SIS model and application thereof in adrenal cortex cancer prognosis

Abstract

The invention discloses an SIS model and application thereof in adrenal cortex cancer prognosis. According to the invention, an AI-driven pathological analysis and multi-omics integration method is adopted for the first time, the ACC is systematically classified, an ACC typing system based on a tumor microenvironment is constructed based on classification, and molecular mechanisms, drug resistance characteristics and potential treatment targets of different subtypes are disclosed. Furthermore, a multi-dimensional drug prediction strategy is combined, candidate drugs for different ACC subtypes are screened out, and a new thought is provided for precise treatment.

Description

Technical Field

[0001] This invention relates to the field of biomedicine, specifically to the SIS model and its application in the prognosis of adrenocortical carcinoma. Background Technology

[0002] Adrenocortical carcinoma (ACC) is a rare and malignant endocrine tumor with an incidence of 0.5-2.0 cases per million people, and it is more common in women. Patient prognosis is closely related to tumor stage; the 5-year survival rate for stages I-III exceeds 50%, while for stage IV it is only 13%. However, due to the heterogeneity and aggressiveness of ACC, the overall 5-year survival rate is only 16% to 47%. Although surgery remains the primary treatment and significantly improves survival, local or metastatic recurrence after surgery is common, mostly occurring within two years. For patients with advanced and metastatic ACC, mitotane is currently the only FDA-approved first-line drug, but its efficacy is limited by the long time to achieve target drug concentrations and serious adverse reactions. Therefore, existing treatment options are insufficient to meet the clinical needs of ACC patients, and there is an urgent need for more precise molecular subtyping and targeted therapy strategies to optimize individualized treatment plans and improve prognosis.

[0003] In recent years, with the development of precision medicine, pathological analysis has gradually evolved from traditional morphological observation towards digitalization and intelligence. The application of high-resolution whole-slide images (WSIs) has greatly improved the efficiency of pathological data acquisition and storage, laying the foundation for the application of artificial intelligence (AI) in pathology. AI-based pathological analysis can mine new imaging features from large-scale data and integrate them with molecular omics data to improve the accuracy of tumor subtyping. However, existing AI pathological analysis is still limited to a single modality and has failed to fully integrate molecular-level information. At the same time, although genomics research has revealed some driving molecules of ACC, its understanding of the tumor microenvironment (TIME) and drug resistance mechanisms is still not comprehensive due to the difficulty of bulk RNA sequencing in resolving cellular heterogeneity. Therefore, single pathological analysis or molecular omics studies cannot fully characterize the molecular features and biological behavior of ACC, limiting the application of precision medicine in this field. Summary of the Invention

[0004] In view of the shortcomings of existing technologies, this invention is the first to employ an AI-driven pathological analysis and multi-omics integration approach to systematically classify ACC. Through multimodal fusion, this invention not only refines the molecular subtyping of ACC based on pathological characteristics, but also utilizes single-cell sequencing (scRNA-seq) to analyze the cellular heterogeneity of ACC, overcoming the limitations of bulk sequencing methods. Furthermore, this invention constructs an ACC subtyping system based on the tumor microenvironment, revealing the molecular mechanisms, drug resistance characteristics, and potential therapeutic targets of different subtypes. Further, by combining a multi-dimensional drug prediction strategy, candidate drugs targeting different ACC subtypes are screened, providing new insights for precision medicine.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] The first aspect of this invention provides the application of a reagent for detecting the expression level of biomarkers in a sample in the preparation of prognostic products for adrenocortical carcinoma, wherein the biomarkers are a combination of CYP17A1, ABAT, CD52, CORO1A, CXCL12, HCLS1, PLEK, C1QB, SIGLEC1, MS4A4A, F13A1, CD163, AIF1, C1QA, CXCL9, FOLR2, CXCL10, and FCERT1G.

[0007] Furthermore, the reagents include reagents for detecting biomarker mRNA levels and / or reagents for detecting biomarker protein levels.

[0008] Furthermore, the reagents for detecting biomarker mRNA levels include reagents used in any of the following methods: PCR-based detection methods, Southern hybridization methods, Northern hybridization methods, dot hybridization methods, fluorescence in situ hybridization methods, DNA microarray methods, ASO methods, and high-throughput sequencing platform methods.

[0009] Furthermore, the reagents for detecting the levels of biomarker proteins include those used in any of the following methods: hematoxylin-eosin staining, safranin O-fast green staining, Western blotting, enzyme-linked immunosorbent assay (ELISA), radioimmunoassay, sandwich assay, immunohistochemical staining, mass spectrometry, immunoprecipitation analysis, complement fixation analysis, flow cytometry fluorescence sorting, and protein chip analysis.

[0010] Furthermore, the detection is performed on samples from the subject.

[0011] Furthermore, the subject refers to any individual of interest, preferably a living organism suffering from or suspected of having adrenocortical carcinoma, including humans, other mammals, preferably primates, and particularly preferably humans.

[0012] Furthermore, the samples include tumor tissue, tumor tissue extract, and tumor cell culture.

[0013] Furthermore, the product also includes sample pretreatment reagents.

[0014] Furthermore, the product assesses the prognosis of subjects by detecting the expression levels of biomarkers in the obtained samples. Low expression levels of the biomarkers indicate low lymphocyte infiltration in adrenocortical carcinoma, and the subject has a poor prognosis; high expression levels of the biomarkers indicate high lymphocyte infiltration in adrenocortical carcinoma, and the subject has a better prognosis.

[0015] Furthermore, the product includes a reagent kit, a chip, a test strip, or a device, equipment, or computer-readable storage medium.

[0016] The second aspect of the present invention provides a method for constructing a prognostic model for adrenocortical carcinoma, the method comprising the steps of: obtaining biomarker expression level data and clinical characteristics of subjects as described in the first aspect of the present invention in a sample, and constructing a prognostic model based on the biomarker expression level data and clinical characteristics using an algorithm.

[0017] The biomarker expression level data can be mRNA expression level or protein expression level.

[0018] The clinical characteristics include individuals with a better prognosis and individuals with a poorer prognosis for adrenocortical carcinoma.

[0019] Furthermore, the clinical features also include pathological image features.

[0020] Furthermore, the algorithm includes one or more of the following: principal component analysis, deep neural network, generalized linear model, logistic regression analysis, LASSO regression analysis, nearest neighbor analysis, support vector machine, neural network model, and random forest model.

[0021] Furthermore, the algorithm is principal component analysis.

[0022] Furthermore, the method includes the following steps: obtaining the expression level data of the biomarkers described in the first aspect of the present invention in the sample, performing unsupervised clustering on the sample, extracting the first principal component using principal component analysis as the final scoring basis, and calculating the SIS score by aggregating the principal component scores representing SIS gene feature A and the principal component scores representing SIS gene feature B.

[0023] Furthermore, the method for judging the prognostic model is as follows: based on the constructed prognostic model, an SIS score is obtained. If the SIS score is higher than the cutoff value, a classification result indicating a better prognosis for the subject is output. If the SIS score is lower than the cutoff value, a classification result indicating a worse prognosis for the subject is output.

[0024] In this invention, the term "cutoff value" refers to a value that is statistically relevant to a particular outcome when compared with the analysis results. In a preferred embodiment, the cutoff value is determined based on statistical conclusions from studies comparing populations with better and worse prognoses for adrenocortical carcinoma. Some such studies are shown in the Examples section of this document, but studies from the literature and the experience of users of the methods described herein can also be used to generate or adjust cutoff values. The cutoff value can also be determined by taking into account the patient's genetic background, clinical characteristics, work environment, and other relevant factors and outcomes.

[0025] A third aspect of the present invention provides a method for predicting the sensitivity of a subject to a treatment drug for adrenocortical carcinoma based on the SIS score, the method comprising the following steps: obtaining the subject's SIS score, the SIS score being obtained by a model constructed by the method described in the second aspect of the present invention; comparing the SIS score with a cutoff value to determine whether the subject belongs to the high SIS group or the low SIS group.

[0026] If a subject belongs to the high SIS group, it is predicted that the subject will be sensitive to mTOR inhibitors and PI3K / Akt / mTOR pathway regulators.

[0027] If a subject belongs to the low SIS group, the subject is predicted to be sensitive to PLK1 inhibitors, calmodulin antagonists, dopamine receptor antagonists, squalene cyclase inhibitors, serotonin receptor antagonists, sterol demethylase inhibitors, acetylcholine receptor antagonists, BCR-ABL kinase inhibitors, and norepinephrine reuptake inhibitors.

[0028] Furthermore, the PLK1 inhibitor is BI-2536 or a pharmaceutically acceptable salt thereof.

[0029] Furthermore, the acetylcholine receptor antagonist is mebivelin or a pharmaceutically acceptable salt thereof.

[0030] Furthermore, the BCR-ABL kinase inhibitor is imatinib or a pharmaceutically acceptable salt thereof.

[0031] Furthermore, the norepinephrine reuptake inhibitor is maprotiline or its pharmaceutically acceptable salt.

[0032] Furthermore, the mTOR inhibitors, PI3K / Akt / mTOR pathway regulators, PLK1 inhibitors, calmodulin antagonists, dopamine receptor antagonists, squalene cyclase inhibitors, serotonin receptor antagonists, sterol demethylase inhibitors, acetylcholine receptor antagonists, BCR-ABL kinase inhibitors, and norepinephrine reuptake inhibitors include those that are artificially synthesized or naturally occurring.

[0033] In this invention, the pharmaceutically acceptable salt refers to a salt of an active compound (such as BI-2536) and is prepared by reacting the active compound with a suitable organic or inorganic acid or acid derivative. Pharmaceutically acceptable salts include, but are not limited to, hydrochlorides, sulfates, phosphates, citrates, hydrobromides, acetates, benzoates, benzenesulfonates, tartrates, carbonates, citrates, gluconates, lactates, malates, methanesulfonates, stearates, valerates, nitrates, sodium salts, calcium salts, potassium salts, zinc salts, and meglumine salts.

[0034] The fourth aspect of the present invention provides a computer system for predicting the prognosis of adrenocortical carcinoma. The computer system includes an analysis unit that analyzes an adrenocortical carcinoma prognosis model constructed using the method described in the second aspect of the present invention to obtain an SIS score. The SIS score is compared with a cutoff value, and a higher SIS score indicates a better prognosis, while a lower SIS score indicates a worse prognosis.

[0035] Furthermore, the computer system also includes an input unit and an output unit.

[0036] Furthermore, the input unit is used to input the biomarker expression level data of the first aspect of the present invention in the sample, wherein the biomarker expression level may be the mRNA expression level or the protein expression level.

[0037] Furthermore, the output unit is used to output the classification result of the subject corresponding to the sample, indicating whether the prognosis is good or bad.

[0038] The fifth aspect of the present invention provides a computer device for predicting the prognosis of adrenocortical carcinoma, the computer device including a memory and a processor.

[0039] The memory is used to store program instructions.

[0040] The processor is used to execute program instructions, which, when executed, are used to perform the following operations: obtain the expression level data of the biomarkers described in the first aspect of the present invention in the sample, input the data into the adrenocortical carcinoma prognostic model constructed by the method described in the second aspect of the present invention for analysis, obtain the SIS score, compare the SIS score with the cutoff value to obtain the classification result of the sample; those with higher SIS scores have better prognoses, and those with lower SIS scores have worse prognoses.

[0041] Furthermore, the computer device may include: a display device for displaying information to a user; and a keyboard and pointing device (e.g., a mouse) through which the user provides input to the computer. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including voice input, speech input, or tactile input).

[0042] Advantages and benefits of this invention: This study screened metabolic and immune synergistic regulatory gene clusters through genomic analysis and, combined with digital pathology and artificial intelligence technologies, explored the molecular characteristics of ACC in depth. We proposed a novel molecular classification method based on SIS, revealing the deep molecular heterogeneity of ACC and providing new ideas for personalized treatment. Furthermore, we are the first to reveal the interaction between ACC tumor cells and immune cells through single-cell RNA sequencing, discovering a new immune escape mechanism and providing a theoretical basis for immunotherapy. Integrating multiple datasets and performing multi-dimensional analysis, we found that patients in the high SIS group with better prognosis responded better to ICB, while patients in the low SIS group with poorer prognosis were more suitable for anti-inflammatory hormone therapy. Attached Figure Description

[0043] Figure 1 The diagram shows the tumor immune microenvironment cluster subtype classification of ACC. A) is a heatmap showing the distribution of three different TIME cluster subtypes in 24 immune cells in ACC patients (TCGA+GSE76019+GSE76021); b) is a principal component analysis diagram of the TIME cluster; c) is a diagram showing the difference in the abundance of 24 immune cells in the TIME cluster subtypes; d) is a Kaplan-Meier curve used to predict the overall survival of ACC patients in the TIME cluster subtypes; and eh) is a diagram showing the differences in the expression of CTLA4(e), PD-L1(f), PD-L2(g), and PD-1(h) in ACC patients within the TIME cluster subtypes.

[0044] Figure 2The following graphs illustrate the results of unsupervised clustering analysis to classify ACC patients into Gene clusters A and B based on DEGs. Figure a shows the 18 common DEGs identified from the TIME clusters; figure b is a heatmap of the consensus clustering matrix of DEG-related molecular patterns in TCGA+GSE76019+GSE76021 when k=2; figure c is a heatmap of the consensus clustering matrix of DEG-related molecular patterns in TCGA+GSE76019+GSE76021+GSE3371+GSE10927 when k=2; figure d is a principal component analysis plot showing the distribution of Gene clusters; figure e is a heatmap showing the expression distribution of DEGs; figure f is a heatmap showing the expression distribution of DEGs in different cell types during single-cell RNA sequencing; figure g is a Kaplan-Meier curve used to predict the overall survival of ACC patients in Gene clusters; figure h is a graph showing the differential levels of 24 immune cell infiltration abundances in Gene clusters; and figure i is a heatmap showing the Gene clusters. Figure showing the differential expression of CTLA4, PD-L1, PD-L2, and PD-1 in ACC patients in the cluster subtype.

[0045] Figure 3 This section presents the validation and application of SIS in deep learning, where a) represents the segmentation and background filtering of pathological images; b) represents the color normalization of patches; c) represents the training of a deep learning model using patches from two patient classes in the training set; d) represents the testing of all patches from patients in the test set using the trained model and the statistical results of patient classification; e) represents the five-fold validation AUC plot of ResNet50; f) represents the five-fold validation AUC plot of Vision Transformer; g) represents the spatial patterns related to SIS prediction revealed by WSIs from high-SIS and low-SIS tumor patients in the TCGA test set; h) represents the correlation between lymphocyte infiltration and SIS subtype in WSIs discovered using the model weight visualization method CAMs; and i) represents the degree of lymphocyte infiltration found in the external validation set of WSIs.

[0046] Figure 4This section presents the application and comparison of artificial intelligence in the SIS and C1A / B groupings, where a) ResNet50 and VisionTransformer evaluate the classification performance of SIS; b) ResNet50 and VisionTransformer evaluate the classification performance of C1A / B; c) Five-fold validation AUC plots of ResNet50 and VisionTransformer for C1A / B groupings; d) Pathological features of CAMs associated with Weiss scores in the SIS grouping; e) Pathological features of CAMs associated with Weiss scores found in the external validation set; f) Survival prognosis of patients in the validation set; and g) Survival prognosis of patients in the validation set and TCGA patients.

[0047] Figure 5 The establishment of steroid-related immune score (SIS) groupings and their clinical relevance are shown in the following figures: a) Distribution of different subtypes and pathological parameters of ACC among SIS subgroups; b) Distribution of SIS in COC subtypes; c) Distribution of SIS in C1A / C1B subtypes; d) Distribution of SIS in immune subtypes; e) Kaplan-Meier curve for predicting overall survival of SIS (TCGA+GSE76019+GSE76021); f) Kaplan-Meier curve for predicting overall survival of SIS (TCGA+GSE76019+GSE76021+GSE3371+GSE10927); g) Likert plot showing the SIS subgroups in WeissScoring. Distribution in System; hj is the distribution of clinical characteristics (j) in univariate (h) and multivariate (i) analysis of SIS subgroups in clinical presentation; kn is the Kaplan-Meier curve used to predict the overall survival of ACC patients in Ki-67 (k), EMT (l), TMB (m) and mRNAsi (n) and in SIS stratification.

[0048] Figure 6 The survival prognosis of the SIS subgroup and the distribution of SIS in different ACC subtypes are shown in Figure 1. a) is the Kaplan-Meier curve used to predict the overall survival of SIS (TCGA); b) is the Kaplan-Meier curve used to predict the overall survival of SIS (GSE76019+GSE76021); c) is the Kaplan-Meier curve used to predict the overall survival of SIS (GSE76019+GSE76021+GSE3371+GSE10927); d) is the distribution of SIS in the subgroup of expression (K=4); and e) is the distribution of SIS in methylation typing.

[0049] Figure 7The first part of the graph shows the correlation between SIS and steroid hormone production. Here, ab represents the enrichment analysis of the KEGG (a) and HALLMARK (b) gene sets in the SIS subgroups; c represents common genes screened from the core genes of the two enriched pathways; d represents the distribution of ADS, hormone, and cortisol levels across the SIS subgroups; e represents the correlation between SIS and ADS; f represents the distribution of SIS in different hormones and cortisol; g represents the distribution of ADS in different hormones and cortisol; h represents the correlation between the expression of steroid hormone-related genes and the main target genes of mitotane with SIS and ADS, comparing expression values ​​with low SIS, high ADS, presence of hormones, and presence of cortisol; i represents the genes most associated with SIS screened using Lasso regression, RF, and SVM-RFE machine learning; and j represents the Kaplan-Meier curve used to predict the overall survival of DHCR7.

[0050] Figure 8 The second part of the results on the correlation between SIS and steroid hormone production shows that: a) DHCR7 and mitotane are related to the mechanism pathway; b) DHCR7 protein expression in ACC tissue and normal adrenal tissue detected by IHC; c) DHCR7 expression distribution in single-cell sequencing; d) cell viability assessment in SW-13 cells after treatment with different concentrations of mitotane; and e) cell viability assessment in NCI-H295R cells after treatment with different concentrations of mitotane.

[0051] Figure 9 The graph shows the gene alterations of metabolism-related genes and their prognostic effects. In the graph, a represents the copy number alterations of metabolism-related genes in the SIS subgroup of ACC patients; bd represents the Kaplan-Meier curves used to predict (b) overall survival, (c) disease-specific survival, and (d) progression-free survival of DHCR7 in copy number alterations; eg represents the Kaplan-Meier curves used to predict (e) overall survival, (f) disease-specific survival, and (g) progression-free survival of SOAT1 in copy number alterations; h represents the expression distribution of DHCR7 in tumors and adjacent normal tissues; i represents the expression distribution of SC5D in tumors and adjacent normal tissues; and j represents the Kaplan-Meier curve used to predict the overall survival of SC5D.

[0052] Figure 10The first part of the ACC SIS subgroup immune characteristic results is shown in Figure a. a) Correlation analysis between SIS grouping and tumor microenvironment ESTIMATE score; b) Heatmap of immune cell infiltration in different SIS groups; c) Relationship between SIS and MHC, EC, SC, and CP; d) Distribution and comparison of IPS in SIS subgroups; e) Correlation between SIS and overall activity score; f) Heatmap of immune activity scores in the seven stages of cancer-community circulation under SIS grouping.

[0053] Figure 11 The second part of the immunophenotypic results for the ACC SIS subgroup is shown in Figure a. a) Distribution of five characteristic immune expression scores in the SIS subgroup; b) Correlation analysis of SIS and TIDE in the TCGA cohort for immunotherapy response; c) Correlation analysis of SIS and TIDE in the GEO cohort for immunotherapy response; d) Expression patterns of immune-related genes in the SIS group; e) UMAP analysis of cell type distribution; f) KEGG enrichment pathway analysis of ACC-related exosomes; g) Interaction and expression of ACC exosomes and immune-related cell receptors; h) KEGG enrichment pathway analysis of exosomes and their related receptors.

[0054] Figure 12 The following are the correlation analysis results of SIS grouping and immune characteristics in ACC patients. Figure a shows a heatmap of differential expression patterns in the KEGG pathway among different SIS groups (high SIS and low SIS), with data sourced from the TCGA and GEO projects; figure b shows a heatmap of differential expression in the HALLMARK pathway among different SIS groups (high SIS and low SIS); figure c shows the correlation analysis between SIS (TCGA) and TIDE; figure d shows the correlation analysis between SIS (GEO) and TIDE; figure e shows a heatmap of the correlation analysis between different SIS groups and multiple TIDE-related indicators (such as Dysfunction and Exclusion); figure f shows the differential distribution of DNA damage in SIS subgroups; and figure g shows a Sankey diagram illustrating the relationship between ACC-related survival-immune QTLs and GWAS.

[0055] Figure 13The following are the results of sensitivity drug analysis related to SIS subgroups. Figure a shows a heatmap of GDSC1 and GDSC2 drugs with statistically significant differences in both the TCGA and GEO cohorts within the SIS group, including the pathways affected by these drugs; Figure b shows cMAP drugs consistently associated with SIS from the top 200 SPIED3 drugs associated with SIS, selected from both TCGA and GEO cohorts; Figure c shows cMAP drugs consistently associated with SIS in both TCGA and GEO, selected from those with scores <-90 or >90; Figure d shows four drugs potentially effective for low SIS patients, selected jointly by SPIED3 and cMAP; Figure e shows GO-MF enrichment analysis of genes specifically expressed in tumor cells; and Figure f shows a heatmap of genes associated with iron metabolism and ferroptosis, calcium metabolism, and schizophrenia from single-cell sequencing results. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Example

[0058] I. Experimental Methods

[0059] 1. Adrenal carcinoma data collection

[0060] Gene expression profiles and clinical information for ACC were obtained from the TCGA portal (http: / / cancergenome.nih.gov), with a total of 79 ACC samples used for subsequent analysis. Among them, 55 patients yielded 237 WSIs, all of which were pathological slide images sourced from the TCGA database. Four ACC datasets with prognostic information—GSE33371, GSE10927, GSE76019, and GSE76021—were downloaded from GEO (http: / / www.ncbi.nih.gov / geo). After merging these datasets, the batch effect was removed using the "ComBat" function.

[0061] 2. Calculation of cell abundance in the microenvironment

[0062] We referenced the immune cell features constructed in published articles and used the R package "GSEAbase" to perform single-sample gene set enrichment analysis (ssGSEA) on the expression levels of 24 microenvironment cell subpopulation-related features from three datasets: TCGA, GSE76019, and GSE76021.

[0063] 3. Differentially expressed genes (DEGs) in TIME clusters

[0064] Three common DEGs were identified in the TIME clusters using the R package "limma". DEGs with FDR values ​​< 0.001 and absolute fold changes > 1 were considered significant and used for further analysis.

[0065] 4. Consensus Clustering of DEGs

[0066] To further reveal the biological significance of ACC immunity, we used the "ConsensusClusterPlus" package to calculate the consistency matrix and the consistency cumulative distribution function based on the expression profile data of DEGs to determine the optimal classification (k=2).

[0067] 5. Develop a Steroid-related Immune Score (SIS) solution.

[0068] First, unsupervised clustering was performed on ACC patients based on DEGs, dividing them into two genotype subtypes (Genecluster A and B). Then, based on gene cluster features, the 18 DEGs were further divided into SIS gene feature A (genes positively correlated with cluster features) and SIS gene feature B (genes negatively correlated with cluster features). To improve the stability and interpretability of the model, the Boruta algorithm was used to reduce the dimensionality of SIS gene features A and B, and principal component analysis (PCA) was used to extract the first principal component (PC1) as the final scoring criterion. PC1A represents the principal component score of SIS gene feature A, and PC1B represents the principal component score of SIS gene feature B. Finally, the SIS score was calculated by aggregating PC1A and PC1B (SIS = ∑PC1A + ∑PC1B), a method similar to the Gene Expression Level Index (GGI). Based on the optimal cutoff value (1.041494), 142 ACC patients were divided into a high SIS group and a low SIS group for subsequent analysis. In addition, we provide the PC1 values ​​and their contribution weights involved in the SIS calculation to enhance the transparency and reproducibility of the study.

[0069] 6. Image preprocessing, data partitioning, and external validation

[0070] Image preprocessing included segmentation, color normalization, and filtering of patches with chromatic aberrations to ensure the quality and consistency of the model input data. First, the WSI image was segmented into 256×256 pixel patches at a 10x magnification using the OpenSlide library, and patches with more than 50% background (RGB color values ​​below 220) were removed. Subsequently, color normalization was performed using the Macenko method, and patches with chromatic aberrations were further filtered out to reduce the impact of technical biases. After preprocessing, a total of 931,162 patches were generated, with the number of patches generated for each WSI ranging from 544 to 7609.

[0071] The dataset comprises 55 ACC patients from TCGA (20 in the highSIS group and 35 in the lowSIS group). We employ five-fold cross-validation to optimize the robustness and generalization ability of the model. Data splitting follows a case-level partitioning, with each fold containing 4 highSIS patients and 7 lowSIS patients. During each training iteration, the ratio of training, validation, and test sets is 3:1:1, and the final result is the average of five test results. During training, all slides and patches for each case inherit their case labels, and the classification model is trained at the patch level. Since each case contains multiple slides, and each slide contains thousands of patches, we employ a resampling method for data balancing to reduce the impact of class imbalance. Approximately 24,000 patches are used per fold for training.

[0072] External Validation: To further validate the model's generalization ability, we additionally included 20 ACC patients from the First and Second Affiliated Hospitals of Dalian Medical University as an independent external validation set. This dataset underwent the same preprocessing procedure and inference was performed on the trained model to evaluate its classification performance on independent cohorts.

[0073] 7. Integration of model prediction results, transfer learning, and evaluation metrics

[0074] During the testing phase, the model makes predictions for all patches across all cases in the test set and calculates the predicted class and prediction confidence for each patch. For all patches of the same slide, we calculate the final classification result for that slide using a weighted average of prediction confidence. Furthermore, the final classification result for each patient is determined by the weighted average of all their slides, thereby improving the model's stability and reliability at the case level.

[0075] To mitigate the risk of overfitting from few-shot deep learning training, we employ transfer learning during model training. Specifically, we perform transfer learning based on ResNet50, fine-tuning the pre-trained ImageNet model to leverage existing feature representations and improve generalization capabilities. Furthermore, we explore ViT (Vision Transformer-B16) based on a self-attention mechanism to further evaluate the impact of different model architectures on classification tasks.

[0076] The model's classification performance was evaluated using five-fold cross-validation and quantified using multiple metrics, including accuracy, AUC (area under the receiver operating characteristic curve), recall, and specificity. All results were calculated based on the mean of the five-fold cross-validation test set at the case level and were further evaluated on an external validation set to ensure the robustness and generalization ability of the classification model.

[0077] 8. Patient selection and clinical sample collection

[0078] We retrospectively selected 10 patients with acute myeloid choriocarcinoma (ACC) who underwent surgery at the Department of Urology, Second Affiliated Hospital of Dalian Medical University between January 1, 2012 and March 1, 2024, and 10 patients with ACC who underwent surgery at the Department of Urology, First Affiliated Hospital of Dalian Medical University between January 1, 2012 and March 1, 2024. We obtained hematoxylin and eosin (H&E) slide data and related clinical information from these patients. H&E slides were digitized at 40x magnification (0.25 μm / pixel) using a panoramic digital pathology slide scanner (BL-006, Songming MedicalTech), and image preprocessing, data segmentation, and model prediction validation were performed according to the methods described above. Fresh ACC specimens were collected during surgical resection for subsequent single-cell sequencing. The study was conducted in accordance with the Declaration of Helsinki. All tissue samples included in this study were approved by the Ethics Committee of the Second Affiliated Hospital of Dalian Medical University (No. KY2024-032-01) and the Ethics Committee of the First Affiliated Hospital of Dalian Medical University (No. PJ-KS-KY-2025-27), and written informed consent was provided for each patient.

[0079] 9. Single-cell RNA sequencing and related analysis

[0080] Fresh ACC specimens collected surgically were obtained from the Second Affiliated Hospital of Dalian Medical University and transported to the laboratory on ice using MACS tissue storage medium (Miltenyi Biotec). After preparing single-cell suspensions, cell counting and viability determination were performed using a Countstar Fluorescence Cell Analyzer (Countstar), adjusting the cell concentration to 300-600 cells / μL. The cell suspensions were loaded onto a 10×Genomics Chromium Controller to generate single-cell gel bead emulsions according to the manufacturer's protocol. A single-cell RNA-seq library was constructed using the single cell 3' Library and Gel Bead Kit V3.1 (10xGenomics, 1000121), and sequencing was performed using an Illumina Novaseq 6000. Data were then processed using CellRanger (version 3.0.2), and the gene expression matrix was imported into Seurat (version 3.0) for quality control and downstream analysis. In Seurat 3.0, cells with more than 200 genes and mitochondrial gene expression ≤25% were retained. Dimensionality reduction was performed using PCA, and visualization was achieved via t-SNE. After screening genes based on avgUMI≥1 & p_val_adj≤0.05, the top 5 genes with avg_log2FC were selected as candidate marker genes for this cell subpopulation, sorted from largest to smallest. Genes specifically expressed in Cancer_cell (avg_log2FC>0.4) were selected, and their potential to become exosomes were queried from genecards. Immune cell receptors were obtained from CellphoneDB (v5.0.0) (https: / / www.cellphonedb.org / ). HitPredict (https: / / www.hitpredict.org / # / ) is an experimentally validated protein-protein interaction website; we manually queried the interaction relationships between tumor exosomes and immune cell receptors and generated heatmaps.

[0081] 10. Immunohistochemistry (IHC)

[0082] The sections were sequentially dewaxed by immersing them in xylene, anhydrous ethanol, and ethanol of different concentrations, and then rinsed with water. Microwave antigen retrieval was performed using 1×EDTA (pH 8.0) retrieval solution, followed by heating and allowing the solution to return to room temperature. The sections were incubated in 3% H2O2 at room temperature for 30 minutes, followed by washing with PBS. The tissue was circled with a histochemical pen, and 3% BSA was added for serum blocking at 37°C for 30 minutes. After blocking, the serum was removed, and anti-DHCR7 rabbit polyclonal antibody (1:200, ZEN-BIOSCIENCE, 822232) was added to the tissue and incubated overnight at 4°C. Then, horseradish peroxidase (HRP)-labeled anti-rabbit secondary antibody (JiJia Biotechnology, J0046) was incubated at 37°C for 1 hour, followed by washing three times with PBS. DAB working solution was added for color development, and the specific brown expression was observed under a microscope before rinsing with water. After staining with hematoxylin, differentiating with hydrochloric acid and alcohol, inverting with blue solution, and dehydrating with anhydrous ethanol and n-butanol, the slides were mounted with neutral resin and air-dried. DHCR7 immunohistochemistry of normal adrenal tissue was obtained from THE HUMAN PROTEIN ATLAS (https: / / www.proteinatlas.org / ).

[0083] 11. Calculation of common tumor markers and acquisition of Genomic alterations information

[0084] Common tumor markers include TMB, EMT, and mRNA si. We filtered out germline mutations annotated in the dbsnp and ExAC databases. Then, we defined and calculated the TMB for each sample as the total number of coding variants divided by the length of exons (38 million), where detected variants were considered frameshift deletions, in-frame deletions, frameshift insertions, in-frame insertions, missense mutations, nonsense mutations, discontinuities, and silences. Patients were divided into high and low TMB groups based on median TMB values. We evaluated EMT scores for ACC patients using EMT gene features constructed from published articles. Patients were divided into high and low EMT groups based on median EMT values. The OCLR machine learning algorithm was used to calculate mRNA si for ACC. Patients were divided into high and low mRNA si groups based on median mRNA si values. Genomic alterations information was obtained from cbioportal (https: / / www.cbioportal.org / ), compared according to SIS grouping, and genes with p<0.05 were selected (two-sided Fisher Exact test).

[0085] 12. Gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA)

[0086] We used the R package "GSVA" to perform a comprehensive scoring of the gene set before conducting differential analysis between subgroups, and used GSEA software (version 4.3.0) to perform gene set enrichment analysis between subgroups. The gene set was obtained from the Molecular Signatures Database (h.all.v2022.1.Hs and c2.cp.kegg.v2022.1.Hs).

[0087] 13. Cell culture, transfection, and pharmacological assays

[0088] Human ACC cell lines (SW13, NCI-H295R) were purchased from Procell Life Science & Technology. SW13 cells were cultured in DMEM medium (EallBio, 03.1002C) containing 10% FBS (PAN Seratech, ST30-3302) and 1% P / S (Procell, PB180120). NCI-H295R cells were cultured in DMEM / F12 medium (EallBio, 03.2001C) containing 10% FBS, 1% P / S, and 0.5% insulin-transferrin-selenium supplement (ITS-G) (Procell, PB180429). When cell confluence reached 60-80%, cells were digested and collected. After cell counting, cells were seeded into 96-well plates. After cell attachment, control wells were treated with different concentrations of mitotane, and an equal volume of DMSO was added. After 48 hours of treatment, 100 μL of 10% CCK-8 solution (ApexBio, K1018) was added to each well, and the mixture was incubated at 37 °C for 2 hours. The absorbance was then measured at 450 nm.

[0089] One day before transfection, seed cells in 6-well plates until cell confluence reaches 40-60%. Following the manufacturer's instructions, mix siRNA with Lipofectamine... TM 2000 (Life Technologies) solution was mixed and added to the wells of the plate. After incubation for 6 hours, the medium was replaced with complete medium containing 10% FBS. After 48 hours of culture, cells were collected and DHCR7 expression levels were detected by qRT-PCR to verify the knockdown effect. Transfected cells were counted and seeded into 96-well plates, and treated with the drug at the IC50 concentration determined above. After 48 hours of treatment, 100 μL of 10% CCK-8 solution (ApexBio, K1018) was added to each well, and the plates were incubated at 37°C for 2 hours. The absorbance was measured at 450 nm. The siRNA sequence used is shown below:

[0090] si-DHCR7-1: CCCUGACUUCUGCCAUAAGUU (SEQ ID NO: 1);

[0091] AACUUAUGGCAGAAGUCAGGG (SEQ ID NO: 2).

[0092] si-DHCR7-2:GCCUUAUCUUUACACGCCUGCA (SEQ ID NO: 3);

[0093] UGCAGCGUGUAAAGAUAAGGC(SEQ ID NO:4)

[0094] si-NC (negative control): UUCUCCGAACGUGUCACGUTT (SEQ ID NO:5);

[0095] ACGUGACACGUUCGGAGAATT(SEQ ID NO:6)

[0096] 14. Immune-related indicators

[0097] The ESTIMATE algorithm defines ssGSEA for each patient as a stromal and immune score based on features associated with stromal tissue and immune cell infiltration, and combines the stromal and immune scores to form the ESTIMATE score. We used the R package "estimate" to score each ACC sample. To ensure data consistency, we selected the TumorImmune MicroEnvironment cell composition Database (TIMEDB) (https: / / timedb.deepomics.org / ) to obtain the relative abundance of immune cells in ACC patients under different algorithms (CIBERSORT, ABIS, MCPcounter, xCell, and ImmuCellAI). (TIP)

[0098] The anticancer immune status in the ACC immune cycle was analyzed using the Cancer Immunome Atlas (https: / / tcia.at / ). ACC immunophenoscores (IPS), including MHC molecules, immunomodulators, effector cells (activated CD8+ and CD4+ T cells, Tem CD8+ and Tem CD4+ cells), and suppressor cells (Tregs and MDSCs), were obtained via The Cancer Immunome Atlas (https: / / tcia.at / ). Differences between ACC subgroups were assessed by referencing five identified representative signatures of pan-cancer immunity from published articles. Furthermore, a TIDE score was calculated for each ACC tumor sample using TIDE (http: / / tide.dfci.harvard.edu / faq / ) to predict its response to immune checkpoint blockade. The impact of genetic variations on ACC immune infiltration was explored using CancerImmunityQTL (http: / / www.cancerimmunityqtl-hust.com / ). DNA damage scores were generated from TCGA aneuploid AWG using ABSOLUTE.

[0099] 15. Prediction of SIS-related susceptibility drugs

[0100] The Genomics of Drug Sensitivity in Cancer (GDSC; https: / / www.cancerrxgene.org) is a website used to assess drug sensitivity and response in cancer cells. We used the "oncoPredict" package to assess drug sensitivity estimates for ACC patients to GDSC1 and GDSC2 drugs, selecting drugs with significant differences between SIS subgroups (p<0.05). We input differentially expressed genes (|logFC|>1) from the SIS subgroups into CMap (https: / / clue.io / ) and SPIED3 (http: / / 92.205.225.222 / HGNC-SPIED3-QF.py). In CMap, we selected drugs with a |Score|>90, and in SPIED3, we selected the top 100 drugs with the highest positive and negative correlations.

[0101] 16. Statistical Analysis

[0102] All statistical analyses were performed in R 4.2.1 software. Difference analyses were primarily performed using the "limma" package. For continuous variables, Student's t-test was used; for categorical variables, the chi-square test was used.2 Tests were performed. The nonparametric two-tailed Wilcoxon rank-sum test was used for comparisons between two groups, while the Kruskal-Wallis test was used for comparisons between two or more groups. Correlation was assessed using the Spearman coefficient. The statistical significance of survival rates between different risk groups was calculated using the log-rank test. Univariate and multivariate Cox regression analyses were performed to explore important prognostic factors.

[0103] II. Experimental Results

[0104] 1. Establishment of the TIME subtype

[0105] Recent literature suggests that future ACC research should focus on understanding the tumor microenvironment. Therefore, we used gene set enrichment analysis (GSEA) to analyze 24 microenvironment cell subpopulations and, based on single-sample GSEA (ssGSEA), divided ACC patients (TCGA+GSE76019+GSE76021) into three clusters (…). Figure 1 The results showed that TIME cluster A had the lowest ssGSEA score, while TIME cluster C had the highest score. The differences in scores among the three groups across 23 microenvironment cell subsets (excluding plasma cells) were statistically significant. Patients with TIME cluster B had the best prognosis. Figure 1 d). Furthermore, the expression of four key immune checkpoints (PD-1, PD-L1, PD-L2, and CTLA4) was lowest in TIME cluster A. Figure 1 eh).

[0106] 2. Genotypes based on differentially expressed genes (DEGs)

[0107] We analyzed the differential genes among the TIME clusters and identified 18 common DEGs ( Figure 2 a). Subsequently, we performed unsupervised hierarchical clustering analysis on these 18 genes and found that the clustering stability was optimal when k=2. Figure 2 b). To verify the validity of this classification, we added two independent GEO datasets (GSE33371 and GSE10927) to the original dataset and repeated the cluster analysis. The results show that the classification pattern is consistent with the previously obtained pattern. Figure 2 c). We named these two groups Gene clusters A and B, and plotted expression heatmaps for the 18 DEGs. Figure 2 e). Single-cell sequencing data showed that these 18 genes were highly expressed in specific cell subpopulations and were associated with immunity and metabolism. Figure 2 f). Patients in Genecluster B group had a better prognosis. Figure 2g), this group not only showed high infiltration abundance in 19 microenvironment cell subsets, but also showed significant high expression in four key immune checkpoints (g). Figure 2 Finally, we calculated the Steroid-related Immune Score (SIS) using principal component analysis (PCA) (Table S1). Based on the optimal cutoff value (1.041494), we divided the 142 ACC patients into a high SIS group and a low SIS group.

[0108] 3. Artificial intelligence validation of SIS and related pathological features

[0109] Deep neural networks have become an important tool for medical image analysis, capable of identifying features that are difficult to detect with the naked eye, such as microsatellite instability (MSI), tumor mutational burden (TMB), and gene expression status. This study used deep learning technology to analyze whole-slice images (WSIs), validating the effectiveness of SIS grouping and exploring its pathological features to better understand the uniqueness of ACC (Acute Coronary Syndrome). Figure 3 We used two mainstream deep learning networks—ResNet50 and Vision Transformer-B16—for validation and comparison (Table 1). Five-fold cross-validation results showed that the AUC for SIS classification reached 0.8 ± 0.01, with ResNet50 performing best (AUC = 0.8214, accuracy = 0.71). Figure 3 Furthermore, ResNet50 also performed well in ACC binary classification C1A / B model prediction (AUC = 0.848, accuracy = 0.74) (S Figure 3 bc), further demonstrating the effectiveness of SIS grouping in pathology. To more intuitively illustrate SIS grouping and its classification probability, we visualized it in the form of a heatmap in the original slides ( Figure 3 g). Using Class Activation Maps (CAMs) technology, we visualized pathological features associated with SIS grouping. The results showed that SIS was associated with sinus invasion and necrosis in the Weiss score, and unexpectedly with lymphocytic infiltration. Figure 3 h, Figure 4 d). To further validate the clinical applicability of the model, we selected the one-fold model with the best ResNet50 performance and validated it on ACC patients from the First and Second Affiliated Hospitals of Dalian Medical University. The results showed that patients in the High SIS group were associated with high lymphocyte infiltration, while patients in the Low SIS group had almost no lymphocyte infiltration; both groups exhibited sinus invasion and necrosis characteristics. Figure 3 i, Figure 4e). This indicates that the model performs consistently on the external validation set and has potential clinical application value. Further prognostic follow-up analysis revealed that patients in the high SIS group had significantly longer survival times than those in the low SIS group. However, due to the limited follow-up time and number of patients, the p-value did not reach statistical significance (p>0.05). When these patients were combined with those in the TCGA cohort for survival analysis, the p-value decreased, which to some extent suggests that these patients conformed to the prognostic characteristics of the SIS subgroup ( Figure 4 fg, Figure 6 a).

[0110] Table 1. Results of two models for classifying SIS and C1A / B subtypes on WSIs

[0111]

[0112] 4. Novel ACC grouping – SIS and clinical relevance

[0113] Several subtypes have been identified in the ACC dataset, including the COC subtype and the C1A / C1B subtype. Figure 5 a). Comparative analysis revealed that patients with high SIS were more associated with the better-prognostic COC1 and indolent C1B subtypes, while patients with low SIS were associated with the poorer-prognostic COC2, COC3, and aggressive C1A subtypes. Significant differences in SIS were found among COC subtypes, C1A / C1B subtypes, expression subtypes, and methylation subtypes. Figure 5 bc). In the TCGA immunophenotyping analysis, ACC patients were mainly distributed in the C3 group (best prognosis) and the C4 group (worst prognosis). The C3 group was mostly composed of patients with high SIS, while the C4 group was mainly composed of patients with low SIS. The SIS level in the C3 group was significantly higher than that in the C4 group. Figure 5 d). Patients in the high SIS group showed good prognosis in both the TCGA and / or GEO datasets. Figure 5 ef, Figure 6 Univariate and multivariate Cox regression analyses confirmed that SIS is an independent prognostic indicator for ACC patients. Figure 5 hi). In clinical parameters, patients in the low SIS group were mainly concentrated in women, T4, N1, M1, Stage III, and Stage IV. Figure 5 j). The Weiss score plays an important role in the pathological diagnosis of ACC. Patients with low SIS account for more than 50% of Weiss score-related indicators, such as necrosis, mitotic rate >5 / 50 HPF, venous invasion, and sinus (lymphatic) invasion. Figure 5g) Epithelial-mesenchymal transition (EMT), tumor mutational burden (TMB), stem cell index based on mRNA expression (mRNAsi), and Ki-67 are common biomarkers in oncology. ACC patients with high EMT, high TMB, high mRNAsi, and high Ki-67 generally have a poorer prognosis, with the low SIS group within this stratification having the worst survival prognosis, and vice versa. Figure 5 kn).

[0114] 5. The low SIS group is closely related to steroid synthesis.

[0115] Because the low SIS group has a large number of patients and a poor prognosis, with over 75% exhibiting abnormal hormone secretion, hormone-suppressing drugs should be the first-line consideration. Mitotan is the only FDA-approved drug for inhibiting corticosteroid synthesis. Although it has been used clinically for over 50 years, its specific mechanism remains unclear. Current evidence suggests that it limits steroid production by inhibiting the activity of steroid-producing enzymes, thereby preventing the conversion of cholesterol into steroids.

[0116] KEGG GSEA and GSVA analyses showed that the high SIS group was enriched in immune-related pathways, while the low SIS group was enriched in steroid biosynthesis pathways. HALLMARK GSEA and GSVA analyses further revealed that the high SIS group was particularly strong in immune pathways, while the low SIS group was enriched in cholesterol homeostasis pathways. Figure 7 ab, Figure 10 (ab). We hypothesize that the high SIS group may be more related to immune response, while the low SIS group may be related to adrenal function. SIS was significantly negatively correlated with adrenal cortex differentiation markers (ADS) (r = -0.62). Figure 7 e). The majority of patients in the high SIS group were predicted to have low ADS (71.4%) and lack secretory function of abnormal hormones (61.5%) and cortisol (80.8%). Conversely, the majority of patients in the low SIS group were predicted to have high ADS (66%) and possess secretory function of abnormal hormones (78.7%) and cortisol (57.4%). Figure 7 d). In the Cortisol Present group, the proportion of patients with low SIS and high ADS was 84%, while in the Hormone Present group, the proportion of patients with low SIS was 79%, higher than the 70% in high ADS. Therefore, we believe that low SIS is more effective in assessing adrenal function than high ADS. Figure 7fg). Through HALLMARK_CHOLESTEROL_HOMEOSTASIS and KEGG_STEROID_BIOSYNTHESIS analysis, we screened out 9 core genes related to steroid hormone synthesis (fg). Figure 7 c). Mitotan is a major drug that inhibits the synthesis of ACC steroid hormones. It primarily inhibits SOAT1, which is associated with cholesterol storage, and CYP11A1 and CYP11B1, which are associated with the conversion of cholesterol to cortisol and aldosterone. Figure 9 a). Among the 12 core genes, DHCR7, CYP51A1, SOAT1, and CYP11A1 were highly associated with SIS and ADS. DHCR7 showed significant and stable differential expression in the hormone-related subgroup, while other genes exhibited greater fluctuations. Figure 7 h). Except for EBP, CYP11A1, and CYP11B1, the frequency of genomic alterations for other genes was higher in the low SIS group than in the high SIS group. In particular, DHCR7, SOAT1, and FDFT1 showed no genomic alterations in the high SIS group, while they had ≥10% genomic alterations in the low SIS group. Figure 9 a). Of these 12 genes, only patients with DHCR7 genomic alterations showed poor prognosis in overall survival (OS), progression-free survival (PFS), and disease-specific survival (DSS). Figure 9 bd), while patients with SOAT1 genomic alterations only showed poor prognosis in PFS. Using Lasso regression, RF, and SVM-RFE machine learning methods, the metabolic genes most associated with SIS characteristics—DHCR7 and SC5D (bd)—were further screened. Figure 7 i). In whole-cancer expression analysis, DHCR7 showed the highest expression in ACC, while SC5D expression was at a moderate level. Nevertheless, high expression of both was associated with poorer patient prognosis. Figure 9 After comprehensive analysis, we believe that DHCR7 is a potential molecular marker for ACC. IHC results showed that the protein expression level of DHCR7 was high in tumor tissues. Figure 8 b). Single-cell RNA sequencing results showed that DHCR7 was specifically expressed in tumor cells ( Figure 8 c). CCK-8 assay results showed that NCI-H295R cells (IC50 = 4.684 μM) were more sensitive to mitotane than SW-13 cells (IC50 = 6.918 μM). Figure 8 e). In DHCR7 knockdown cells, mitotane sensitivity was significantly increased. Figure 8 de).

[0117] 6. The high SIS group is closely related to immunity.

[0118] GSEA and GSVA results showed that patients in the high SIS group had more pronounced tumor immune characteristics. Therefore, we further analyzed the correlation between SIS and ESTIMATE score and immune cell infiltration abundance. The results showed that, in the same dataset, the correlation between the high SIS group and ESTIMATE score was significantly higher than that in the low SIS group, with the correlation exceeding 0.8 in both datasets. Figure 12 a). Regarding the abundance of immune cell infiltration, both patient groups showed a correlation with increased monocytes, decreased neutrophils, decreased basophils, decreased natural killer cells, decreased naïve CD8 T cells, and increased cytotoxic cells. In the high SIS group, the abundance of T CD8 cells and Tfh cells increased with increasing SIS, while in the low SIS group, the abundance of macrophages, macrophage M1, macrophage M2, and activated NK cells increased with lower SIS. Figure 10 b). Through machine learning analysis, pan-cancer immunology studies revealed four immune-related factors: MHC molecules (MHC), immunomodulators (CP), effector cells (ECs such as activated CD8+ T cells and CD4+ T cells, TemCD8+ and TemCD4+ cells), and suppressor cells (SCs such as Tregs and MDSCs). ACC analysis showed that patients in the high SIS group had higher levels of MHC and EC, while CP and SC levels were lower. Figure 10 c). Meanwhile, patients in the high SIS group had significantly higher immune phenotypes (IPS). Figure 10 d). Tumor immunophenotyping (TIP) showed that the high SIS group had higher activity scores in CD8 T cells, macrophages, NK cells, and immune cells infiltrating the tumor. Figure 10 e), and the overall activity score showed a strong positive correlation with SIS (r = 0.69). Figure 10 f). Analysis of five immune expression characteristics showed that high SIS scores were higher and statistically significant ( Figure 11 a). Furthermore, previous studies have shown that immune infiltration is associated with DNA damage. Our analysis showed that patients in the high SIS group had lower SNV neoantigens, non-silent mutation rates, copy number variation (CNV) burden (fragment number), and homologous recombination defects, in addition to exhibiting low proliferation rates (a). Figure 12 f). Quantitative immune trait loci (immunQTLs) are used to assess the impact of common genetic variations on immune infiltration. Survival-associated immune QTL (FDR < 0.05)-GWAS analysis showed that QTLs regulated by T cells (Tregs) were the most numerous, followed by CD8 T cells, and the fewest were resting CD4 memory T cells (Tregs). Figure 12 g). Among these GWAS diseases, calcium-related QTLs are the most numerous, all originating from Tregs. Schizophrenia is the second most prevalent, involving T cell regulation (Tregs), CD8 T cells, macrophage M1, and resting CD4 memory T cells, representing the broadest range of involvement.

[0119] TIDE is an important indicator used to predict tumor response to immune checkpoint blockade (ICB). Both the TCGA and GEO datasets show a significant negative correlation between SIS and TIDE. Figure 12 In the high SIS group, over 70% of patients responded to ICBs, while over 65% of patients in the low SIS group were resistant to ICBs. Furthermore, among patients resistant to ICBs, the low SIS group accounted for over 80%. Figure 11 bc). The two main immune escape mechanisms involved in TIDE are associated with T cell dysfunction and T cell rejection, respectively. Our data show that the immune escape mechanism of ACC is mainly associated with T cell dysfunction. Cancer-associated fibroblasts (CAFs), myeloid-derived suppressor cells (MDSCs), and tumor-associated macrophages (TAMs) M2 subtype are the three main cell types that suppress T cell infiltration. Patients in the high SIS group showed a significant negative correlation with TAM M2. In addition, the immune characteristics of ACC were also strongly correlated with interferon-γ response and T cell inflammatory phenotype (Merck18). Figure 12 e). We further investigated the expression of three classes of molecules involved in tumor escape mechanisms in the low SIS group. Most MHC molecules were expressed at low levels in the low SIS group, thus avoiding T cell recognition; immunosuppressive factors (such as TGFBR1) may be upregulated to achieve tumor escape; and immunostimulatory factors (such as RAET1E) may be downregulated to avoid immune attack. Figure 11 d).

[0120] To further understand the relationship between tumors and the tumor microenvironment, we selected a low SIS patient (with abnormal hormone secretion) for single-cell RNA sequencing analysis. The results showed that the patient's cell subsets were mainly concentrated in cancer cells (78.95%), macrophages (15.44%), progenitor cells (3.42%), mesenchymal cells (1.17%), T cells (0.65%), and endothelial cells (0.38%). Figure 11 e). We screened 79 exosome-related genes specifically expressed in tumor cells (avg_log2FC>0.4). KEGG analysis showed that these genes were enriched not only in hormone synthesis and secretion pathways, but also in metabolic reprogramming and energy metabolism pathways, immune evasion and tumor microenvironment pathways, and drug metabolism and resistance pathways. Figure 11f). Analysis using HitPredict and CellphoneDB revealed interactions between these exosomes and receptors on immune cells, with over 65% of these interactions involving T cells and macrophages. Most of these exosomes were highly expressed in the low SIS group in the TCGA and GEO datasets. Figure 11 g). These interacting genes are significantly enriched in tumors, extracellular matrix, and immunity. Figure 11 h). Unexpectedly, some genes (such as AXL, HGF, PGFFB, and PDGFRB) were found to be associated with resistance to EGFR tyrosine kinase inhibitors.

[0121] 7. Drug sensitivity prediction based on SIS grouping

[0122] We used the GDSC1 and GDSC2 datasets to predict drug sensitivity in SIS groups. The results showed that drugs acting on the PI3K / Akt / mTOR pathway (90%) were more sensitive in patients with high SIS. Of all drugs more sensitive in the high SIS group, 54.5% acted on the Autophagy pathway, 40.9% on the PI3K / Akt / mTOR pathway, and 18.2% on the Protein Tyrosine Kinase / RTK pathway. Furthermore, GDSC1 and GDSC2 jointly predicted that BI-2536 (a PLK1 inhibitor) was more sensitive in patients with low SIS in the TCGA and GEO datasets. Figure 13 a). SPIED3 predictions also showed that MTOR inhibitors were more sensitive in patients with high SIS, consistent with GDSC results. Figure 13 b). cMAP analysis primarily predicted drugs that were more effective in patients with low SIS, including calmodulin antagonists, dopamine receptor antagonists, squalene cyclase inhibitors, serotonin receptor antagonists, and sterol demethylase inhibitors. These drugs all inhibit the synthesis of steroid hormones, consistent with our analysis that patients with low SIS exhibit adrenal function characteristics. Figure 13 c). Meanwhile, SPIED3 and cMAP jointly predicted the efficacy of four classes of drugs in patients with low SIS: acetylcholine receptor antagonists (mebivelin), dopamine receptor antagonists, BCR-ABL kinase inhibitors (imatinib), and norepinephrine reuptake inhibitors (maprotiline). It is noteworthy that all dopamine receptor antagonists are used to treat schizophrenia ( Figure 13 d). Furthermore, single-cell RNA sequencing results showed that the screened tumor cell-specific genes were significantly enriched in calcium-binding and iron-binding pathways in GO-MF enrichment analysis; these pathways are both associated with schizophrenia. Figure 13e). Among them, the CHP1 gene is involved in both calcium metabolism and promotes ferroptosis. The PCLO gene is not only related to calcium metabolism but is also significantly associated with schizophrenia. Figure 13 f).

[0123] Regarding drug treatment, our study predicts that patients in the high SIS group will be more sensitive to drugs affecting the PI3K / Akt / mTOR pathway, while patients in the low SIS group will be more sensitive to other drugs. In addition to PLK1 inhibitors predicted by GDSC1 and GDSC2, cMAP data show that several drugs that inhibit steroid hormone synthesis may be effective in the low SIS group, including clozapine. As a drug for treating schizophrenia, clozapine has been shown to inhibit aldosterone secretion by inhibiting the D4 receptor. Furthermore, our study found that schizophrenia is associated with Survival-ImmunQTLs of various immune cells in the ACC. Single-cell RNA sequencing results showed that ACC-specific genes are enriched in iron and calcium metabolism pathways, which are highly associated with schizophrenia. Therefore, we speculate that drugs for treating schizophrenia may have the potential for combination therapy with mitotane. On the other hand, drugs that inhibit calcium metabolism may be effective for patients within the SIS group. Notably, Ca... 2+ The levels of these drugs were most closely associated with Survival-Immun QTLs in ACC immune cells. Therefore, drugs that inhibit calcium metabolism and promote ferroptosis may have important research value in the tumor immunity and tumor progression of ACC.

[0124] The present invention has been described in detail above. Those skilled in the art will recognize that the invention can be practiced in a wide range of ways with equivalent parameters, concentrations, and conditions without departing from its spirit and scope, and without requiring unnecessary experiments. Although embodiments have been provided, it should be understood that further modifications can be made to the invention. In summary, according to the principles of the invention, this application is intended to include any changes, uses, or improvements to the invention, including changes made using conventional techniques known in the art that depart from the scope disclosed herein.

Claims

1. The application of a reagent for detecting the expression level of biomarkers in a sample in the preparation of prognostic products for adrenocortical carcinoma, characterized in that, The biomarkers are a combination of CYP17A1, ABAT, CD52, CORO1A, CXCL12, HCLS1, PLEK, C1QB, SIGLEC1, MS4A4A, F13A1, CD163, AIF1, C1QA, CXCL9, FORR2, CXCL10, and FCERT1G.

2. The application according to claim 1, characterized in that, The reagents include reagents for detecting biomarker mRNA levels and / or reagents for detecting biomarker protein levels; Preferably, the reagents for detecting the mRNA level of the biomarker include reagents used in any of the following methods: PCR-based detection methods, Southern hybridization methods, Northern hybridization methods, dot hybridization methods, fluorescence in situ hybridization methods, DNA microarray methods, ASO methods, and high-throughput sequencing platform methods; Preferably, the reagents for detecting the levels of biomarker proteins include reagents used in any of the following methods: hematoxylin-eosin staining, safranin O-fast green staining, Western blotting, enzyme-linked immunosorbent assay (ELISA), radioimmunoassay, sandwich assay, immunohistochemical staining, mass spectrometry, immunoprecipitation analysis, complement fixation analysis, flow cytometry fluorescence sorting, and protein chip analysis.

3. The application according to claim 1, characterized in that, The test is performed on a sample from the subject; Preferably, the sample includes tumor tissue, tumor tissue extract, and tumor cell culture. Preferably, the product further includes sample pretreatment reagents.

4. The application according to claim 1, characterized in that, The product assesses the prognosis of subjects by detecting the expression levels of biomarkers in the obtained samples. Low expression levels of the biomarkers indicate low lymphocyte infiltration in adrenocortical carcinoma, and the subject has a poor prognosis; high expression levels of the biomarkers indicate high lymphocyte infiltration in adrenocortical carcinoma, and the subject has a better prognosis. Preferably, the product includes a reagent kit, a chip, a test strip, or a device, equipment, or computer-readable storage medium.

5. A method for constructing a prognostic model for adrenocortical carcinoma, characterized in that, The method includes the following steps: obtaining biomarker expression level data and subject clinical characteristics in the sample as described in claim 1, and constructing a prognostic model based on the biomarker expression level data and clinical characteristics using an algorithm; The biomarker expression level data can be mRNA expression level or protein expression level; The clinical characteristics include individuals with a better prognosis and individuals with a poorer prognosis for adrenocortical carcinoma.

6. The method according to claim 5, characterized in that, The clinical features also include pathological image features; Preferably, the algorithm includes one or more of the following: principal component analysis, deep neural network, generalized linear model, logistic regression analysis, LASSO regression analysis, nearest neighbor analysis, support vector machine, neural network model, and random forest model; Preferably, the algorithm is principal component analysis.

7. The method according to claim 5, characterized in that, The method includes the following steps: obtaining the expression level data of the biomarker as described in claim 1 in the sample, performing unsupervised clustering on the sample, extracting the first principal component using principal component analysis as the final scoring basis, and calculating the SIS score by aggregating the principal component scores representing SIS gene feature A and the principal component scores representing SIS gene feature B. Preferably, the method for judging the prognostic model is as follows: based on the constructed prognostic model, an SIS score is obtained; if the SIS score is higher than the cutoff value, a classification result indicating a better prognosis for the subject is output; if the SIS score is lower than the cutoff value, a classification result indicating a worse prognosis for the subject is output.

8. A method for predicting the sensitivity of subjects to treatment drugs for adrenocortical carcinoma based on SIS scores, characterized in that, The method includes the following steps: obtaining the subject's SIS score, the SIS score being obtained by a model constructed using the method described in any one of claims 5-7; comparing the SIS score with a cutoff value to determine whether the subject belongs to the high SIS group or the low SIS group; If a subject belongs to the high SIS group, it is predicted that the subject will be sensitive to mTOR inhibitors and PI3K / Akt / mTOR pathway regulators; If the subject belongs to the low SIS group, it is predicted that the subject will be sensitive to PLK1 inhibitors, calmodulin antagonists, dopamine receptor antagonists, squalene cyclase inhibitors, serotonin receptor antagonists, sterol demethylase inhibitors, acetylcholine receptor antagonists, BCR-ABL kinase inhibitors, and norepinephrine reuptake inhibitors. Preferably, the PLK1 inhibitor is BI-2536 or a pharmaceutically acceptable salt thereof; Preferably, the acetylcholine receptor antagonist is mebiveline or a pharmaceutically acceptable salt thereof; Preferably, the BCR-ABL kinase inhibitor is imatinib or a pharmaceutically acceptable salt thereof; Preferably, the norepinephrine reuptake inhibitor is maprotiline or a pharmaceutically acceptable salt thereof.

9. A computer system for predicting the prognosis of adrenocortical carcinoma, characterized in that, The computer system includes an analysis unit, which uses the adrenocortical carcinoma prognostic model constructed by the method described in any one of claims 5-7 to perform analysis, obtain an SIS score, compare the SIS score with the cutoff value, and find that a higher SIS score indicates a better prognosis and a lower SIS score indicates a worse prognosis. Preferably, the computer system further includes an input unit and an output unit; Preferably, the input unit is used to input the biomarker expression level data of the sample as described in claim 1, wherein the biomarker expression level may be mRNA expression level or protein expression level; Preferably, the output unit is used to output the classification result of the subject corresponding to the sample, indicating whether the prognosis is good or bad.

10. A computer device for predicting the prognosis of adrenocortical carcinoma, characterized in that, The computer device includes a memory and a processor; The memory is used to store program instructions; The processor is used to execute program instructions, which, when executed, are used to perform the following operations: obtain the expression level data of the biomarker as described in claim 1 in the sample, input the data into the adrenocortical carcinoma prognostic model constructed by the method described in any one of claims 5-7 for analysis, obtain the SIS score, compare the SIS score with the cutoff value to obtain the classification result of the sample; those with higher SIS scores have better prognoses, and those with lower SIS scores have worse prognoses.

Images