Construction method and application of human head and neck tumor prognosis and treatment response prediction model based on programmed cell death characteristic gene

Abstract

The invention relates to the technical field of biomedicine, and particularly discloses a construction method and application of a head and neck tumor prognosis and treatment response prediction model based on a programmed cell death characteristic gene. According to the method, data of an scRNA-seq data set related to the head and neck tumors is utilized, and on the basis of genes related to currently known cell death modes, differential expression genes related to the cell death modes in the head and neck tumors are obtained through differential expression analysis; on the basis, screening out a characteristic gene which is remarkably related to prognosis of the head and neck tumors and is used for representing programmed cell death of the head and neck tumors; therefore, a head and neck tumor related risk score CDRscore model is constructed; further verifying the effectiveness of the CDRscore model through an independent verification set, and verifying the distribution of 10 genes in the CDRscore model by using space transcriptome sequencing of laryngeal squamous cell carcinoma tissues. Therefore, the CDRscore can be used as a new index for predicting postoperative prognosis and immunotherapy response, and can assist in clinical treatment decision of human head and neck tumors.

Description

Technical Field

[0001] The present invention belongs to the field of biomedicine, and specifically relates to a method for constructing and applying a human head and neck tumor prognosis and treatment response prediction model based on programmed cell death characteristic genes. Background Art

[0002] Laryngeal leukoplakia, also known as vocal cord leukoplakia, is a keratinized hyperplastic lesion of the laryngeal mucosal epithelium, which appears as white patches or plaques. It often occurs in the vocal cord mucosa, involving the unilateral vocal cord more than the bilateral vocal cord, and a part of the anterior commissure may also be affected. The main symptoms of patients are persistent or intermittent hoarseness, and there may also be discomfort in the throat, foreign body sensation, and chronic cough. Laryngeal leukoplakia has a tendency to malignant transformation and can be regarded as a precancerous lesion. Early active intervention can help prevent its development into squamous cell carcinoma. However, the mechanism leading to the transformation of laryngeal leukoplakia to squamous cell carcinoma is not yet clear, which greatly limits clinical treatment and prognosis.

[0003] Pathologically, laryngeal leukoplakia is a lesion with abnormal growth or tissue maturation and hyperkeratosis of the laryngeal mucosal epithelium, which is an atypical hyperplasia of epithelial cells. The atypical hyperplasia of epithelial cells in laryngeal leukoplakia is usually divided into three levels: ① Mild: the number of atypical cells is small and limited to the lower 1 / 3 of the epithelial layer; ② Moderate: the atypical cells exceed 1 / 2 or 2 / 3 of the epithelial layer; ③ Severe: the atypical cells involve the entire layer of the epithelial layer, which can better reflect the biological behavior of the lesion. At present, the pathological community mostly adopts the World Health Organization (WHO) squamous epithelial dysplasia classification method for the pathological tissue classification of laryngeal leukoplakia, which is divided into simple epithelial hyperplasia, mild, moderate and severe dysplasia (carcinoma in situ) and invasive carcinoma. Since the judgment of epithelial atypical hyperplasia is dominated by subjective factors, the latest expert consensus advocates the use of a dichotomy of "low grade (mild to moderate dysplasia)" and "high grade (severe dysplasia and carcinoma in situ)" to judge the severity.

[0004] Studies have shown that about 18% of laryngeal leukoplakia patients can transform into squamous cell carcinoma, and the average time of malignant transformation is 5.8 years. Multivariate analysis shows that pathology is a risk factor affecting the canceration of laryngeal leukoplakia, and its malignant transformation rate is directly related to the presence or absence of cell atypia and the degree of atypia. That is, pathological classification above moderate dysplasia is the strongest prognostic factor for the canceration of laryngeal leukoplakia, and as the degree of atypia increases, the time of canceration gradually shortens. Smoking and alcoholism are widely recognized risk factors for the onset of laryngeal leukoplakia. A domestic study of 263 laryngeal leukoplakia patients showed that more than half of the patients were long-term smokers and long-term drinkers, but the degree of smoking and drinking was not correlated with the degree of pathological atypia or squamous cell carcinoma transformation. In addition, the size of the lesion range is also a factor that restricts the canceration of laryngeal leukoplakia. The lesion range exceeds more than 1 / 2 of the vocal cord on one side and is prone to canceration. However, the degree of pathological cell atypia alone is not sufficient to predict the canceration of laryngeal leukoplakia. Many studies have shown that simple hyperplasia can also develop into squamous cell carcinoma. In addition, the pathological classification based on the state of epithelial dysplasia is often a description of a phenomenon and cannot deeply analyze the mechanism of squamous cell carcinoma transformation of laryngeal leukoplakia. Therefore, it is urgent to explore the molecular mechanism of its carcinogenesis.

[0005] Head and neck cancers (HNCs) are a heterogeneous group of tumors. Depending on the anatomical location of their occurrence, they are mainly divided into several types, such as the oral cavity, oropharynx, hypopharynx, larynx and vocal cords. Most HNCs originate from the mucosal epithelium, so they are also called head and neck squamous cell carcinoma (HNSC). HNSC is the most common squamous cell carcinoma and the most common head and neck cancer, and its incidence ranks seventh among cancers worldwide. According to the 2020 Global Cancer Observatory (GLOBOCAN) statistics, there are more than 930,000 new cases of HNSC each year, and approximately 460,000 deaths. HNSC accounts for 3% of the number of new cancer cases and all deaths worldwide. The incidence of HNSC varies from country to country and is usually associated with exposure to tobacco-derived carcinogens or excessive alcohol consumption. In addition, infection with human papillomavirus (HPV), especially HPV-16, can easily lead to the occurrence of HNSC, so HNSC can be divided into HPV-negative and HPV-positive HNSC. In general, the risk of HNSC in men is 2 to 4 times that of women. The median age of diagnosis of non-virus-related HNSC is 66 years old, while the median age of diagnosis of HPV-related oropharyngeal cancer and Epstein-Barr virus-related nasopharyngeal cancer is ~53 years old and ~50 years old, respectively. In the past 30 years, the survival rate of HNSC has improved slightly. When analyzing all age groups and anatomical sites recorded in the Surveillance, Epidemiology, and End Results registry (SEERregistry) supported by the National Cancer Institute (NCI), the 5-year survival rate of HNSC increased from 55% during 1992 to 1996 to 66% during 2002 to 2006. The improved survival rate was partly attributed to the emergence of HPV-associated HNSC.

[0006] Currently, most of the methods used for clinical postoperative prognosis of tumors are based on pathological examination, involving the number, size, tissue integrity, lymph node metastasis, and pathological grade of tumors. The TNM (Tumor Node Metastasis) staging system is currently the most widely used tumor staging system in the world, which can well reflect the progression of tumors. In the past, the TNM system was also used for staging and postoperative prognosis of HNSC. It was found that approximately 30% to 40% of HNSC patients were TNM stage I / II at diagnosis and could be cured after surgery or radiotherapy (RT). However, more than 60% of HNSC patients were initially diagnosed as locally advanced (LA). The Union for International Cancer Control (UICC) and the American Joint Committee on Cancer (AJCC) defined LA-HNSC as T3-4 or N1-3 in the 8th edition of the TNM staging system in 2017, and added additional information related to HPV positivity to the TNM staging system. LA-HNSC has a high risk of local recurrence and a poor prognosis, and is usually treated with a combination of surgery, radiotherapy, and systemic therapy. Evidence shows that HNSC gradually develops from cell atypia to varying degrees of atypical hyperplasia and eventually forms invasive HNSC, but most patients are diagnosed with advanced HNSC without obvious clinical symptoms of precancerous lesions. Currently, the 5-year survival rate of LA-HNSC patients after standard treatment is only about 40%. Therefore, screening molecular markers for the early occurrence of HNSC and establishing an early warning model for the occurrence and development of HNSC on this basis may provide strong support for the occurrence, postoperative prognosis, and precise treatment of HNSC.

[0007] Cell death is a fundamental physiological process that controls various life phenomena including growth, development, aging and disease. Unlike accidental cell death (ACD), regulated cell death (RCD), also known as programmed cell death (PCD), can be triggered by specific signal transduction mechanisms such as TP53, KRAS, NOTCH and hypoxia signaling and metabolic reprogramming. In 2018, the Nomenclature Committee on Cell Death (NCCD) identified 12 different modes of cell death: intrinsic apoptosis, extrinsic apoptosis, mitochondrial permeability transition-driven necrosis (MPT-driven necrosis), necroptosis, ferroptosis, pyroptosis, poly (ADP ribose) polymerase (PARP)-dependent cell death (Parthanatos, PDCD), endocytic cell death, NET-mediated cell death, lysosome-dependent cell death, autophagy-dependent cell death, immunogenic cell death, cellular senescence, and mitotic catastrophe. It is worth noting that other types of cell death have been recently revealed, including autosis, cuproptosis, anoikis, disulfidptosis, alkaliptosis, oxeiptosis, and mitotic cell death. Although a variety of cell death-related genes have been identified as biomarkers for diagnosis or prognosis of certain cancers, the role and status of different types of cell death patterns in different HNSC subtypes are not clear. Summary of the invention

[0008] The purpose of the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a method and application for constructing a model for predicting the prognosis and treatment response of human head and neck tumors based on programmed cell death characteristic genes. We were the first to analyze the expression and distribution characteristics of the currently known 18 different programmed cell death patterns in the five main different head and neck tumor tissues in humans, including thyroid cancer, laryngeal cancer, oropharyngeal cancer, nasopharyngeal cancer and oral squamous cell carcinoma; on this basis, 100 different machine learning language algorithm combinations were used to screen out 10 programmed cell death characteristic genes that were significantly associated with the prognosis of head and neck tumors; thereby, a quantitative individual programmed cell death-related risk score CDRscore was constructed; and it was further verified by independent validation sets and spatial transcriptome sequencing of laryngeal squamous cell carcinoma tissues. Therefore, CDRscore can be used as a new indicator for predicting postoperative prognosis and immunotherapy response, which will assist in clinical treatment decisions for head and neck tumors.

[0009] In order to achieve the above object, the present invention provides the following technical solutions:

[0010] A method for constructing a model for predicting prognosis or treatment response of head and neck tumors, comprising:

[0011] Step 1: Based on the currently known genes related to cell death patterns, using the scRNA-seq dataset data related to head and neck tumors, differential expression analysis was performed to obtain differentially expressed genes related to cell death patterns in head and neck tumors.

[0012] The details are as follows:

[0013] First, we established scRNA-seq maps based on the HNC-related scRNA-seq datasets GSE103322, GSE148673, GSE150321, GSE162025, GSE172577, and GSE181919 from GEO.

[0014] Secondly, we analyzed the single-cell transcriptome sequencing data. We processed a total of 326,178 single cells based on 6 GEO datasets (GSE103322, GSE148673, GSE150321, GSE162025, GSE172577, GSE181919). Seurat (V5.0.3) software was used for scRNA-seq analysis. We used Seurat (V5.0.3) for scRNA-seq analysis. First, we excluded low-quality cells (i.e., cells with less than 500 genes per cell, more than 5% of mitochondrial genes, or a logarithm of less than 3 per gene UMI (unique molecular identifier)), and used scDblFinder (V1.16) to identify and exclude potential double cells. Then, we normalized and scaled the expression matrix using the default settings of Seurat. To reduce the computational burden, we used a sketch-based analysis method to reduce the dimensionality of the data through principal component analysis (PCA) and uniform manifold approximation and projection (UMAP). Next, we applied unsupervised clustering analysis to identify cell subpopulations (Clusters) and applied the results to the full dataset. We used Seurat's FindAllMarkers() function with |log2Fold Change|>1 and adjusted p-value<0.001 as the threshold to identify marker genes within each subpopulation. Finally, we annotated the cell subpopulations based on the classical marker genes associated with programmed cell death patterns in previously published studies.

[0015] Step 2: Analyze the differentially expressed genes related to cell death patterns in head and neck tumors to obtain multiple programmed cell death genes related to the prognosis of head and neck tumors; then screen the multiple programmed cell death genes related to the prognosis of head and neck tumors to obtain characteristic genes that characterize programmed cell death of head and neck tumors. Step 2 includes:

[0016] (1) Based on step 1, on the basis of the TCGA-HNSC cohort, univariate Cox analysis was used to identify programmed cell death genes associated with the prognosis of head and neck tumors from the gene expression profiles of cell death patterns;

[0017] (2) Select the algorithm combination with the highest average C-index in the training set and validation set from a variety of machine learning language algorithm combinations as the optimal model, and use LASSO regression to select the variables in the optimal model to screen out candidate genes;

[0018] (3) Performing multifactorial stepwise Cox regression analysis on the candidate genes in step (2) to determine the characteristic genes that characterize programmed cell death in head and neck tumors.

[0019] The details are as follows:

[0020] Based on the TCGA-HNSC cohort, we identified 154 prognostic genes (prognosis-related programmed cell death genes in head and neck cancer) from the gene expression profiles of 18 CDMs (including alkaliptosis, anoikis, apoptosis, autophagy, autosis, cuproptosis, disulfidptosis, entotic cd, ferroptosis, immunogenic, lysosomal_cd, mitotic_cd, mpt, necroptosis, netotic_cd, oxeiptosis, parthatos, and pyroptosis) using univariate Cox analysis.

[0021] Then, in order to develop a prediction model with high accuracy and stability, we first integrated 10 machine learning algorithms and generated 100 algorithm combinations. These 10 machine learning algorithms included Random survival forest (RSF), elastic net, LASSO regression, Ridge regression, stepwise Cox regression, Cox boost, partial least squares regression of Cox model, supervised principal components analysis (SuperPC), generalized boosted regression modeling, and survival support vector machine. The TCGA-HNSC cohort was used as the training set, and the 100 algorithm combinations were fitted using the Leave-one-out cross validation (LOOCV) framework. The GSE65858 dataset and GSE41613 dataset were used as the validation set, and the consistency index (C-index) of all algorithm combinations in the training set and validation set was calculated, and the model with the highest average C-index in the training set and validation set was selected as the optimal model to obtain candidate genes. The results show that the optimal model is the combination model of random survival forest (RSF) and supervised principal component analysis (SuperPC), which has the highest average C-index (0.651), and the C-index of this combination model is in the leading position in all validation datasets.

[0022] LASSO regression was used to further refine the variables in the optimal model, and the optimal λ was obtained when the partial likelihood deviation reached the minimum based on the LOOCV framework. Under the optimal λ, 25 cell death-related genes with non-zero coefficients were screened.

[0023] Subsequently, we performed multivariate stepwise Cox regression analysis on these 25 candidate genes and finally identified 10 genes (MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3, MAP2K7), which were defined as cell death-related signatures (CDR signatures), that is, characteristic genes that characterize programmed cell death in head and neck tumors.

[0024] Step 3: According to the characteristic genes of programmed cell death that characterize head and neck tumors, analyze their respective expression levels and corresponding regression coefficients to construct a risk scoring model for head and neck tumors. Specifically, we analyzed the 10 genes (MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3, MAP2K7) that were finally determined and defined as cell death-related signatures (CDR signatures), and the corresponding regression coefficients of these 10 specific genes of programmed cell death in head and neck tumors, and constructed a cell death-related risk score (CDRscore) model, which is a risk scoring model for head and neck tumors. The model includes the following contents or formulas:

[0025] CDRscore=(4.985107*10 -4 )*NDFIP1 EXP +(-1.068404*10 -03 )*MAP2K7 EXP +(1.441470*10 -04 )*SLC2A3 EXP +(1.123539*10 -04 )*EFNB2 EXP +(3.678527*10 -04 )*HPRT1 EXP +(8.106291*10 -04 )*DDX19A EXP +(-3.859801*10 -04 )*BTG3 EXP +(8.725573*10 -05 )*HK1 EXP +(4.235660*10 -04 )*PCMT1 EXP+(8.672*10 -04 )*MRPL10 EXP .

[0026] Among them, the number in () is the expression level of the characteristic gene that characterizes programmed cell death of head and neck tumors, and the name of the specific gene follows (). The number obtained by adding these result values ​​is the value of the programmed cell death risk score (CDRscore).

[0027] Next, we used the cell death-related risk score (CDRscore) model to study the construction and calibration of survival analysis and prediction nomograms. We calculated the risk score of each patient in the TCGA-HNSC cohort based on the CDR signature, and then divided the patients into high-risk and low-risk groups according to the optimal cutoff value determined by the survminer package; Kaplan-Meier curve analysis was performed using the survminer package to study whether the overall survival of the high-risk group and the low-risk group showed significant differences (log-rank test, p<0.05); the receiver operating characteristic (ROC) curve was drawn using the timeROC package to analyze and evaluate the accuracy of the CDR score in predicting the overall survival of HNC patients. The area under the curve (AUC) of CDRscore was compared with other clinical characteristics, including age, stage, T stage, and N stage. Multivariate Cox regression analysis was performed using the regplot package, and a nomogram was constructed to integrate the risk score with clinical information, including age, stage, T stage, N stage, and gender. The calibration curve was used to evaluate the precise discrimination and accuracy of the nomogram, and the DCA curve was used to evaluate its net clinical benefit.

[0028] Immediately afterwards, we also recruited data from two other datasets (GSE65858 and GSE41613) to verify the clinical value of our CDRscore model. The results confirmed that the CDRscore model well distinguished the OS of patients, because the OS of cases in the high-risk group was significantly lower than that in the low-risk group.

[0029] Finally, the difference between high and low risks of the CDRscore model was verified based on ST sequencing of LSCC and normal control paraffin samples.

[0030] Based on the above content, the present invention further provides the use of a programmed cell death characteristic gene detection reagent in the preparation of head and neck tumor screening and / or prediction products, wherein the programmed cell death characteristic genes include the following 10 genes: MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3 and MAP2K7. The detection reagent is a reagent for detecting the expression amount of the programmed cell death characteristic gene.

[0031] The present invention also provides a device for predicting the prognosis or treatment response of head and neck tumors, characterized in that the device comprises:

[0032] A module for acquiring characteristic gene data of programmed cell death characterizing head and neck tumors, used to acquire the expression levels of programmed cell death characteristic genes, wherein the programmed cell death characteristic genes include MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3 and MAP2K7;

[0033] A calculation module, used to calculate the CDRscore value of each tumor tissue sample, specifically, to calculate the CDRscore value of each tumor tissue sample according to a cell death-related risk score (CDRscore) model;

[0034] The output module is used to output the risk type of the tumor tissue sample according to the CDRscore value of each tumor tissue sample. The output module divides the patients into a high CDRscore group and a low CDRscore group according to the optimal cutoff value determined by the survminer package.

[0035] The beneficial effects of the present invention are:

[0036] The present invention mainly analyzes the expression and distribution characteristics of 18 different currently known programmed cell death patterns in 5 different head and neck tumor tissues in humans, including thyroid cancer, laryngopharyngeal cancer, oropharyngeal cancer, nasopharyngeal cancer and oral squamous cell carcinoma, and annotates cell subpopulations according to the published classic marker genes; through differential expression analysis, the differentially expressed genes of programmed cell death genes in head and neck tumors are obtained; on this basis, 100 different machine learning language algorithm combinations are used to screen out 10 programmed cell death characteristic genes that are significantly related to the prognosis of head and neck tumors; thus, a quantitative individual programmed cell death-related risk score CDRscore is constructed; and it is further verified by independent validation sets and spatial transcriptome sequencing of laryngeal squamous cell carcinoma tissues. Therefore, CDRscore can be used as a new indicator for predicting postoperative prognosis and immunotherapy response, which will assist in clinical treatment decisions for head and neck tumors.

[0037] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0039] Figure 1 : Interpretation of the distribution of CDMs in different HNC subtypes. (A) Illustration of patient recruitment. TH: thyroid cancer, HC: laryngopharyngeal cancer, OP: oropharyngeal cancer, NP: nasopharyngeal carcinoma, OC: oral squamous cell carcinoma. (B) UMAP projection after aggregating 326,178 cells from six different datasets. (C) UMAP projection after aggregating HNC subtypes (left) and violin plots of the scores of different CDMs in different HNC subtypes (right).

[0040] Figure 2 :Description of the different distribution of CDMs in cell subpopulations based on scRNA-seq maps. (A) The percentage stacked bar chart shows the distribution of the proportion of CDMs of each cell subtype in each tissue. (B) The distribution of CDMs in fibroblasts, B cells, T / NK cells, and epithelial / malignant cells in different tissues of HNC.

[0041] Figure 3 :T cells from tumor patients exhibit high levels of cell death. (A) UMAP projection of HNC T / NK subtypes based on scRNA-seq profiles. (B) The percentage stacked bar chart indicates the proportion of T / NK subtypes derived from different tissues. (C) The figure shows the difference in cell death patterns of T / NK subtypes in tumor tissues and patient PBMCs.

[0042] Figure 4 :Distribution of different CDMs in HNCs by RNA-seq profiles. (A) Box plots show the differences in CDMs between tumor tissues and normal tissues in all TCGA-HNSC RNA-seq profiles. (B) Radar plots show the correlation of different CDMs with MSI (left) and TMB (right). (C) Heat maps show the correlation of tumor immune scores and tumor stemness indicators with different CDMs.

[0043] Figure 5 : C-index of 100 algorithm combinations on all datasets and average C-index.

[0044] Figure 6 : Overall construction of CDRscore. (A) LASSO regression analysis of cell death-related genes screened by RSF and superPC. (B) The figure shows the optimal value determined by log(λ) value on the horizontal axis of LASSO regression analysis. (C) The coefficients of the CDR signature finally obtained by stepwise Cox regression.

[0045] Figure 7 : Validation of the CDRscore model in TCGA-HNSC. (A) Risk score (top), patient survival status (middle), and indicator gene transcription distribution (bottom) of TCGA-HNSC high / low risk groups based on CDRscore. (B) TCGA-HNSC cohorts were grouped according to CDRscore and Kaplan-Meier curves of overall survival were drawn. (C) ROC curves and AUC values ​​of risk scores and clinical data.

[0046] Figure 8 : Validation of the CDRscore model in the GEO dataset. The GSE65858 HNC cohort (A) and the GSE41613 HNC cohort (B) were grouped according to the CDRscore, and the Kaplan-Meier curves of OS (left) and the ROC curves and AUC values ​​of the risk score at 1, 3, and 5 years (right) were plotted.

[0047] Fig. 9 :ST sample cell subset annotation and CDR signature gene expression. (A) Deconvolution ST images of normal samples (top) and LSCC tumor samples (bottom), each point includes a pie chart and shows the potential cell composition. The pie chart on the right shows the total proportion of cells in the two samples. (B) ST images show the transcription of CDRsignature genes in normal and tumor tissues.

[0048] Fig.10 : The cell subpopulation dominated by malignant cells has the highest risk score. (A) The median importance of different cell death patterns in predicting cell types in normal tissue (left) and tumor tissue (middle), and high risk scores can be seen in malignant cells (right). (B) ST images show four different subpopulations in tumor samples. (C) Statistical analysis of risk scores of the four different subpopulations. (D) The proportion of different cell types in the four subpopulations. ***p<0.0001. DETAILED DESCRIPTION

[0049] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention.

[0050] The content involving patients in this embodiment has obtained the clinical research approval ethics demonstration of the Scientific Research Project Review Ethics Committee of the 989th Hospital of the Joint Logistics Department of the People's Liberation Army of China, and the ethics number is: 2021-78.

[0051] Sources of relevant data in the embodiments: TCGA-HNSC cohort: from TCGA database ( https: / / portal.gdc.cancer.gov / ) was retrieved from the TCGA-HNSC open dataset, and a total of 528 cases were collected. In terms of patient gender, there were 386 males and 142 females. In terms of primary sites, there were 132 cases of tongue cancer, 117 cases of laryngeal cancer, 71 cases of lip cancer, 56 cases of floor of mouth cancer, 46 cases of tonsil cancer, 43 cases of oral cancer, 24 cases of root of tongue cancer, and 39 cases of other head and neck cancers. The 528 cases corresponded to 566 samples, including 522 tumor tissues and 44 corresponding normal tissues. The data of these TCGA-HNSC open datasets include RNA-seq expression matrix, clinical information, and somatic mutation information.

[0052] 8 datasets from GEO: The following 8 datasets were retrieved and screened from the Gene Expression Omnibus (GEO, https: / / www.ncbi.nlm.nih.gov / geo / ) for analysis. They mainly include RNA-seq data: GSE65858 (containing gene expression of 270 patients with head and neck squamous cell carcinoma), GSE41613 (97 HPV-negative oral squamous cell carcinoma patients); scRNA-seq data: GSE103322: (scRNA-seq of 5902 cells from 18 patients with oral cancer); GSE148673 (scRNA-seq of 46,501 single cells from 21 tumors); GSE150321 (two male patients with GSE162025 (10 patients with LSCC tissues from 6 patients, aged 66 and 64 years old, respectively) performed scRNA-seq on LSCC tissues; GSE162025 (10 patients with NPC tumor-peripheral blood paired samples performed scRNA-seq and T cell receptor sequencing on 176,447 cells; GSE172577 (6 patients with oral squamous cell carcinoma tumor tissue cells) and GSE181919 (23 patients with HNC, 9 normal tissues, 4 leukoplakia tissues, 20 primary cancer tissues, and 4 lymph node metastasis tissues scRNA-seq data.

[0053] Example 1: Construction of a map based on single-cell transcriptome sequencing (scRNA-seq) data to analyze the distribution of cell death-related modes (CDMs) of five HNC subtypes

[0054] The specific steps are as follows:

[0055] 1. We collected scRNA-seq data of five major subtypes of HNC from six different datasets (GSE103322, GSE148673, GSE150321, GSE162025, GSE172577, and GSE181919) from the Gene Expression Omnibus (GEO, https: / / www.ncbi.nlm.nih.gov / geo / ), including thyroid carcinoma (TH), oral squamous cell carcinoma (OC), hypopharyngeal carcinoma (HC), oropharyngeal carcinoma (OP), and nasopharyngeal carcinoma (NP) ( Figure 1 ).

[0056] 2. We used Seurat (V5.0.3) for scRNA-seq analysis. First, we excluded low-quality cells (i.e., cells with less than 500 genes per cell, more than 5% mitochondrial genes, or a logarithmic value of UMI (unique molecular identifier) ​​per gene less than 3), and used scDblFinder (V1.16) to identify and exclude potential double cells.

[0057] 3. We normalized and scaled the expression matrix using the default settings of Seurat. To reduce the computational burden, we used a sketch-based analysis method to reduce the dimensionality of the data through principal component analysis (PCA) and uniform manifold approximation and projection (UMAP).

[0058] 4. Next, we applied unsupervised cluster analysis to identify cell subpopulations (Clusters) and applied the results to the full dataset. We used Seurat's FindAllMarkers() function with |log2 Fold Change|>1 and adjusted p-value<0.001 as the threshold to determine the marker genes within each subpopulation.

[0059] 5. Finally, we annotated the cell subpopulations based on the marker genes of classic cell death-related patterns in previously published studies for screening of differentially expressed genes related to the prognosis of head and neck tumors.

[0060] Using these scRNA-seq data, we systematically analyzed the distribution of 18 different types of CDMs based on the transcription of characteristic genes. We performed UMAP clustering on five different subtypes of HNC and explored the status of fourteen regulatory CDMs in these five different subtypes, including oxygen death, alkali death, mitochondrial permeability transition-driven necrosis (mpt), apoptosis, and ferroptosis. The results showed that disulfide death, necroptosis, and pyroptosis may be the main CDMs of laryngeal pharyngeal carcinoma, while pyroptosis is the main CDM of nasopharyngeal carcinoma. The levels of apoptosis, ferroptosis, and autophagy were significantly upregulated in oropharyngeal cancer samples, while oral squamous cell carcinoma samples showed high levels of mpt as well as apoptosis. Finally, we also found that thyroid cancer samples had high levels of anoikis and apoptosis ( Figure 1 ).

[0061] Laryngeal leukoplakia (LL) is considered a precancerous lesion. We also compared the 14 CDMs in normal laryngeal tissue (NL, n = 9), LL (n = 4), tumor tissue (n = 61), peripheral blood mononuclear cells (PBMCs, n = 10) of HNC patients, and metastatic tumors in lymph nodes (tLN, n = 9) of HNC patients. Figure 2 A). It is noteworthy that the levels of ferroptosis and pyroptosis in LL samples were much higher than those in other samples. In contrast, the levels of autophagy, immunogenic cell death, ferroptosis, and pyroptosis in NL samples were stronger than those in tumor and tLN tissues. HNC mainly uses the TNM staging system to classify the disease in each typical anatomical subsite, so we compared the levels of different cell deaths in tumor tissues of different TNM stages. Compared with stage III, HNC in stages I and II showed higher levels of pyroptosis, necroptosis, mpt, copper death, and anoikis. These combined data indicate that based on scRNA-seq profiles, there are different distributions of CDMs in the five subtypes of HNC ( Figure 2 ).

[0062] Immunotherapy targeting tumor-infiltrating lymphocytes (TILs) has become an important approach for the treatment of HNC. TILs include helper T cells (CD4 + ) and cytotoxic T cells (CD8 + ), regulatory T cells (Tregs), NK cells, etc. We further subdivided the cells in the T / NK subsets into eight subsets through automatic annotation combined with manual labeling, including CD8 + T cells, CD4 + T cells, NK cells, Th17 cells, Th1 cells, Tregs, CD8 + Exhausted T cells (Tex) and proliferative T cells (Tprolif) Figure 3 A). Among them, CD8 + Tex, Tregs, Th1 and Tprolif are mainly derived from tumor tissues, CD4 + T and NK cells are mainly derived from PBMCs ( Figure 3B), indicating that HNCs have a high infiltration of TILs. The role and anti-tumor activity of T cells are affected by the tumor microenvironment, so we compared the differences in CDMs between tumor-infiltrating T cells (from tumor tissue) and circulating T cells (from PBMCs) in tumor patients, and focused on CD4 + T cells, CD8 + T cells, NK cells and Tprolif, found that CD4 + T cells, CD8 + Copper death was significantly upregulated in T cells and Tprolif ( Figure 3 C).

[0063] Example 2: Screening of programmed cell death genes significantly associated with prognosis of head and neck tumors

[0064] The specific steps are as follows:

[0065] 1. Based on Example 1, we identified 154 programmed cell death genes associated with the prognosis of head and neck tumors from the gene expression profiles of 18 CDMs using univariate Cox analysis on the basis of the TCGA-HNSC cohort;

[0066] 2. In order to develop a prediction model with high accuracy and stability, we first integrated 10 machine learning algorithms and generated 100 algorithm combinations. These 10 machine learning algorithms include Random survival forest (RSF), elastic network, LASSO regression, Ridge regression, stepwise Cox regression, Cox boost, partial least squares regression of Cox model, supervised principal components analysis (SuperPC), generalized boosted regression modeling, and survival support vector machine. The TCGA-HNSC cohort was used as the training set, and the 100 algorithm combinations were fitted using the Leave-one-out cross validation (LOOCV) framework;

[0067] 3. The GSE65858 dataset and the GSE41613 dataset were used as validation sets, and the consistency index (C-index) in the training set and validation set of all algorithm combinations was calculated. The model with the highest average C-index in the training set and validation set was selected as the optimal model to obtain candidate genes.

[0068] 4. We also recruited data from two other datasets (GSE65858 and GSE41613) to verify the clinical value of our CDRscore model.

[0069] The results show that the optimal model is the combination model of random survival forest (RSF) and supervised principal component analysis (SuperPC), which has the highest average C-index (0.651), and the C-index of this combination model is in the leading position in all validation datasets ( Figure 5 ). LASSO regression was used to further select the variables in the optimal model. Based on the LOOCV framework, the optimal λ was obtained when the partial likelihood deviation reached the minimum value. Under the optimal λ, 25 cell death-related genes with non-zero coefficients were screened. Subsequently, we performed multifactor stepwise Cox regression analysis on these 25 genes and finally identified 10 genes (MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3, MAP2K7), which were defined as cell death-related signatures (CDR signature) ( Figure 6 ).

[0070] Example 3: Establishment of CDRscore risk assessment prediction model

[0071] The specific steps are as follows:

[0072] 1. We constructed a cell death-related risk score (CDRscore) model based on the 10 genes (MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3, and MAP2K7) that were finally identified and defined as cell death-related signatures (CDR signatures).

[0073] 2. The prediction model includes the following contents or formulas:

[0074] CDRscore=(4.985107*10 -4 )*NDFIP1 EXP +(-1.068404*10 -03 )*MAP2K7 EXP +(1.441470*10 -04 )*SLC2A3 EXP +(1.123539*10 -04 )*EFNB2 EXP +(3.678527*10 -04 )*HPRT1 EXP +(8.106291*10 -04 )*DDX19A EXP +(-3.859801*10 -04 )*BTG3 EXP+(8.725573*10 -05 )*HK1 EXP +(4.235660*10 -04 )*PCMT1 EXP +(8.672*10 -04 )*MRPL10 EXP .

[0075] Among them, () is the expression level of the prognosis-related programmed cell death characteristic gene, and () is followed by the name of the specific gene. The number obtained by adding up these result values ​​is the value of the programmed cell death risk score (CDRscore).

[0076] 3. Calculate the risk score for each patient in the TCGA-HNSC cohort, determine the optimal median value using the survminer package, and divide the patients into high-risk and low-risk groups.

[0077] 4. Describe the relationship between risk score, patient status (alive or dead), overall survival (OS), and transcription of CDR signature.

[0078] Our results suggest that the OS of patients in the high-risk group was significantly lower than that in the low-risk group, p<0.0001( Figure 7 The area under the receiver operating characteristic curve was 0.772, which was higher than that of other clinical classifications ( Figure 7 We also recruited data from two other datasets (GSE65858, GSE41613) to verify the clinical value of our CDRscore model. The results confirmed that the CDRscore model distinguished the OS of patients well, because the OS of the high-risk group cases was significantly lower than that of the low-risk group ( Figure 8 ). In conclusion, we established a CDRscore model based on CDR signature, which was more accurate than other clinical information in predicting the prognosis of HNC.

[0079] Example 4: Spatial transcriptome sequencing analysis of the distribution of CDR signature genes in laryngeal squamous cell carcinoma tissues 1. Sample preparation

[0080] 1) Samples are re-embedded according to experimental groups:

[0081] a) Preheat the oven to 60°C, melt and filter the slice paraffin wax (melting point 56°C-58°C).

[0082] b) Place the formalin-fixed paraffin-embedded laryngeal squamous cell carcinoma specimens provided by the hospital in a tissue embedding box of appropriate size, immerse in section paraffin, and place in a 60°C constant temperature oven for wax immersion overnight.

[0083] c) After wax immersion overnight, the tissue block is removed, a preheated metal mold is taken, a small amount of paraffin is poured into the mold, and the tissue block is clamped from the embedding box. According to the experimental requirements, the tissue blocks of the same experimental group are arranged in order at the bottom of the mold with the cross-section facing downward, and photographed and marked. The tissue is gently flattened to ensure that there are no bubbles, paraffin is added to the upper edge of the metal mold, and the marked embedding frame is tilted to cover it, and then paraffin is added to submerge the upper edge of the embedding frame, and the wax block is left to solidify at room temperature.

[0084] d) After the wax block is completely solidified, place the wax block and the metal mold in a -20℃ refrigerator and freeze them for 30 minutes. The wax block and the metal mold can then be easily separated.

[0085] e) Store re-embedded paraffin samples at 4°C prior to sequencing, away from direct light to ensure uniform cooling and maintain RNA integrity.

[0086] 2) Expose tissue surface: Use RNaseZap TM Wipe the work area and all surfaces with RNase decontamination solution. Place the tissue wax block on the paraffin slicer specimen holder at room temperature, set the slice thickness to 15 micrometers (Micrometers, μm), and cut the tissue wax block at a speed of 15 μm / revolution until the area of ​​interest is exposed.

[0087] 2. Sample RNA Quality Assessment

[0088] 1) Extract RNA from paraffin sections using RNeasy FFPE Kit:

[0089] a) Set the paraffin sectioner to slice at 10 μm thickness, collect 10 μm thick tissue sections, immediately place the sections into a pre-cooled 1.5 ml (Millilitre, mL) centrifuge tube, add 160 μL (Microlitre, μL) of deparaffinization solution, vortex vigorously for 10 seconds (Seconds, s), and briefly centrifuge to allow the sample to reach the bottom of the tube.

[0090] b) Incubate at 56°C for 3 minutes and cool to room temperature. If the deparaffinized solution becomes waxy or solid after cooling, it means that too little deparaffinized solution was added or the sample carried too much paraffin. If this happens, add additional deparaffinized solution and repeat the 56°C incubation for 3 minutes.

[0091] c) Add 150 μL of PKD buffer to the centrifuge tube and vortex to mix. Centrifuge at 11,000×g for 1 min.

[0092] d) Add 10 μL of proteinase K to the lower clear phase and mix gently with a pipette.

[0093] e) Incubate at 56°C for 15 min, then at 80°C for 15 min. Vortex the liquid in the centrifuge tube every 3-5 min.

[0094] f) Transfer the lower uncolored phase to a new 2 mL centrifuge tube.

[0095] g) Incubate on ice for 3 min. Centrifuge at 20,000×g for 15 min.

[0096] h) Transfer the supernatant to a new 1.5 mL centrifuge tube, taking care not to disturb the pellet.

[0097] i) Use an RNase-free needle and syringe to inject 550 μL of RNase-free water into the vial of lyophilized DNase I (1,500 units). Invert the vial and mix gently without vortexing to prepare the RNase-free DNase I solution.

[0098] j) Add DNase Booster Buffer (about 16 μL), which is equivalent to one tenth of the total volume of the supernatant transferred in step f), and 10 μL of RNase-free DNase I solution to the tube, invert the solution to mix, and centrifuge briefly.

[0099] k) Incubate at room temperature for 15 min.

[0100] l) Add 320 μL of RBC buffer to the tube and mix the lysate thoroughly.

[0101] m) Add 720 μL of 100% ethanol to the tube and mix using a pipette, then proceed immediately to the next step. Precipitation may occur after adding ethanol, but it will not affect RNA collection.

[0102] n) Transfer 700 μL of sample to an RNeasy MinElute spin column (pink), including the pellet, in a 2 mL collection tube, gently close the cap, and centrifuge at no less than 8,000 × g for 15 s.

[0103] o) Repeat the previous step until all samples have passed through the column.

[0104] p) Prepare RPE Buffer by adding 4 volumes (44 mL) of ethanol (96% to 100%) to the bottle containing 11 mL of RPE Buffer.

[0105] q) Add 500 μL of RPE Buffer to the column, gently close the lid, and centrifuge at no less than 8,000 × g for 15 seconds.

[0106] r) Place the RNeasy MinElute spin column into a new 2 mL collection tube. Open the spin column cap and centrifuge at maximum speed for 5 minutes.

[0107] s) Leave at room temperature for 15 min to ensure the column membrane is dry.

[0108] t) To elute the RNA, transfer the column to a new 1.5 mL centrifuge tube, add 14 to 30 μL of RNase-free water to the column, and centrifuge at maximum speed for 1 min.

[0109] u) Store RNA on ice or at -80°C until ready to assess RNA quality.

[0110] 2) Use the Agilent RNA 6000Pico kit and the Agilent 2100 bioanalyzer to evaluate the DV200 of the sample RNA. The DV200 of the sample RNA ≥ 30% is considered to be a good indicator of RNA integrity and can be used for the next step of the experiment. 3. Hematoxylin-eosin staining, imaging, and de-crosslinking

[0111] 1) Slice production:

[0112] a) Before you begin, use RNaseZap TM Wipe all surfaces and work areas with RNase decontamination solution. Trim the wax block to ensure that the tissue area is less than 6.5 × 6.5 square millimeters (mm 2 ). Fill the water bath with ultrapure water and make sure the temperature is set at 42 °C and slide a laboratory wipe over the water surface to ensure there are no bubbles and particles.

[0113] b) Carefully wipe excess grease from the MX35 Ultra paraffin blade with 100% ethanol. Wait until the ethanol evaporates before continuing to wipe. Always use a new blade for sectioning. Secure the blade in the microtome's disposable blade holder and place a knife guard on the blade to minimize damage. Make sure the clearance angle is set to 10°.

[0114] c) Freeze the wax block for 10 to 30 minutes and then fix it in the specimen holder of the microtome, aligning it with the blade.

[0115] d) Set the microtome to 5 μm tissue sections and begin sectioning.

[0116] e) For tissue blocks with exposed tissue, discard the first few slices and start collecting the last few slices. When collecting the slices, place the tip of the paintbrush slightly above and parallel to the blade. While rotating the wheel handle, gently touch the edge of the slice with the paintbrush to lift the slice. With the help of the paintbrush, pick up the slice. Immediately place it on the surface of the water bath.

[0117] f) Allow the section to float on the surface of the water bath until it is mostly flat and wrinkle-free. The floating time will depend on the sample type. If the section is too wrinkled or wrinkled, allow it to float longer.

[0118] g) Float the sections onto regular glass slides and inspect the quality under a microscope to select regions of interest for section placement and imaging.

[0119] h) For Visium Spatial slides stored at -20°C, equilibrate at room temperature for at least 30 minutes (no more than 24 hours). Place tissue sections on the Visium Spatial slide, placing only one tissue section per capture area without any cracks, tears or folds in the tissue.

[0120] i) Place slides containing tissue sections in a low humidity environment, such as a desiccator, and avoid direct exposure to light. Visium Spatial slides containing tissue sections are incubated at 42°C for 3 hours, dried in a desiccator overnight, and can be stored in a desiccator at room temperature for up to 2 weeks.

[0121] 2) Hematoxylin-eosin staining and imaging:

[0122] a) Incubate the prepared sections at 60°C for 2 hours, then cool to room temperature, and soak in xylene-I for 10 minutes, xylene-II for 10 minutes, anhydrous ethanol-I for 3 minutes, anhydrous ethanol-II for 3 minutes, 96% ethanol-I for 3 minutes, 96% ethanol-II for 3 minutes, and 70% ethanol for 3 minutes, then put them in ultrapure water for 20 seconds, and immediately perform H&E staining and cover the sections.

[0123] b) Add 1 mL of hematoxylin to the sliced ​​tissue and incubate at room temperature for 3 min. Tilt the slices and discard the reagent. Then immerse the slices in 5 times the volume of ultrapure water, then transfer them to 10 times the volume of ultrapure water, and then transfer them to 15 times the volume of ultrapure water to wash away the hematoxylin.

[0124] c) Add 1 mL of anti-blue buffer to the tissue and incubate at room temperature for 1 min. Tilt the slices and discard the reagents, then immerse the slices in 15 times the volume of ultrapure water to wash away the anti-blue buffer.

[0125] d) Immerse the sections in alcohol eosin solution. Incubate at room temperature for 2 minutes. Do not use diluted eosin. Tilt the sections to discard the reagent and then immerse in 15 times the volume of ultrapure water to wash away the alcohol eosin solution.

[0126] e) Prepare coverslip mounting medium: Dilute glycerol with nuclease-free water to a final concentration of 85%. Vortex to mix and centrifuge briefly to remove air bubbles.

[0127] f) Place the sections on a flat, clean, non-absorbent work surface and add 100-150 μL of coverslip mounting medium to the tissue, evenly covering all tissue sections on the slide, applying the coverslip at an angle to one end of the slide. Slowly lower the slide without introducing air bubbles. Allow the glycerol to diffuse and settle.

[0128] g) Once coverslipping is complete, images are acquired using a light microscope and an M8 slide scanner.

[0129] 3) Remove the coverslip and remove the crosslinks:

[0130] a) Prepare 800 mL of ultrapure water in a beaker.

[0131] b) Immerse the slide sideways / horizontally in a beaker containing 800 mL of water with the coverslip completely sideways until the coverslip slowly separates from the slide.

[0132] c) Gently immerse the slide in 30x water, ensuring all mounting medium is removed.

[0133] d) Dry the slides at 37°C for 3 min to allow the slides to dry and immediately de-crosslink.

[0134] e) Prepare 0.1N HCl solution using Nuclease-free water. Thaw the de-crosslinking buffer (from the Visium FFPE Reagent Kit v2-Small) at room temperature, dilute the de-crosslinking buffer 10-fold with Nuclease-free water, vortex to mix, and centrifuge briefly.

[0135] f) Place the slide into the Visium CytAssist Tissue Slide Box.

[0136] g) Add 150 μL of 0.1N HCl along the edge of the well to evenly cover the tissue section without creating bubbles. Tap the cassette gently to ensure even coverage.

[0137] h) Remove the hydrochloric acid in the well. Add 100 μL of 0.1N HCl along the edge of the well to evenly cover the tissue section without creating bubbles to ensure uniform coverage.

[0138] i) Seal the slices and place them in a 42°C oven for 15 min to decolorize.

[0139] j) Open the sliding seal and remove all hydrochloric acid from the edges.

[0140] k) Add 150 μL of diluted de-crosslinking buffer along the side of the wells and aspirate the diluted de-crosslinking buffer.

[0141] l) Add 100 μL of dilution de-crosslinking buffer along the side of the well.

[0142] m) Reseal the sections and place the box at 95°C for 60 min to de-crosslink and cool at 22°C for 10 min.

[0143] 4. Construction of spatial transcriptome library

[0144] 1) Probe construction:

[0145] The Visium Spatial Gene Expression Slide has 2 capture zones. Each capture zone is 6.5 x 6.5 mm 2 , defined by the reference frame (reference frame + capture area is 8×8mm 2 ). The capture zone has approximately 5,000 gene expression spots, and the primers for each expression spot include:

[0146] Illumina TruSeq Read 1 sequencing primer;

[0147] 16nt spatial barcode (all primers at a specific position share the same spatial barcode);

[0148] 12nt unique molecular identifier (UMI);

[0149] 30 nt poly(dT) sequence (capture ligation product);

[0150] The probe set used is: Visium Human Transcriptome Probe Set v2.0.

[0151] 2) Probe hybridization:

[0152] a) Prepare the prehybridization mix just before use. Prepare the probe hybridization solution (from the Visium Human Transcriptome Probe Kit v2-Small) by adding reagents in the order listed, keeping at room temperature, vortexing, and centrifuging.

[0153]

[0154]

[0155] b) Remove the Visium cassette, remove the Visium slide, and remove all de-crosslinking buffer from the wells.

[0156] c) Add 100 μL of prehybridization mixture along the edge of the well to evenly cover the tissue section without introducing bubbles. Seal the Visium slide and incubate at room temperature for 15 minutes.

[0157] d) Remove all pre-hybridization mixture from the wells and add 100 μL of room temperature probe hybridization solution to each well. Seal the Visium slide and incubate at 50°C for 16 to 24 hours for probe hybridization.

[0158] 3) Probe connection:

[0159] a) Preheat FFPE Post-Hyb Wash Buffer (495 μL / each sample) to 50°C.

[0160] b) Prepare 2×SSC buffer.

[0161]

[0162] c) After hybridization is complete, remove the slide seal and remove all probe hybridization mixture from the wells.

[0163] d) Immediately add 150 μL of pre-warmed FFPE Post-Hyb Wash Buffer to each well, seal the Visium slide, and incubate at 50°C for 5 min.

[0164] e) Remove the slide seal and remove all FFPE Post-Hyb Wash Buffer from the wells.

[0165] f) Repeat the above two steps twice, for a total of three washes.

[0166] g) Add 150 μL of 2×SSC buffer to each well and wait for the Visium slide to cool to room temperature (about 3 min) before proceeding to the next step.

[0167] h) Prepare the probe ligation mix and place on ice.

[0168]

[0169] i) Remove all 2×SSC buffer from all wells. Add 60 μL of probe ligation mixture along the side of the wells, seal the Visium slide, and ligate at 37°C for 1 hour.

[0170] j) Remove the Visium slide and remove all probe ligation mixtures from all wells. Immediately add 100 μL of post-ligation wash buffer to each well, seal and incubate at 57°C for 5 minutes.

[0171] k) Remove the Visium slide and remove all post-ligation wash buffer from all wells. Immediately add 100 μL of post-ligation wash buffer to each well, seal and incubate at 57°C for 5 minutes.

[0172] l) Remove the Visium slide and remove all post-ligation wash buffer from all wells. Add 150 μL 2×SSC buffer to each well and wash twice.

[0173] m) Allow the slide to cool to room temperature and proceed to the next step.

[0174] 4) RNA digestion and tissue removal

[0175] a) Ensure the Visium CytAssist is powered on, clean, and ready to run an experiment.

[0176] b) Prepare the probe release mixture as soon as possible before use. The probe release mixture is viscous. Mix well with the pipette until the solution is homogeneous. Avoid creating bubbles. Keep on ice.

[0177]

[0178] c) Prepare 1× PBS solution and keep it at room temperature.

[0179]

[0180] d) Prepare 10% eosin solution before use, vortex to mix and centrifuge.

[0181]

[0182] e) Using a pipette, remove all of the 2×SSC buffer from each well of the tissue slide cassette.

[0183] f) Remove the tissue slide from the Visium CytAssist Tissue Slide Box.

[0184] g) Add 150 μL of 10% eosin solution to evenly cover each tissue section.

[0185] h) Incubate at room temperature for 1 min.

[0186] i) Place the slide at an angle over a waste container to remove the 10% eosin solution.

[0187] j) Place the slide over a liquid waste container and rinse with 1 mL of 1× PBS. Do not pipette directly onto a paper towel.

[0188] k) Repeat the previous step twice, washing three times in total.

[0189] l) Load the Visium CytAssist tissue slide into the groove of the Visium slide stage and close the Visium slide lock.

[0190] m) Remove the probe release mixture from ice. Pipette mix the tissue removal enzyme and centrifuge briefly. Add 4.7 μL of tissue removal enzyme to 70.3 μL of probe release mixture and pipette mix. Immediately add the mixture to each of the spaced wells on the VisiumCytAssist tissue slide, adding 25 μL to each well. The time between adding the tissue removal enzyme and the probe release mixture and starting the Visium CytAssist instrument run should be less than 5 minutes.

[0191] n) Close the lid of the Visium CytAssist Spatial Transcriptome Analysis System and incubate at 37°C for 30 min.

[0192] o) Prepare the probe extension mix, mix by pipetting, and keep on ice.

[0193]

[0194] p) Place the Visium CytAssist slide over a waste container and rinse each section of the slide surrounded by the spacer with 1 mL of 2× SSC. Do not pipette directly into the collection area surrounded by the reference frame.

[0195] q) Repeat the previous step two more times, washing each capture zone three times in total. Place the Visium CytAssist slide into a new Visium sealed box and immediately perform probe extension.

[0196] 5) Probe extension:

[0197] a) Add 75 μL of probe extension mix to each well. Tap the Visium seal to ensure even coverage of the capture area.

[0198] b) Seal the slide and incubate at 45°C for 15 min to extend the probe. After extension, the sample can be stored at 4°C for 24 hours.

[0199] 6) Probe elution:

[0200] a) Prepare 0.08 M KOH Mix, vortex briefly and centrifuge, and keep at room temperature.

[0201]

[0202] b) After the probe is extended, remove the Visium seal cartridge from the adapter and place it on a flat, clean work surface.

[0203] c) Remove the Visium Slide Seal and use a pipette to aspirate all of the Probe Extension Mixture from the wells.

[0204] d) Add 100 μL of 2×SSC buffer to each well. Aspirate all the 2×SSC buffer.

[0205] e) Add 50 μL of 0.08 M KOH Mix to each well and gently tap the Visium seal to ensure even coverage of the capture area.

[0206] f) Incubate at room temperature for 10 min.

[0207] g) All solutions of each sample containing the probe were transferred to 8 tube strips, 3 μL of 1 M Tris-HCl (pH 8.0) was added to each sample, vortexed and centrifuged, and then placed on ice.

[0208] 7) Pre-amplification and magnetic bead screening based on SPRI technology:

[0209] a) Prepare the pre-amplification mix on ice. Add reagents in the order listed. Mix briefly, centrifuge and place on ice.

[0210]

[0211] b) Add 47 μL of pre-amplification mix to the samples in the 8-tube strip, mix briefly and centrifuge.

[0212] c) Incubate in a thermal cycler (heat lid temperature 105°C, total volume 100 μL) according to the following protocol.

[0213]

[0214] d) Vortex to resuspend SPRIselect reagent. Add 120 μL of SPRIselect reagent to each tube and mix by pipetting.

[0215] e) Incubate at room temperature for 5 min.

[0216] f) Place the eight-tube strip on a magnetic rack until the liquid becomes clear.

[0217] g) Remove the supernatant. Add 300 μL of 80% ethanol to the remaining pellet. Wait 30 seconds and remove the ethanol.

[0218] h) Add 200 μL of 80% ethanol to the pellet. Wait 30 seconds and remove the ethanol.

[0219] i) Briefly centrifuge and place on a magnetic stand. Remove any remaining ethanol without disturbing the beads. Air dry for 2 minutes. Do not exceed 2 minutes, as this will reduce elution efficiency.

[0220] j) Remove the magnetic stand and add 105 μL of buffer EB and mix well with a pipette.

[0221] k) Incubate at room temperature for 2 min.

[0222] l) Place the eight-tube strip on a magnetic rack until the solution is clear.

[0223] m) Transfer 100 μL of the sample to a new tube and store at 4°C for up to 72 hours, or at -20°C for up to 4 weeks, or proceed to the next step.

[0224] 8) Library construction:

[0225] a) Prepare qPCR Mix on ice, add reagents in order, vortex and briefly centrifuge, and place on ice.

[0226]

[0227] b) Add 9 μL of qPCR Mix to each well of the qPCR plate (possibly including negative control wells).

[0228] c) Use Nuclease-free water to dilute the sample screened by SPRI technology magnetic beads 5 times, transfer 1 μL of the diluted sample to the qPCR plate well containing qPCR Mix, and if a negative control is used, add 1 μL of Nuclease-free water to the corresponding well, and record the sample position. After mixing with a pipette, cover with qPCR sealing film and centrifuge briefly.

[0229] d) Prepare the qPCR system according to the following scheme and start the program (heated lid temperature 105°C, total system 100 μL).

[0230]

[0231] e) Record the Cq value for each sample.

[0232] f) Select an appropriate set of sample indices to ensure that there is no sample index overlap during multiple sequencing.

[0233] g) Prepare 10x amplification mix.

[0234]

[0235] h) Take 25 μL of the sample from the pre-amplification mixture and add it to the aliquot tubes. Add 70 μL of the amplification mixture to each tube. Add 5 μL of Dual Index Plate TS Set A to each tube. Mix well and centrifuge.

[0236] i) Incubate in a thermal cycler (heat lid temperature 105°C, total volume 100 μL) according to the following protocol.

[0237]

[0238] 5. Sequencing

[0239] Processing method: Visium V4 Slide-FFPE v2.

[0240] Probe set name: Visium Human Transcriptome Probe Set v2.0.

[0241] Number of target genes: 18085.

[0242] Reference transcriptome: GRCh38-2020-A.

[0243] Process version: spaceranger-2.0.0 / spaceranger-2.0.1.

[0244] The Visium chip information used in library construction is as follows:

[0245]

[0246] 6. Data Analysis

[0247] 1) Use Space Ranger to process Visium ST raw-seq data.

[0248] a) Read raw sequencing data (raw-seq) and input the FASTQ file of the Visium spatial transcriptome and the corresponding reference genome.

[0249] b) Conduct quality control.

[0250] c) Alignment to a reference genome.

[0251] d) Quantify the expression of each gene at each spatial location.

[0252] 2) Data scaling and normalization are performed using the scTransform() function.

[0253] a) Load Seurat and the necessary spatial transcriptome sequencing result dataset.

[0254] b) Create a Seurat object using the output file (expression matrix) from Space Ranger.

[0255] c) Run the scTransform() function to calculate the normalized expression of each gene.

[0256] 3) Data integration was performed through canonical correlation analysis (CCA).

[0257] a) Prepare the second Seurat object.

[0258] b) Use the FindIntegrationAnchors() function to find the integration anchors between two Seurat objects.

[0259] c) Integrate the two Seurat objects together using the IntegrateData() function.

[0260] 4) Dimensionality reduction by PCA and U-MAP: Run the RunPCA() function and RunUMAP() function to reduce the dimensionality of the integrated dataset.

[0261] 5) Unsupervised clustering was performed via FindNeighbors() and FindClusters() (resolution = 0.3).

[0262] 6) Identify differently expressed spots based on the Seurat package: Use the FindAllMarkers() function or the FindMarkers() function to identify genes specifically expressed in each cluster and display them visually.

[0263] 7) Seurat and mistyR packages were used for spatial colocalization and cellular niche analysis.

[0264] To further verify the difference between high and low risk in the CDRscore model, we performed ST detection on LSCC (laryngeal squamous cell carcinoma) samples (n=2) and normal laryngeal control samples (n=1) based on 10×GenomicsVisium. We annotated the cell types of the samples using a deconvolution algorithm. The frequencies of epithelial cells and fibroblasts in normal laryngeal tissue were high, while malignant cells, T cells, and B cells were significantly increased in LSCC samples. Except for MRPL10, which was too low to be detected, CDRsignature was expressed in both LSCC and normal tissues ( Fig. 9 ).

[0265] We analyzed the correlation between cells in each spot and the mode of cell death and found that tumors and normal tissues showed different types and levels of cell death, and malignant cells showed the strongest correlation with risk scores ( Fig.10 A). At the same time, dimensionality reduction clustering of tumor samples can divide cells into four different cell subpopulations ( Fig.10 B), the CDRscore model of the present invention was used to calculate the risk scores of different cell subsets and found that the risk scores of Cluster1 and Cluster2 were significantly higher than those of other cell subsets ( Fig.10 C). Combined with the analysis of the proportion of cell types in different cell subpopulations ( Fig.10 D) It was found that the cell subset composed mainly of malignant cells had the highest risk score.

[0266] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. Application of a programmed cell death characteristic gene detection reagent in the preparation of head and neck tumor screening and / or prediction products, characterized in that: The programmed cell death characteristic genes include the following 10 genes: MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3 and MAP2K7.

2. The use according to claim 1, characterized in that The detection reagent is a reagent for detecting the expression amount of programmed cell death characteristic gene.

3. A method for constructing a model for predicting prognosis or treatment response of head and neck tumors, characterized in that: The following steps are involved: Step 1: Using the data from the scRNA-seq dataset related to head and neck tumors, based on the genes related to known cell death patterns, differentially expressed genes related to cell death patterns in head and neck tumors were obtained through differential expression analysis; Step 2: Analyze the differentially expressed genes related to cell death patterns in head and neck tumors to obtain multiple programmed cell death genes related to the prognosis of head and neck tumors; then screen the multiple programmed cell death genes related to the prognosis of head and neck tumors to obtain characteristic genes that characterize programmed cell death of head and neck tumors; Step 3: Based on the characteristic genes that characterize programmed cell death of head and neck tumors, their respective expression levels and corresponding regression coefficients were analyzed to construct a risk scoring model for head and neck tumors.

4. The method for constructing a head and neck tumor prognosis or treatment response prediction model according to claim 3, characterized in that: In step 1, the scRNA-seq datasets related to head and neck tumors are 6 datasets from the Gene Expression Omnibus database: GSE103322, GSE148673, GSE150321, GSE162025, GSE172577, and GSE181919.

5. The method for constructing a head and neck tumor prognosis or treatment response prediction model according to claim 3, characterized in that: Step 2 includes: (1) Based on step 1, univariate Cox analysis was used to identify programmed cell death genes associated with the prognosis of head and neck cancer from the gene expression profiles of cell death patterns on the basis of the TCGA-HNSC cohort; (2) Select the algorithm combination with the highest average C-index in the training set and validation set from a variety of machine learning language algorithm combinations as the optimal model, and use LASSO regression to select the variables in the optimal model to screen out candidate genes; (3) Performing multifactorial stepwise Cox regression analysis on the candidate genes in step (2) to determine the characteristic genes that characterize programmed cell death in head and neck tumors.

6. The method for constructing a head and neck tumor prognosis or treatment response prediction model according to claim 5, characterized in that: The cell death modes described in (1) include alkaliptosis, anoikis, apoptosis, autophagy, autosis, cuproptosis, disulfidptosis, entotic cell death, ferroptosis, immunogenic, lysosomal cell death, mitotic cell death, mpt, necroptosis, netotic cell death, oxeiptosis, parthatos, and pyroptosis.

7. The method for constructing a head and neck tumor prognosis or treatment response prediction model according to claim 3, characterized in that: Signature genes that characterize programmed cell death in head and neck tumors include: MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3, and MAP2K7.

8. The method for constructing a head and neck tumor prognosis or treatment response prediction model according to claim 3, characterized in that: The model is as follows: CDRscore=(4.985107*10 -4 )*NDFIP1 EXP +(-1.068404*10 -03 )*MAP2K7 EXP +(1.441470*10 -04 )*SLC2A3 EXP +(1.123539*10 -04 )*EFNB2 EXP +(3.678527*10 -04 )*HPRT1 EXP +(8.106291*10 -04 )*DDX19A EXP +(-3.859801*10 -04 )*BTG3 EXP +(8.725573*10 -05 )*HK1 EXP +(4.235660*10 -04 )*PCMT1 EXP +(8.672*10 -04 )*MRPL10 EXP ; Among them, the numbers in brackets represent the expression levels of characteristic genes that characterize programmed cell death in head and neck tumors. The number after the brackets is the name of the gene. The number obtained by adding the values ​​is the value of the programmed cell death risk score.

9. A device for predicting prognosis or treatment response of head and neck tumors, characterized in that: The device comprises: A module for acquiring data of characteristic genes of programmed cell death characterizing head and neck tumors, which is used to acquire the expression levels of characteristic genes of programmed cell death characterizing head and neck tumors, wherein the characteristic genes of programmed cell death include MRPL10, DDX19A, NDFIP1, PCMT1, HPRT1, SLC2A3, EFNB2, HK1, BTG3 and MAP2K7; The calculation module is used to calculate the CDRscore value of each tumor tissue sample. Specifically, the CDRscore value of each tumor tissue sample is calculated based on the cell death-related risk scoring model: CDRscore=(4.985107*10 -4 )*NDFIP1 EXP +(-1.068404*10 -03 )*MAP2K7 EXP +(1.441470*10 -04 )*SLC2A3 EXP +(1.123539*10 -04 )*EFNB2 EXP +(3.678527*10 -04 )*HPRT1 EXP +(8.106291*10 -04 )*DDX19A EXP +(-3.859801*10 -04 )*BTG3 EXP +(8.725573*10 -05 )*HK1 EXP +(4.235660*10 -04 )*PCMT1 EXP +(8.672*10 -04 )*MRPL10 EXP ; The numbers in brackets represent the expression levels of characteristic genes that characterize programmed cell death in head and neck tumors, and the numbers after the brackets are the names of the genes; The output module is used to output the risk type of the tumor tissue sample according to the CDRscore value of each tumor tissue sample.

10. A device for predicting prognosis or treatment response of head and neck tumors according to claim 9, characterized in that: The output module divides patients into high CDRscore group and low CDRscore group according to the optimal cutoff value determined by the survminer package.

Images