Integrating Single-Cell Sequencing and Machine Learning to Uncover the Role of Mitophagy in Subtyping and Prognosis of Esophageal Cancer

doi:10.21203/rs.3.rs-4917245/v1

Download PDF

Research Article

Integrating Single-Cell Sequencing and Machine Learning to Uncover the Role of Mitophagy in Subtyping and Prognosis of Esophageal Cancer

https://doi.org/10.21203/rs.3.rs-4917245/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

Globally, esophageal cancer stands as a prominent contributor to cancer-related fatalities, distinguished by its grim prognosis. Mitophagy has a significant impact on the process of cancer progression. This study investigates the prognostic significance of mitophagy-related genes (MRGs) in esophageal carcinoma (ESCA) with the aim of elucidating molecular subtypes. By analyzing RNA-seq data from The Cancer Genome Atlas (TCGA), 6451 differentially expressed genes (DEGs) were identified. Cox regression analysis narrowed this list to 14 MRGs with notable prognostic implications. ESCA patients were classified into two distinct subtypes (C1 and C2) based on these genes. Furthermore, leveraging the differentially expressed genes between Cluster 1 and Cluster 2, ESCA patients were classified into two novel subtypes (CA and CB). Notably, patients in C2 and CA subtypes exhibited inferior prognosis compared to those in C1 and CB (p < 0.05). Functional enrichments and immune microenvironments varied significantly among these subtypes, with C1 and CB demonstrating higher immune checkpoint expression levels. Employing machine learning algorithms like LASSO regression and Random Forest, alongside multivariate COX regression analysis, two core genes: HSPD1 and MAP1LC3B were identified. A robust prognostic model based on these genes was developed and validated in two external cohorts. Additionally, single-cell sequencing analysis provided novel insights into esophageal cancer microenvironment heterogeneity. Through Coremine database screening, Icaritin emerged as a potential therapeutic candidate to improve esophageal cancer prognosis. Molecular docking results indicated favorable binding efficacies of Icaritin with HSPD1 and MAP1LC3B, enhancing the comprehension of the underlying molecular mechanisms of esophageal cancer and offering therapeutic avenues.

esophageal cancer

mitophagy

single-cell sequencing

machine learning

prognosis

subtype

Esophageal carcinoma, a malignant tumor, predominantly comprises esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC), originating from the epithelial cells of the esophageal lining^[1]. This disease imposes a considerable global health challenge, ranking as the sixth foremost contributor to cancer-related mortality and the ninth in the incidence of cancers globally, underscoring its widespread and severe nature^[2]. Despite some therapeutic advancements in endoscopic interventions, surgical resections, chemotherapy, radiotherapy, and multimodal treatment regimens, clinical outcomes for ESCA patients remain discouraging, with mortality rates persistently high^[3]. Therefore, there is an urgent requirement for identifying prognostic biomarkers and construct robust prognostic models to enhance patient status, implement personalized treatment strategies, and ultimately improve the survival rates of those suffering from this serious disease. The infiltration of immune cells has a well-established correlation with cancer prognosis, particularly in hard-to-treat tumors like esophageal cancer^[4]. Statistical evidence suggests that immunotherapy has significantly improved treatment outcomes in these refractory cancers^[5]. However, individual responses to cancer immunotherapy vary greatly^[6]. Consequently, identifying novel and distinctive genes holds promise for advancing more targeted and effective treatment strategies, thereby offering valuable prospects for ESCA.

Under stressors such as nutrient deprivation, reactive oxygen species (ROS), and cellular senescence, mitochondria can experience varying degrees of damage^[7]. Mitophagy, a targeted autophagy mechanism that specifically eliminates damaged mitochondria within autophagosomes for fusion with lysosomes and subsequent disposal, is essential for preserving both mitochondrial integrity and cellular stability^[8]. This concept was first formally proposed by Lemasters in 2005, who observed that the onset of mitophagy is triggered by the depletion of mitochondrial membrane potential and the subsequent opening of the mitochondrial permeability transition pore ^[9]. Several diseases exhibit a close correlation with aberrant mitophagy, implicating its involvement in their underlying pathogenesis, including neurodegenerative disorders, hematological malignancies, cardiovascular diseases, and even cancer^[10–11]. Notably, mitophagy exhibits a dual and complex part in the development of tumors, both promoting and inhibiting cancer progression. Distinct cancer types, such as ovarian, lung, colorectal, and breast cancers, display heterogeneous levels of mitophagy, indicating its diverse involvement across various malignancies. Among the four main mitophagy pathways in mammals, the PINK1/Parkin-mediated mitophagy is considered a potential tumor suppression mechanism. Research has shown that 25% of colorectal cancer tumors exhibit focal PARK2 loss^[12], and specific PARK2 mutations or deletions have been detected across multiple cancer types, suggesting that PARK2 mutations may drive tumor progression by eliminating the E3 ubiquitin ligase activity of Parkin. However, the role of PARK2 loss in cancer may depend on the specific context and could even be contradictory. For instance, in melanoma, Parkin loss inhibits Mfn2 ubiquitination, inducing apoptosis and suppressing melanoma development and metastasis. BNIP3, a protein with pro-apoptotic properties, augments mitophagy by impeding the fusion of dysfunctional mitochondria, thereby expediting their elimination from the cell. The function of BNIP3 as a tumor inhibitor or promoter remains controversial in various cancer contexts. Similarly, the role of Nix-mediated mitophagy may exhibit complexity and diversity across different cancer types and stages^[13]. The dual nature of mitophagy emphasizes its potential as a novel therapeutic target. Inducing mitophagy may provide an alternative mechanism for cancer cell death, creating hopeful avenues for cancer treatment. Inhibitory strategies targeting mitophagy regulators such as PINK1, Parkin, BNIP3, and FUNDC1 may offer new therapeutic options for cancer intervention^[14]. Current research has primarily concentrated on the role of mitophagy in ovarian, lung, as well as breast cancers, while studies on esophageal cancer (ESCA) remain relatively limited. Therefore, there is an urgent need to explore the relationship between mitophagy and esophageal cancer to uncover new therapeutic strategies for this malignancy.

This study aims to identify unique genes associated with mitophagy by comparing ESCA patients with a control group. We then employ univariate and multivariable Cox regression analyses, along with machine learning algorithms, to determine two feature genes with prognostic value. These findings provide a basis for establishing a prognostic scale for predicting outcomes in esophageal cancer patients. Furthermore, we validate the reliability of the model through ROC curves, DCA curves, and calibration curves. Additionally, we utilize consensus clustering methods to identify two distinct subtypes of esophageal cancer patients and explore their potential characteristics through enrichment analysis, immune infiltration analysis and immune checkpoints comparison. Moreover, using single-cell sequencing technology, we investigate the expression patterns of these two feature genes in esophageal cancer cells. Eventually, we conducted traditional Chinese medicine prediction and molecular docking for the core genes. This comprehensive approach not only enhances our comprehension of the molecular mechanisms driving esophageal cancer but also provides insightful perspectives for tailoring individualized therapeutic approaches to this condition.

2.1 Data Collection and Preprocessing

The flowchart of this study is visually depicted in Fig. 1,outlining its methodology and progression.

RNA sequencing (RNA-seq) data for 173 individuals diagnosed with esophageal carcinoma (ESCA) was obtained from The Cancer Genome Atlas (TCGA) public database (https://portal.gdc.cancer.gov). In addition, a collection of mitophagy-related genes (MRGs) was extracted from the GeneCards database (https://www.genecards.org/), yielding a total of 385 genes for further analysis. For external validation, datasets GSE26886 and GSE20347 were sourced from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/). The characteristics of these validation datasets are summarized in Table 1.

Table 1

External validation data set from a GEO information (GSE26886 and GSE20347)
Datatsets	Platform	Total sample number	Normal sample number	ESCA sample number
GSE26886	GPL570	69	20	49
GSE20347	GPL571	34	17	17

2.2 Identification of Differentially Expressed Genes Related to Mitophagy

The "limma" package^[15] were utilized for differential expression analysis. The parameters for differential expression analysis were set as follows: |log2FC| > 1.2, and FDR < 0.05. The genes identified through this screening were cross-referenced with mitophagy-related genes (MRGs) sourced from the GeneCards database to determine the common genes, designated as differentially expressed mitophagy-related genes (DEMRGs).

2.3 Cox regression analysis

Univariate and multivariate Cox proportional hazards regression analyses were employed to screen critical genes closely associated with ESCA prognosis. Additionally, we applied forest plots to visually represent the results.

2.4 Machine Learning

During the construction of the prognostic model, two advanced machine learning techniques were employed: LASSO^[16] regression and Random Forest^[17]. Lasso regression, an extension of logistic regression, incorporates a shrinkage parameter to minimize classification error while selecting significant variables, thereby providing an effective approach to filter through numerous feature variables for optimal model performance. Random forest, as an ensemble of decision trees, excel in survival analysis by adeptly identifying important features and elucidating patient outcomes, illustrating their extensive applicability in this domain.

2.5 Development and verification of a risk assessment model

To develop a risk model, we employed LASSO regression to calculate the coefficients of each gene, as shown in the formula \(\:\:\text{R}\text{i}\text{s}\text{k}\:\text{s}\text{c}\text{o}\text{r}\text{e}=\varSigma\:\left(gene1\:expression\times\:coefficient1+gene2\:expression\times\:coefficient2\right)\),where represents the expression of MRDEGs. Patients were segregated into high- and low-risk cohorts, utilizing the median risk score as the demarcation point. To validate this prognostic model, external datasets, GSE26886 and GSE20347, sourced from the GEO database, were analyzed. To further assess the predictive precision of the risk model, decision curve analysis (DCA^)[18]and time-dependent receiver operating characteristic (ROC) curve ^[19]evaluations were performed using the "ggDCA" and "timeROC" R packages. Kaplan-Meier curves were generated and log-rank tests administered to evaluate the distinction in overall survival (OS) between the two risk groups, subsequently assessing the clinical outcome predictive potential of the risk model in ESCA. Utilizing the "rms" and "survival" R packages, a comprehensive nomogram was developed that integrated risk score and clinicopathological characteristics to facilitate personalized predictions. A calibration curve was generated to assess the predictive accuracy of the nomogram.

2.6 Consensus Clustering

Using the "ConsensusClusterPlus" R package^[20] unsupervised clustering of DEMRGs was conducted to identify distinct subtypes of ESCA patients. To visualize the distribution of patients across these clusters, Uniform Manifold Approximation and Projection (UMAP) analysis was performed utilizing the "umap" R package. Subsequently, principal component analysis (PCA) was applied to the subtypes derived from the secondary clustering analysis, expounding upon their inherent structure. Both UMAP and PCA serve as effective dimensionality reduction techniques, enabling the mapping of high-dimensional data into lower-dimensional space, thereby making it convenient for a clearer representation of the data's characteristics and insights into ESCA subtypes.

2.7 Immune Infiltration

To investigate the distribution of immune cells within ESCA patient subtypes and the association between risk score and tumor-infiltrating immune cells, three algorithms were employed: TIMER, CIBERSORT and ESTIMATE^[21]. These algorithms aided in estimating the relevance between the prognostic model and immune cell infiltration patterns within the tumor microenvironment (TME). Data on immune infiltration was sourced from TIMER^[22](https://cistrome.shinyapps.io/timer/), with results from TIMER, CIBERSORT compared. Moreover, the ESTIMATE algorithm evaluated the immune and stromal components of the TME, generating stromal scores, immune scores, and ESTIMATED scores^[23]. An increase in these scores indicates a greater presence of corresponding components within the TME, providing critical insights into the tumor's immunological landscape and potential implications for patient prognosis.

2.8 Enrichment Analysis

The "ClusterProfiler" R package was employed to conduct Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses to elucidate the biological processes (BP), molecular functions (MF),cellular components (CC), and potential signaling pathways associated with ESCA subtypes which was visualized in Metascape. To capture subtle alterations in pathway activity, Gene Set Variation Analysis (GSVA) was conducted, transforming gene expression profiles across samples into gene set enrichment scores, thus facilitating the assessment of gene set activity^[24]. Complementarily, Gene Set Enrichment Analysis (GSEA) was conducted to evaluate the differential expression of predefined gene sets between distinct biological states, providing a comprehensive overview of the transcriptional landscape in ESCA^[25].

2.9 Mitophagy Expression in Single-Cell RNA Sequencing Tumor Data

TISCH2 (http://tisch.compgenomics.org), a robust single-cell RNA-seq TME database, enables an exhaustive analysis of heterogeneity across datasets and cell types^[26]. This study explores the immune-related prognostic features of ESCA at the single-cell level, utilizing data from TISCH2, specifically GSE160269 and GSE173950.

2.10 Prediction of Traditional Chinese medicine and molecular docking

Map the mitophagy key genes obtained in the aforesaid steps onto the Coremine medical ontology information retrieval platform (www.coremine.com/medical/), and select the top 30 traditional Chinese medicine drugs that may have a regulatory effect on the mitophagy key genes in patients suffering from esophageal cancer. Furthermore, to procure the target protein result files, searches were conducted on the PDB database (https://www.rcsb.org/) and the AlphaFold database (https://alphafold.ebi.ac.uk/). From the PubChem database (https://pubchem.ncbi.nlm.nih.gov/), the active ingredient structure files were retrieved and converted into PDB format utilizing Open Babel 2.3.2 software. PYMOL 2.3.4 software was then employed to meticulously prepare the receptor protein by eliminating water molecules and ligands. Following this, the receptor protein underwent essential modifications, including hydrogenation and charge balancing, using AutoDockTools software. Subsequently, molecular docking simulations were conducted between the receptor protein and small molecule ligands utilizing AutoDock Vina 1.1.2^[27]. Finally, Pymol was utilized to visualize the docking outcomes, focusing on those with favorable binding energies.

3.1 Identification of Prognostic Mitophagy-Related Genes (MRGs)

Differential expression analysis of tumor and normal samples sourced from the TCGA database revealed significant alterations in gene expression profiles in esophageal squamous cell carcinoma (ESCA). Specifically, a total of 6451 differentially expressed genes (DEGs) were identified, with 1455 genes downregulated and 4996 genes upregulated (Fig. 2A).

Furthermore, we identified 136 genes common to both the mitophagy-related genes and differentially expressed genes (DEGs) (Fig. 2B). Subsequently, a profound KEGG enrichment analysis of these genes prominently converged on mitophagy and autophagy pathways(Fig. 2C).To evaluate the prognostic value of these genes, we conducted a univariate Cox regression analysis incorporating survival status, survival duration, and gene expression profiles, ultimately identifying 14 genes with significant prognostic value, the results were visually presented in a forest plot(Fig. 2D).Moreover, a correlation heatmap was also generated to illustrate the intricate relationships among these 14 genes(Fig. 2E).

3.2 Identification of MRG-related Subtypes Based on 14 MRGs

Consensus clustering of 153 ESCA patients from the TCGA cohort was conducted utilizing the expression profiles of the 14 genes identified via univariate Cox regression analysis. Adjusting the clustering parameter (k) within a range from 2 to 10, we identified the optimal clustering configuration at k equals 2 (Fig. 3A-C), effectively stratifying patients into two distinct clusters: Cluster 1 (n = 71) and Cluster 2 (n = 82). UMAP visualizations corroborated the clear demarcation between these clusters (Fig. 3D). The Kaplan-Meier survival analysis revealed a statistically significant enhancement in the overall survival rate among patients belonging to Cluster 1, in contrast to those in Cluster 2 (p < 0.05). (Fig. 3E), underscoring the capability of MRGs to stratify ESCA patients into subtypes with notable survival disparities.

3.3 Immunological Landscape of the Two Subtypes

Our comprehensive immunological analyses elucidated substantial disparities in immune landscape between the two molecular subtypes. Utilizing the Estimate algorithm, we discerned notably heightened stromal, immune, and overall estimate scores among ESCA patients in Cluster 1(Fig. 4A–C), indicating a more favorable prognosis. The CIBERSORT algorithm further identified an increased proportion of T cells CD4 + naive (p < 0.05) and activated natural killer (NK) cells (p < 0.05) in Cluster 1; however, no significant differences were detected in other immune cell infiltrations (Fig. 4D). Additionally, TIMER algorithm analysis confirmed higher infiltration levels of neutrophils and myeloid dendritic cells in Cluster 1, while no significant variation in CD8 + T cell infiltration was noted (Fig. 4E-G), accentuating the distinct immunological profiles of these two subtypes.

3.4 Differentially Expressed Genes (DEGs) and their Functional Annotations

Distinct sets of differentially expressed genes (DEGs) were discerned between the two clusters., uncovering a total of 164 DEGs. Among these, 70 genes exhibited reduced expression in Cluster 1 compared to Cluster 2, while 94 genes showed increased expression in Cluster 1 (Fig. 5A). Subsequent KEGG enrichment analysis elucidated several prominent signaling pathways, including the p53 signaling pathway, ECM-receptor interaction, and protein digestion and absorption (Fig. 5B). Correspondingly, GO enrichment analysis indicated that DEGs were significantly enriched in biological processes such as the cell cycle, extracellular matrix organization, and the positive regulation of cell migration (Fig. 5C). Furthermore, enrichments of crucial molecular functions were observed (Fig. 5D). Protein-protein interaction (PPI) analysis unveiled three sub-modules, all tightly linked to tumor development and immunity, suggesting a potential interplay between immune response and the role of mitophagy in esophageal cancer (Fig. 5E). In order to gain deeper insights into the relationships between enriched pathways and the prognostic outcomes of ESCA patients, Gene Set Enrichment Analysis (GSEA) was conducted, demonstrating that focal adhesion and ECM receptor interaction were remarkably expressed in Cluster 1, whereas lysine degradation and homologous recombination were notably enriched in Cluster 2 (Fig. 5F).

3.5 Further Identification of ESCA Based on DEGs Between Subtypes

As noted previously, differential analysis between clusters 1 and 2 led to the identification of 164 genes exhibiting differential expression patterns (Fig. 6A). Based on these genetic markers, ESCA patients were categorized into distinct subgroups. The optimal clustering was achieved at k = 2 (Fig. 6B-D). Principal Component Analysis (PCA) revealed that these genes effectively differentiated the two subtypes—Cluster A (CA) and Cluster B (CB) (Fig. 6E).

3.6 Discerning biological functions and immune microenvironments across distinct gene clusters.

Our investigation into the expression of 14 genes with notable prognostic significance across the two subtypes revealed distinctive expression profiles. (Fig. 7A).The results of Gene Set Variation Analysis (GSVA) highlighted the predominant involvement of CA in NADPH oxidase H2O2 forming activity. Conversely, CB associates mainly with negative regulation of the vitamin D biosynthetic process and pathways, including the TGF-Beta signaling pathway and WNT signaling pathway (Fig. 7B, C). The CIBERSORT algorithm disclosed higher infiltration levels of CD4 + naive T cells, follicular helper T cells, and activated myeloid dendritic cells in Cluster B, alongside lower infiltration levels of CD4 + memory resting T cells, CD4 + memory activated T cells, and resting NK cells (Fig. 7D). Additionally, the TIMER algorithm indicated higher infiltration levels of CD4 + T cells in Cluster A, along with lower levels of B cells, Neutrophils and Myeloid dendritic cells (Fig. 7E).

3.7 Construction of Risk Model

We initially applied LASSO regression analysis, utilizing 10-fold cross-validation to identify the optimal model. The optimal λ value was determined, leading to the identification of 7 genes (Fig. 8A-B). Furthermore, the random forest algorithm identified six genes with importance scores exceeding 1.4 as central features (Fig. 8C). Notably, three genes—TMEM43, HSPD1, and MAP1LC3B—were common to both machine learning approaches (Fig. 8D). Following this, multivariate Cox regression analysis revealed that HSPD1 (HR=1.58, 95% CI: 1.05-2.38) and MAP1LC3B (HR=2.86, 95% CI: 1.45-5.65) emerged as pivotal differentially expressed mitophagy-related genes (DEMRGs) (Fig. 8E). Based on these two critical DEMRGs, a risk model was constructed, with the risk score calculated as follows: Risk score=(0.366419869×HSPD1 expression)+(0.397483984×MAP1LC3B expression). Patients in the TCGA cohort were subsequently stratified into high-risk (n=77) and low-risk (n=76) groups based on the median risk score. Moreover, the coefficient value of the model was determined (Fig. 8F). The relationship between risk score and survival status was further investigated, with scatter plots illustrating that patient mortality increased alongside rising risk score (Fig. 8G). To emphasize the prognostic accuracy of the model, a time-dependent ROC analysis was conducted, yielding AUC values of 0.74, 0.71, and 0.83 for 1, 3, and 5 years, respectively (Fig. 8H), indicating the model's robust predictive capability. Kaplan-Meier analysis demonstrated that the high-risk group experienced a poorer prognosis compared to the low-risk group in the TCGA cohort (Fig. 8I).

3.8 Independence of the Established Risk Model

We further delved into the intricate relationship between the risk score and various clinical characteristics through rigorous subgroup and regression analyses, assessing the independence of our developed risk model. Our findings indicated no significant differences in risk score across different age cohorts (Fig. 9A), genders (Fig. 9B), M stages (Fig. 9C), T stages (Fig. 9D), N stages (Fig. 9E), or overall staging (Fig. 9F), suggesting a lack of correlation between the risk score and these clinical factors. Notably, even when stratifying patients based on age (Fig. 9G-H) and Gender (Fig. 9I-J), our risk model maintained its robust predictive power, with individuals exhibiting lower risk score showing a more favorable prognosis. These outcomes highlight the exceptional utility of our risk model in prognosticating the outcomes of esophageal cancer patients.

3.9 Construction and Calibration of an Integrated Nomogram

To enhance predictive performance, we constructed a nomogram model for evaluate the prognosis of esophageal squamous cell carcinoma (ESCA) (Fig. 10A), which allows for individualized predictions of survival probabilities at 1-year, 3-year, and 5-year intervals. Each risk factor was assigned a score, and the cumulative scores of these indicators were utilized to calculate the total score for predicting ESCA prognosis for each patient. The calibration plot revealed a high degree of concordance between observed and ideal values (Fig. 10B). Examination of the calibration curve provides a comprehensive evaluation of the model's efficacy, substantiating its applicability for clinical application. According to the model, the c-index value was determined to be 0.68 (Fig. 10C). Decision Curve Analysis (DCA) revealed that the nomogram surpasses in delivering superior clinical net benefit (Fig. 10D). These discoveries imply that the risk score-based nomogram can function as a potent prognostic instrument for prediction in clinical scenarios. Furthermore, to authenticate the accuracy of the model, validation was conducted utilizing the GSE26886 and GSE20347 datasets (Fig. 10E-F), with all AUC values exceeding 0.7.

3.10 Comparison of Functional Enrichment and Immune Infiltration distinction Between High and Low-Risk Cohort

To investigate the underlying mechanisms of the risk model, Gene Set Enrichment Analysis (GSEA) was employed to examine the functional enrichment profiles within high-risk and low-risk cohorts. The high-risk group exhibited significant enrichment in spliceosome activity, Alzheimer disease pathways, and the pentose phosphate pathway, while the VEGF, Notch, and ERBB signaling pathways were more prominent in the low-risk group (Fig. 11A). Additionally, we conducted GSEA to elucidate the roles of the two hub genes, HSPD1 and MAP1LC3B, in esophageal carcinoma. In esophageal cancer, GSEA revealed diverse roles for HSPD1 (Fig. 11B), including enrichment in critical processes such as "apoptosis," "DNA replication," and "Adherens junction formation," while also being involved in essential signaling pathways, including the p53 signaling pathway and focal adhesion pathways. For MAP1LC3B (Fig. 11C), it was implicated in the immune process of "natural killer cell mediated cytotoxicity" and displayed enrichment in the "mismatch repair" process, contributing to critical signaling mechanisms, encompassing focal adhesion and the ERBB signaling pathway. Additionally, the differences in genetic mutation profiles between the high-risk and low-risk groups were unveiled through the illustration of the mutation waterfall plot (Fig. 11D). CIBERSORT analysis indicated substantial disparities in immune cell infiltration patterns across high-risk and low-risk groups (Fig. 11E). Correlation analysis demonstrated a significant positive correlation between Mast cell activated and Macrophage in the high-risk group, while a negative correlation was observed between resting myeloid dendritic cells and Monocyte (p < 0.05, Fig. 11F). Conversely, in the low-risk group, Macrophage M0, M2 and Neutrophil exhibited positive correlations, while Neutrophil displayed a negative correlation with Monocyte (p < 0.05, Fig. 11G).

3.11 Exploring the Correlation Between Three Types of Grouping Patterns

We employed the risk score obtained from our risk model to construct box plots for contrasting risk score among the MRGs cluster and the DEGs patterns (Fig. 12A-B). The results revealed that the C2 and CA groups demonstrated notably elevated risk score in comparison to the C1 and CB groups. Additionally, a Sankey diagram was employed as a visual tool to illustrate the correlation between distinct clusters and risk score (Fig. 12C). Furthermore, we assessed the association between various immune checkpoints, revealing that levels of CD274 and TGFB1 were enhanced in the C1 and CB groups in comparison to the C2 and CA groups (Fig. 12D-E).

3.12 Single-Cell Sequencing Data Search for Gene Expression in Tumors

A comprehensive search of single-cell sequencing data for tumor gene expression was performed. We initially downloaded esophageal cancer single-cell datasets GSE160269 and GSE173950 from the TISCH2 database. The GSE160269 dataset, after dimension reduction, was annotated into 31 cell clusters, revealing 13 distinct cell types. In contrast, the GSE173950 dataset was annotated into 22 cell clusters, identifying 12 cell types, including both tumor and immune cells, and illustrating the proportional representation of different cell components in each patient (Fig. 13A-D). UMAP plots and violin plots were utilized to depict the expression profiles of relevant genes across different annotated cell types (Fig. 13E-J). The results demonstrated the expression of genes such as HSPD1 and MAP1LC3B across various annotated cell types, including tumor cells.

3.13 Prediction of Potential Traditional Chinese medicine and molecular docking

The two screened pivotal genes were imported into the Coremine database, and the top 30 corresponding traditional Chinese medicines were selected. Among them, the traditional Chinese medicine where both genes appeared simultaneously was Icaritin, suggesting that this traditional Chinese medicine has considerable potential for improving the prognosis of patients with esophageal cancer. Icaritin (Epimedium flavonoids) was selected to perform molecular docking with HSPD1 and MAP1LC3B proteins respectively (Fig. 14A-B).

A binding energy value of less than -4.25 kcal·mol-1 signifies a certain level of binding capability between the two entities, while a value less than -5.0 kcal·mol-1 indicates a more superior binding ability^[28]. The results showed that the core targets HSPD1 and MAP1LC3B had a better binding ability with Icaritin. The binding energy of Icaritin with HSPD1 was -9.3, and the binding energy with MAP1LC3B was -6.8, as shown in the Table 2.

Table 2 Binding Energy of Molecular Docking

Numerator ID	Name	Receptor protein	PDB ID	Binding ability[a]
5318980	Icaritin	HSPD1	4PJ1	-9.3
5318980	caritin	MAP1LC3B	6LAN	-6.8

[a] kcal·mol-1

Esophageal cancer ranks among the deadliest cancers globally, which has seriously endangered human health^[29]. It is detrimental for those suffering from esophageal cancer to wait for treatment to begin, regardless of whether the disease is primary or metastatic^[30]. Although immunotherapy for ESCA have been launched around the world in recent years, immunotherapies do not work well or benefit all patients in the majority of cases^[31]. The problems such as the high invasiveness of esophageal cancer, the unsatisfactory clinical treatment effect and the poor prognosis are still huge challenges. Mitophagy is thought to be significant to recycle mitochondrial mass and remove damaged mitochondria^[32]. Especially, the potential role in esophageal cancer, which is a highly invasive cancer, still needs to be further clarified.

This study is the first to systematically evaluate the role of mitophagy in esophageal cancer. We identified mitophagy genes related to the prognosis of ESCA based on various advanced algorithms and screened 14 differentially expressed genes for related subtype identification. Consensus clustering was performed for ESCA patients in the TCGA cohort, and patients were divided into C1 and C2 with significantly different prognoses with k = 2 as the boundary. Notable disparities in immune infiltration were observed between the two clusters when analyzed using the CIBERSORT, TIMER, and ESTIMATE algorithms. Cluster 1 had a higher proportion of immune cell infiltration levels, which also implied a better prognosis for Cluster 1. After that, esophageal cancer patients were categorized into two distinct molecular subtypes, CA and CB, on the basis of the differentially expressed genes between C1 and C2. Comprehensive immunological analysis and biological enrichment analysis showed that the variations observed in immune infiltration levels and levels of pathway enrichment between the two clusters suggested the prospect of immunotherapy for ESCA. evaluated the differential expression of immune checkpoint genes across the two subtypes. The results revealed that the two subtypes, C1 and CB, had higher immune checkpoint expression levels, which provided new insights for immune checkpoint-based therapies in esophageal cancer. All in all, we combined mitophagy with the prognosis and classification of esophageal cancer, a fresh insight into the complexities of esophageal cancer heterogeneity.

Our initial findings prompted the employment of two machine learning algorithms - LASSO regression and random forest - in conjunction with univariate and multivariate COX regression analysis. Ultimately, two pivotal genes, HSPD1 and MAP1LC3B, were identified and a prognostic model was constructed based on these findings. Our prognostic model was proved by the external validation set to have good predictive efficacy. The findings indicated a direct correlation between an elevated risk score and a poorer prognosis for patients. At the same time, through functional enrichment analysis, immune infiltration analysis and visualization of gene mutations, we revealed the complex differences between the high-risk and low-risk groups. In addition, we also constructed a nomogram for forecasting the prognosis of esophageal cancer patients, validated by calibration curves to exhibit robust predictive accuracy.

HSPD1 is a gene encoding the chaperonin family. During the cancer development process, the downregulation of HSPD1 is closely related to cancer cell apoptosis^[33]. The risk score's rise coincides with heightened mortality in ESCA patients, supporting the model's validity. In addition, as an effective prognostic marker, HSPD1 serves as a pivotal factor in cancer progression and has survival-promoting or apoptosis-inducing functions depending on the tumor type^[34]. Beatrice Parma et al. discovered that mitochondrial HSPD1 targeting is related to the metabolic damage of non-small cell lung cancer, and HSPD1 is widely expressed in NSCLC tumors and cells^[35]. Yu Zhang et al. identified the correlation between HSPD1 and mitochondrial autophagy in pituitary adenoma^[36]. Seon-Kyu Kim et al. found that the prognostic marker EHMT2 inhibits apoptosis by controlling the expression of HSPD1^[37]; The other hub gene MAP1LC3B, as a ubiquitin-like modifier involved in the formation of autophagosomes, meets the cellular energy requirements and prevents excessive ROS production by eliminating mitochondria to the basal level. In the study on the association between the expression pattern of MAP1LC3B and malignancy, an increase in expression level was related to a decrease in patient survival time. The viewpoint of our study is consistent with the histological experimental results^[38]. Haishun Qu et al. integrated CTH and MAP1LC3B to construct a prognostic model for gastric cancer. Survival analysis revealed a marked decline in survival rates among the high-risk group.

In the last few years, There has been a boom in single-cell sequencing in recent years^[39]. Single-cell sequencing technology and analytical tools have allowed oncologists to gain a better understanding of the tumor immune microenvironment and how it affects the antitumor immune response^[40]. Additionally, we further utilized the advanced single-cell sequencing technology to visualize the expression patterns of the two genes, HSPD1 and MAP1LC3B, in esophageal cancer tissues. The results suggested that HSPD1 and MAP1LC3B might be closely related to the immune microenvironment of esophageal cancer. Based on the above discoveries, we see hope in the treatment through immunotherapy in the new therapeutic approaches.

Chinese traditional medicine has a firmly established position in preventing and treating cancer. ^[41]. At the end of this study, based on the two genes, HSPD1 and MAP1LC3B, we predicted the traditional Chinese medicine - Icaritin through the Coremine database, which is greatly important for improving the prognosis of esophageal cancer patients. And the effective component - Icaritin was used to perform molecular docking with the two molecules, HSPD1 and MAPL1C3B. The results indicated a favorable docking interaction. In the study of Yang et al, Icaritin could be considered a promising agent for treating and preventing OSCC^[42]. It is suggested that Icaritin may have a positive effect in improving the prognosis of patients with esophageal cancer, but its clinical effect still needs further study.

This study has made significant progress on multiple levels. First of all, we deeply analyzed the core role of mitophagy in the development of esophageal cancer. A novel interplay among immune cells, tumor cells, and cancer stem cells is revealed within the ESCA tumor microenvironment^[43], which provides a novel perspective for the deeper comprehension of the disease. Secondly, we applied advanced machine learning techniques, which not only improved the identification accuracy of pivotal genes and signaling pathways, but also significantly enhanced the accuracy and reliability of prognosis assessment^[44]. Furthermore, combined with standardized single-cell sequencing data, we provided a detailed molecular map for revealing the heterogeneity of the tumor microenvironment^[45].

Nevertheless, this research also has several constraints. Our conclusion is based on the secondary bioinformatics analysis of public database data. Although the model performed well in the external validation set, there is a lack of direct evidence from animal experiments and cell experiments. Because the samples were derived from a retrospective study, there may be selective bias, which may affect the universality of the analysis results^[46]. Therefore, well-designed prospective studies are needed in the future to further validate our findings. Meanwhile, to understand the clinical significance of mitophagy more comprehensively, more clinical variables need to be considered. However, the collection of relevant data in the current public databases is still not perfect. Although we have preliminarily revealed the mechanism of mitophagy, its detailed molecular mechanism, regulatory network and interaction with other biological processes still need to be further explored.

To sum up, through the comprehensive application of advanced technical means and methods, this study systematically revealed the important role of mitophagy in esophageal cancer and its close relationship with prognosis and classification. This discovery not only broadens our knowledge of the molecular mechanism of esophageal cancer, but also provides important molecular markers and theoretical basis for achieving precision medicine for esophageal cancer. Future studies will further explore the mechanism of mitophagy, optimize technical methods and promote clinical transformation and application, in order to bring more effective treatment options and longer survival periods for patients.

By leveraging advanced technological approaches, this study systematically uncovered the pivotal role of mitophagy in esophageal cancer progression and its close association with prognosis and subtyping. The constructed prognostic model, featuring the key genes HSPD1 and MAP1LC3B, exhibited excellent predictive efficacy in external validation sets, while marked discrepancies were observed in gene expression patterns, functional enrichment, and immune cell infiltration between risk groups. We also successfully identified two subtypes (C1, C2 and CA, CB). Furthermore, single-cell sequencing illuminated the expression profiles of target genes in the tumor microenvironment, offering novel perspectives for immunotherapy applications. The predicted traditional Chinese medicines and their molecular docking outcomes presented promising avenues for improving esophageal cancer prognosis. This study significantly enriched our comprehension of esophageal cancer molecular mechanisms and provided essential molecular markers and theoretical frameworks for advancing precision medicine in esophageal cancer. Future endeavors will further explore mitophagy mechanisms, refine technical methodologies, and expedite clinical translations to offer more effective therapeutic options and prolonged survival for patients.

Funding

This study was funded by Detail Project of Precision Medicine Joint Fund of Natural Science Foundation of Hebei Province (Project No.H2021406066); National Natural Science Foundation Project Incubation Fund of Chengde Medical College (Project No.202416)

Conflict of interest

The authors declare no pertinent conflicts of interest with regard to the matter presented in this article.

Author Contribution

The conception and manuscript drafting of this study were conducted by FT. XH, SW, and ZW, MH were responsible for the data acquisition process. The comprehensive review and final approval of the manuscript were carried out by YL. YG oversaw all aspects of the manuscript including data interpretation, critical evaluation of the article, and ensuring its scientific rigor. Following a rigorous examination of the findings , all authors endorsed the manuscript for submission in its definitive version .

Data Availability

In this research, public datasets were scrutinized, including the TCGA-ESCA cohort sourced from the TCGA repository (accessible via http://cancergenome.nih.gov/) and GSE26886 along with GSE20347 retrieved from the GEO database (located at http://www.ncbi.nlm.nih.gov/geo/). The methodologies and software tools utilized for analysis are comprehensively detailed in the "Materials and Methods" section. The necessary data and materials were readily accessible when required.

Akhuj A, Athawale V, Fating T (2024) A Combat Journey of Rehabilitation in Pre- and Post-chemotherapy for Esophagus Carcinoma. Cureus 16(4):e58202. https://doi.org/10.7759/cureus.58202
Reijneveld EAE, Bor P, Dronkers JJ et al (2022) Impact of curative treatment on the physical fitness of patients with esophageal cancer: A systematic review and meta-analysis. Eur J Surg Oncol 48(2):391–402. https://doi.org/10.1016/j.ejso.2021.08.015
Qu HT, Li Q, Hao L et al (2024) Esophageal cancer screening, early detection and treatment: Current insights and future directions. World J Gastrointest Oncol 16(4):1180–1191. https://doi.org/10.4251/wjgo.v16.i4.1180
Sui X, Chen C, Zhou X et al (2023) Integrative analysis of bulk and single-cell gene expression profiles to identify tumor-associated macrophage-derived CCL18 as a therapeutic target of esophageal squamous cell carcinoma. J Exp Clin Cancer Res 42(1):51. https://doi.org/10.1186/s13046-023-02612-5
Silk AW, Barker CA, Bhatia S et al (2022) Society for Immunotherapy of Cancer (SITC) clinical practice guideline on immunotherapy for the treatment of nonmelanoma skin cancer. J Immunother Cancer 10(7):e004434. https://doi.org/10.1136/jitc-2021-004434
Liu X, Chen Y, Zhang S, Dong L (2021) Gut microbiota-mediated immunomodulation in tumor. J Exp Clin Cancer Res 40(1):221. https://doi.org/10.1186/s13046-021-01983-x
Lu Y, Li Z, Zhang S, Zhang T, Liu Y, Zhang L (2023) Cellular mitophagy: Mechanism, roles in diseases and small molecule pharmacological regulation. Theranostics 13(2):736–766. https://doi.org/10.7150/thno.79876
D'Arcy MS (2024) Mitophagy in health and disease. Molecular mechanisms, regulatory pathways, and therapeutic implications. Apoptosis. https://doi.org/10.1007/s10495-024-01977-y
Lemasters JJ (2005) Selective mitochondrial autophagy, or mitophagy, as a targeted defense against oxidative stress, mitochondrial dysfunction, and aging. Rejuvenation Res 8(1):3–5. https://doi.org/10.1089/rej.2005.8.3
Abeliovich H (2007) Mitophagy: the life-or-death dichotomy includes yeast. Autophagy 3(3):275–277. https://doi.org/10.4161/auto.3915
Youle RJ, Narendra DP (2011) Mechanisms of mitophagy. Nat Rev Mol Cell Biol 12(1):9–14. https://doi.org/10.1038/nrm3028
Su L, Zhang J, Gomez H et al (2023) Mitochondria ROS and mitophagy in acute kidney injury. Autophagy 19(2):401–414. https://doi.org/10.1080/15548627.2022.2084862
Song C, Pan S, Zhang J et al (2022) Mitophagy: A novel perspective for insighting into cancer and cancer treatment. Cell Prolif 55(12):e13327. https://doi.org/10.1111/cpr.13327
Panigrahi DP, Praharaj PP, Bhol CS et al (2020) The emerging, multifaceted role of mitophagy in cancer and cancer therapeutics. Semin Cancer Biol 66:45–58. https://doi.org/10.1016/j.semcancer.2019.07.015
Ritchie ME, Phipson B, Wu D et al (2015) limma powers differential expression analyses for RNA-sequencing and microarray studies. Nucleic Acids Res 43(7):e47. https://doi.org/10.1093/nar/gkv007
Engebretsen S, Bohlin J (2019) Statistical predictions with glmnet. Clin Epigenetics. ;11(1):123. Published 2019 Aug 23. https://doi.org/10.1186/s13148-019-0730-1
Alderden J, Pepper GA, Wilson A et al (2018) Predicting Pressure Injury in Critical Care Patients: A Machine-Learning Model. Am J Crit Care 27(6):461–468. https://doi.org/10.4037/ajcc2018525
Vickers AJ, Cronin AM, Elkin EB et al (2008) Extensions to decision curve analysis, a novel method for evaluating diagnostic tests, prediction models and molecular markers. BMC Med Inf Decis Mak 8:53 Published 2008 Nov 26. https://doi.org/10.1186/1472-6947-8-53
Miao Y, Yuan Q, Wang C et al (2023) Comprehensive Characterization of RNA-Binding Proteins in Colon Adenocarcinoma Identifies a Novel Prognostic Signature for Predicting Clinical Outcomes and Immunotherapy Responses Based on Machine Learning. Comb Chem High Throughput Screen 26(1):163–182. https://doi.org/10.2174/1386207325666220404125228
Zhang Y, Liu N, Wang S (2018) A differential privacy protecting K-means clustering algorithm based on contour coefficients. PLoS ONE 13(11):e0206832. https://doi.org/10.1371/journal.pone.0206832
Yoshihara K, Shahmoradgoli M, Martínez E et al (2013) Inferring tumour purity and stromal and immune cell admixture from expression data. Nat Commun 4:2612. https://doi.org/10.1038/ncomms3612
Li T, Fan J, Wang B et al (2017) TIMER: A Web Server for Comprehensive Analysis of Tumor-Infiltrating Immune Cells. Cancer Res 77(21):e108–e110. https://doi.org/10.1158/0008-5472.CAN-17-0307
Becht E, Giraldo NA, Lacroix L et al (2016) Estimating the population abundance of tissue-infiltrating immune and stromal cell populations using gene expression [published correction. appears Genome Biol 17(1):249. https://doi.org/10.1186/s13059-016-1070-5
Hänzelmann S, Castelo R, Guinney J (2013) GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 14:7 Published 2013 Jan 16. https://doi.org/10.1186/1471-2105-14-7
Subramanian A, Tamayo P, Mootha VK et al (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102(43):15545–15550. https://doi.org/10.1073/pnas.0506580102
Han Y, Wang Y, Dong X et al (2023) TISCH2: expanded datasets and new tools for single-cell transcriptome analyses of the tumor microenvironment. Nucleic Acids Res 51(D1):D1425–D1431. https://doi.org/10.1093/nar/gkac959
Maleeva G, Wutz D, Rustler K et al (2019) A photoswitchable GABA receptor channel blocker. Br J Pharmacol 176(15):2661–2677. https://doi.org/10.1111/bph.14689
Jiang Wenjie L, Jiaqing N, Yang et al To explore the mechanism of pine-rhubarb-patchcompatibility in renal clear cell carcinoma based on network pharmacology and molecular docking. J Ningxia Med Univ (04), 413–421. 10.16050/j.cnki.issn1674-6309.2024.04.014
Moon H, Zhu J, Donahue LR et al (2019) Krt5⁺/Krt15⁺ foregut basal progenitors give rise to cyclooxygenase-2-dependent tumours in response to gastric acid stress. Nat Commun 10(1):2225 Published 2019 May 20. https://doi.org/10.1038/s41467-019-10194-0
Foster JM, Sleightholm R, Patel A et al (2019) Morbidity and Mortality Rates Following Cytoreductive Surgery Combined With Hyperthermic Intraperitoneal Chemotherapy Compared With Other High-Risk Surgical Oncology Procedures. JAMA Netw Open 2(1):e186847 Published 2019 Jan 4. https://doi.org/10.1001/jamanetworkopen.2018.6847
Li Y, Yang C, Liu Z et al (2022) Integrative analysis of CRISPR screening data uncovers new opportunities for optimizing cancer immunotherapy. Mol Cancer 21(1):2 Published 2022 Jan 2. https://doi.org/10.1186/s12943-021-01462-z
Li R, Xin T, Li D et al (2018) Therapeutic effect of Sirtuin 3 on ameliorating nonalcoholic fatty liver disease: The role of the ERK-CREB pathway and Bnip3-mediated mitophagy. Redox Biol 18:229–243. https://doi.org/10.1016/j.redox.2018.07.011
Kim SK, Kim K, Ryu JW et al (2019) The novel prognostic marker, EHMT2, is involved in cell proliferation via HSPD1 regulation in breast cancer. Int J Oncol 54(1):65–76. https://doi.org/10.3892/ijo.2018.4608
Yun CW, Kim HJ, Lim JH et al (2019) Heat Shock Proteins: Agents of Cancer Development and Therapeutic Targets in Anti-Cancer Therapy. Cells 9(1):60 Published 2019 Dec 24. https://doi.org/10.3390/cells9010060
Parma B, Ramesh V, Gollavilli PN et al (2021) Metabolic impairment of non-small cell lung cancers by mitochondrial HSPD1 targeting. J Exp Clin Cancer Res 40(1):248. https://doi.org/10.1186/s13046-021-02049-8
Zhang Y, Ma X, Liu C et al (2024) Identification of HSPD1 as a novel invasive biomarker associated with mitophagy in pituitary adenomas. Transl Oncol 41:101886. https://doi.org/10.1016/j.tranon.2024.101886
Kim SK, Kim K, Ryu JW et al (2019) The novel prognostic marker, EHMT2, is involved in cell proliferation via HSPD1 regulation in breast cancer. Int J Oncol 54(1):65–76. https://doi.org/10.3892/ijo.2018.4608
Spirina LV, Avgustinovich AV, Bakina OV et al (2022) LC3B, mTOR, AMPK Are Molecular Targets for Neoadjuvant Chemotherapy in Gastric Cancers. Curr Issues Mol Biol 44(7):2772–2782. https://doi.org/10.3390/cimb44070190
Xu J, Liao K, Yang X et al (2022) Correction to: Using single-cell sequencing technology to detect circulating tumor cells in solid tumors. Mol Cancer 21(1):100. https://doi.org/10.1186/s12943-022-01564-2
Jia Q, Chu H, Jin Z et al (2022) High-throughput single-сell sequencing in cancer research. Signal Transduct Target Ther 7(1):145. https://doi.org/10.1038/s41392-022-00990-4
Liu H, Huang M, Xin D, Wang H, Yu H, Pu W (2024) Natural products with anti-tumorigenesis potential targeting macrophage. Phytomedicine 131:155794. https://doi.org/10.1016/j.phymed.2024.155794
Yang JG, Lu R, Ye XJ et al (2017) Icaritin Reduces Oral Squamous Cell Carcinoma Progression via the Inhibition of STAT3 Signaling. Int J Mol Sci 18(1):132 Published 2017 Jan 12. https://doi.org/10.3390/ijms18010132
Yang K, Xie Y, Xue L et al (2023) M2 tumor-associated macrophage mediates the maintenance of stemness to promote cisplatin resistance by secreting TGF-β1 in esophageal squamous cell carcinoma. J Transl Med 21(1):26. https://doi.org/10.1186/s12967-022-03863-0
Romero-Brufau S, Whitford D, Johnson MG et al (2021) Using machine learning to improve the accuracy of patient deterioration predictions: Mayo Clinic Early Warning Score (MC-EWS). J Am Med Inf Assoc 28(6):1207–1215. https://doi.org/10.1093/jamia/ocaa347
Xie S, Cai Y, Chen D et al (2023) Single-cell transcriptome analysis reveals heterogeneity and convergence of the tumor microenvironment in colorectal cancer. Front Immunol 13:1003419 Published 2023 Jan 4. https://doi.org/10.3389/fimmu.2022.1003419
Hughes MSA, Beganovic M (2019) Introduction of Selection Biases Due to Loss to Follow-Up in Infectious Disease Retrospective Outcomes Studies. Antimicrob Agents Chemother 63(12):e01681–e01619. https://doi.org/10.1128/AAC.01681-19

No competing interests reported.

Download PDF

Editorial decision: Revision requested
03 Nov, 2024
Reviews received at journal
02 Nov, 2024
Reviewers agreed at journal
22 Oct, 2024
Reviewers agreed at journal
15 Oct, 2024
Reviewers agreed at journal
14 Sep, 2024
Reviewers invited by journal
14 Sep, 2024
Editor assigned by journal
16 Aug, 2024
Submission checks completed at journal
16 Aug, 2024
First submitted to journal
15 Aug, 2024

You are reading this latest preprint version

Integrating Single-Cell Sequencing and Machine Learning to Uncover the Role of Mitophagy in Subtyping and Prognosis of Esophageal Cancer

Status:

Version 1

Abstract

Figures

1 Introduction

2 Materials and Methods

2.1 Data Collection and Preprocessing

2.2 Identification of Differentially Expressed Genes Related to Mitophagy

2.3 Cox regression analysis

2.4 Machine Learning

2.5 Development and verification of a risk assessment model

2.6 Consensus Clustering

2.7 Immune Infiltration

2.8 Enrichment Analysis

2.9 Mitophagy Expression in Single-Cell RNA Sequencing Tumor Data

2.10 Prediction of Traditional Chinese medicine and molecular docking

3 Results

3.1 Identification of Prognostic Mitophagy-Related Genes (MRGs)

3.2 Identification of MRG-related Subtypes Based on 14 MRGs

3.3 Immunological Landscape of the Two Subtypes

3.4 Differentially Expressed Genes (DEGs) and their Functional Annotations

3.5 Further Identification of ESCA Based on DEGs Between Subtypes

3.6 Discerning biological functions and immune microenvironments across distinct gene clusters.

3.8 Independence of the Established Risk Model

4 Discussion

5 Conclusion

Declarations

Funding

Author Contribution

Data Availability

References

Additional Declarations

Status:

Version 1