruminococcus bromii is: tgtctcagtcccaatgtggc; The upstream primer sequence for detecting Porphyromonas somerae is: atgatgtaggcggaatgcgt; The downstream primer sequence for detecting Porphyromonas somerae is: gtaagctgccttcgcaatcg; The upstream primer sequence for detecting Agathobacter rectalis is: aacgtgctacaatggcgta; The downstream primer sequence for detecting Agathobacter rectalis is: caccttccgatacggctacc; The upstream primer sequence for detecting Actinomyces oris is: ggaccggtttttgctggttc; The downstream primer sequence for detecting Actinomyces oris is: aaaacacccaaaggcgcatc; The upstream primer sequence for detecting Subdoligranulum variabile is: actcctgtcgttagggacga; The downstream primer sequence for detecting Subdoligranulum variabile is: tctacgcattccaccgctac; The upstream primer sequence for detecting Prevotella buccalis is: gcacggtaaacgatggatgc; The downstream primer sequence for detecting Prevotella buccalis is: tgtaacacgtgtgtagcccc; The upstream primer sequence for detecting Parabacteroides merdae is: gaggaaggtcccccacattg; The downstream primer sequence for detecting Parabacteroides merdae is: gtaagctgccttcgcaatcg; The upstream primer sequence for detecting *Beduinibacterium massiliense* is: agagatcgggaggaacacca; The downstream primer sequence for detecting Beduinibacterium massiliense is: gtttgctacccacgctttcg; The upstream primer sequence for detecting Parabacteroides goldsteinii is: gtgaggtaacggctcaccaa; The downstream primer sequence for detecting Parabacteroides goldsteinii is: ccttcatccttcacgcgact; The upstream primer sequence for detecting Ruminococcus champanellensis is: gacgatcagtagccggactg; The downstream primer sequence for detecting Ruminococcus champanellensis is: caatattccgcactgctgcc.

Images