Predictive value of MHC-related genes in cervical cancer: implications for immunotherapy and prognostic nomogram development

Abstract

Background

While cervical cancer (CC) is increasingly detected at earlier stages due to improved screening, clinical dilemmas persist in stratifying precancerous lesions and managing advanced disease. Major histocompatibility complex-related genes (MHCRGs), which regulate antigen presentation and immune surveillance, may serve as promising biomarkers for prognosis and therapeutic guidance in CC.

Methods

We identified MHCRGs via literature review and constructed a prognostic signature using LASSO and multivariate Cox regression based on TCGA-CESC data. Bioinformatic analyses assessed associations between the MHCRG signature, immune cell infiltration, immunotherapy response, and drug sensitivity. Validation was conducted using external GEO datasets (GSE29570, GSE63514), and further confirmed through qRT-PCR and Western blot in clinical CC samples. Functional assays, including gene knockdown in CC cell lines, were performed to explore the biological roles of key MHCRGs in tumor progression.

Results

A five-gene signature (CANX, HLA-DMB, HLA-DPB1, PSMB6, PSMB7) was identified as predictive of prognosis, immune infiltration, and therapeutic responsiveness. CANX was significantly upregulated, while HLA-DMB and HLA-DPB1 were downregulated in high-risk groups across all datasets and clinical tissues. qRT-PCR and WB results aligned with bioinformatic predictions. Functional experiments demonstrated that silencing CANX suppressed CC cell proliferation and invasion. These findings support the oncogenic role of CANX and the immune-regulatory potential of the other MHCRGs. The MHCscore remained an independent prognostic factor in multivariate analysis.

Conclusions

We propose a robust MHCRG-based gene signature, validated through both bioinformatic analyses and in vitro functional experiments. These findings offer translational potential for improving risk stratification and guiding immunotherapy in CC patients.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-03460-9.

Introduction

Cervical cancer (CC) remains a global public health issue, ranking as the fourth most common malignancy in women. In 2022, approximately 660,000 new cases and 350,000 deaths were reported worldwide, with the burden disproportionately affecting low- and middle-income countries [1]. The introduction of Human papillomavirus (HPV) vaccination and cytological screening programs has significantly reduced incidence and mortality in many regions, leading to earlier detection and better outcomes [2]. However, clinical challenges remain at both ends of the disease spectrum: for early-stage or precancerous lesions such as cervical intraepithelial neoplasia (CIN), it remains difficult to distinguish those likely to progress; while for advanced or recurrent CC, treatment options are often limited and prognoses remain poor [3].

To address these challenges, reliable biomarkers are needed to guide clinical decision-making—whether to inform risk-based surveillance in early-stage cases or to predict therapeutic responses in advanced-stage patients. Immune-related gene signatures have shown promise in this regard. Among them, major histocompatibility complex-related genes (MHCRGs) have emerged as critical regulators of tumor immunogenicity. The MHC-I antigen presentation pathway, essential for CD8 + T cell-mediated cytotoxicity, can be disrupted by tumor cells to escape immune detection [4]. Recent research has demonstrated that alterations in MHCRGs could influence tumor immune evasion, patient outcomes, and immunotherapeutic responses across various malignancies [5]. In hepatocellular carcinoma (HCC), MHCRGs are linked to tumor mutation burden and immune checkpoint inhibitor efficacy [6]. In CC cancers, MHC II is progressively dysregulated to evade host anti-tumor responses during the transition from low-grade squamous intraepithelial lesion (LSIL) to high-grade squamous intraepithelial lesion (HSIL) and ultimately to squamous cell carcinoma (SCC) [7]. HPV-positive CCs exhibit higher expression of genes linked to MHC Class I and Class II antigen presentation (AP) compared to HPV-negative patients [8]. In addition, deletion of MHC class I expression in individuals with CC might diminish the effectiveness of checkpoint inhibitors that target the PD-1/PD-L1 axis [9]. MHCRGs serve a pivotal function in tumor immunity and CC progression.

The integration of bioinformatics tools with clinical phenotypes offers promising avenues for enhancing CC management. By dissecting the genetic landscape and immune profiles of CC through high-throughput data analysis, researchers could identify novel biomarkers for prognosis and potential therapeutic targets [10]. Furthermore, understanding the molecular heterogeneity associated with different CC phenotypes could lead to personalized medicine strategies that markedly improve patient outcomes [11]. Such advancements in deciphering disease mechanisms at the molecular level are pivotal for developing next-generation diagnostics and therapeutics tailored to individual patient profiles.

Our study employs bioinformatics methods to analyze the cancer genome atlas-cervical squamous cell carcinoma and endocervical adenocarcinoma (TCGA-CESC) datasets for genetic alterations, immune infiltration patterns, drug sensitivity (DST) profiles, and prognostic indicators. By constructing a prognostic model (PM) based on MHCRGs through Lasso regression with cross-validation, we seek to categorize patients into risk cohorts with distinct survival outcomes. Additionally, we explore differential gene expression between these risk cohorts to elucidate underlying biological processes contributing to disease progression. Against this backdrop, we investigated the prognostic and functional relevance of MHCRGs in CC, aiming to identify biomarkers with potential utility across both early lesion risk stratification and advanced disease management.

Materials and methods

Data acquisition and processing

The CC dataset (TCGA-CESC) [12] was downloaded using the TCGAbiolinks R package from TCGA database (https://portal.gdc.cancer.gov/). After excluding samples lacking clinical data, 306 CC samples with clinical information were retained. RNA-seq expression data were normalized to FPKM values. Clinical data were obtained from UCSC Xena [13] (http://genome.ucsc.edu). Microsatellite Instability (MSI) data were obtained from cBioPortal [14] (https://www.cbioportal), and immune scores were calculated using the Tumor Immune Dysfunction and Exclusion (TIDE) [15] website (http://tide.dfci.harvard.edu) tool. From the TCGA portal (https://portal.gdc.cancer.gov/), we procured “Masked Somatic Mutation” data for CC in patients with single nucleotide polymorphisms (SNPs). The SNP data underwent preprocessing utilizing VarScan software, and visualization of somatic mutations was performed utilizing the maftools R [16, 17] package. To investigate the Copy Number Variations (CNV) of TCGA-CESC patients, we procured the “Masked Copy Number Segment” data of the individuals utilizing R’s TCGAbiolinks package. Then, we executed GISTIC 2.0 analysis on the procured and processed CNV segments, and all the default parameters were used in the process of GISTIC 2.0 analysis [18]. Additionally, we acquired the expression profile datasets GSE29570 [19] and GSE63514 [20] of CC patients from the GEO database [21] utilizing the R package GEOquery [22]. GSE29570 includes 45 CC biopsies and 17 matched normal exocervix samples (the outer portion of the cervix lined by squamous epithelium) profiled with GPL6244 (Affymetrix Human Gene 1.0 ST Array), while GSE63514 contains 28 CC tissue samples and 76 cervical intraepithelial neoplasm samples profiled with GPL570 (Affymetrix Human Genome U133 Plus 2.0 Array). Expression data from the CC samples in both datasets were used for validation (Table 1).

Table 1.

List of CC information

	TCGA-CESC	GSE29570	GSE63514
Platform	TCGA	GPL96	GPL570
Species	Homo sapiens	Homo sapiens	Homo sapiens
Tissue	cervical epitheilia tissue	cervical epitheilia tissue	cervical epitheilia tissue
Samples in CC group	306	45	28
Reference	–	The Amerindian mtDNA haplogroup B2 enhances the risk of HPV for cervical cancer: de-regulation of mitochondrial genes may be involved	Molecular transitions from papillomavirus infection to cervical precancer and cancer: Role of stromal estrogen receptor signaling

To analyze immune-related genes, we obtained a set of 20 MHCRGs by searching “Major histocompatibility complex” in PubMed [23], all present in TCGA-CESC, GSE29570, and GSE63514 (Table 2). Using GeneCards [24], 12 immune checkpoint genes (ICGs) with protein-coding status and relevance score > 10,000, and 19 protein-coding HLA genes (symbols starting with “HLA”) were also retrieved.

Table 2.

Major histocompatibility complex related genes list

Major histocompatibility complex related genes
B2M	CALR	CANX	ERAP1	ERAP2	HLA-DMA
HLA-DMB	HLA-DOA	HLA-DOB	HLA-DPA1	HLA-DPB1	HLA-DQB1
PSMB10	PSMB5	PSMB6	PSMB7	PSMB8	PSMB9
TAP2	TAPBP

TCGA the cancer genome atlas, CES cervical endocervical adenocarcinoma and squamous cell carcinoma.

Construction of MHCRGs prognostic model (PM)

To develop a PM for MHCRGs in CC, we employed Least Absolute Shrinkage and Selection Operator (LASSO) regression on the TCGA-CESC dataset. We utilized 10-fold cross-validation to determine the optimal penalty parameter (lambda) that minimizes the cross-validated error, and performed 1000 iterations to improve the model’s stability and reduce overfitting [25]. LASSO regression, an extension of linear regression incorporating an L1 penalty term, is widely used for feature selection and model regularization. The LASSO regression results were visualized, and the resulting prognostic model stratified samples based on risk scores and survival outcomes. We also investigated the expression patterns of MHCRGs across risk groups using risk factor plots. To corroborate our findings, we examined the expression disparities of MHCRGs between high-risk cohorts and low-risk cohorts derived from the Lasso regression model across the TCGA-CESC, GSE29570, and GSE63514 datasets. To identify differentially expressed genes (DEGs) between high-risk and low-risk cohorts in the TCGA-CESC dataset, we used the limma package [26] to analyze transcriptomic expression data. Gene with |log2FC| >1 and adjusted p < 0.05 were considered significantly differentially expressed. Specifically, genes with log2FC > 1 were classified as upregulated, while those with log2FC < −1 were considered downregulated. The results of the differential analysis were visualized using a volcano plot generated with the ggplot2 R package.

Functional and pathway enrichment analyses

We conducted Gene Ontology (GO) [27] and Kyoto Encyclopedia of Genes and Genomes (KEGG) [28] enrichment analyses to explore the biological functions and pathways associated with MHCRGs in the TCGA-CESC dataset. These analyses were performed using the clusterProfiler R package [29], and terms with adjusted p < 0.05 and FDR (q-value) < 0.05 were considered significant. The Benjamini-Hochberg method was used for multiple testing correction.

GSEA and GSVA between risk groups

Gene Set Enrichment Analysis (GSEA) [30] was used to investigate the distribution of predefined gene sets between high-risk and low-risk cohorts in the TCGA-CESC dataset. The analysis was conducted using clusterProfiler with the following settings: seed = 2023, permutations = 1000, gene set size = 10–500. Gene sets were obtained from MSigDB (c2.cp.all.v2022.1.Hs.symbols.gmt), and results with adjusted p < 0.05 and FDR < 0.25 were considered significant. Gene Set Variation Analysis (GSVA) [31] was applied to evaluate pathway activity differences between HR and LR groups. The gene set “h.all.v7.4.symbols.gmt” was downloaded from MSigDB [32] and used to assess enrichment across samples. Pathways with adjusted p < 0.05 were considered significantly enriched.

Immune infiltration analysis via ssGSEA

To evaluate the tumor microenvironment (TME), we applied single-sample Gene Set Enrichment Analysis (ssGSEA) using the GSVA R package. We incorporated 28 immune-related gene sets from a published study [33], representing diverse tumor-infiltrating immune cell types, including CD8⁺ T cells, dendritic cells, macrophages, and regulatory T cells. ssGSEA enrichment scores quantified the infiltration level of each IC type in individual TCGA-CESC specimens. Boxplots were generated to compare immune cell infiltration (ICI) levels between high-risk and low-risk cohorts of the MHCRGs PM. Pearson correlation was used to assess the relationships among immune cells, visualized using ggplot2. Additionally, we analyzed correlations between MHCRGs and ICI levels within the high-risk and low-risk groups, presenting the results as scatter plots.

Immune microenviroment and correlation analysis

We further characterized the immune landscape using the ESTIMATE algorithm [34], which infers immune and stromal cell content based on tumor gene expression profiles. From the TCGA-CESC dataset, we derived immune scores to quantify the level of immune infiltration and compared these scores between high-risk and low-risk groups of the MHCRGs PM. To predict potential response to immune checkpoint blockade (ICB) therapy, we used the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm [35] (http://tide.dfci.harvard.edu). TIDE models two major immune escape mechanisms: T cell dysfunction in cytotoxic T lymphocyte-rich tumors and T cell exclusion in cytotoxic T lymphocyte-poor tumors. We assessed TIDE scores across high-risk and low-risk cohorts and evaluated group differences using the Mann–Whitney U test (Wilcoxon rank-sum test), with p < 0.05 considered statistically significant.

Drug sensitivity analysis

Genomic alterations in cancer significantly influence therapeutic response and can serve as predictive biomarkers for drug efficacy. We utilized the Genomics of Drug Sensitivity in Cancer (GDSC) database [36] to predict drug response in the TCGA-CESC cohort. The pRRophetic algorithm [37] was applied to estimate the half-maximal inhibitory concentration (IC50) of common chemotherapeutics and small-molecule agents based on the FPKM expression matrix. IC50 differences between high-risk and low-risk cohorts of the MHCRGs PM were visualized using group comparison plots.

MHC phenotype scoring via sGSEA

To explore the biological relevance of MHCRGs in CC, we calculated major histocompatibility complex (MHC) phenotype scores for each TCGA-CESC sample using the GSVA R package and the ssGSEA algorithm [38]. Scores were derived from the MHCRG expression matrix. Based on the median MHC score, patients were categorized into high-score and low-score groups for downstream analysis.

Prognostic clinical correlation analysis

We conducted univariate and multivariate Cox regression analyses to evaluate the prognostic value of MHC scores and clinical features in TCGA-CESC patients. Variables with prognostic significance were incorporated into a multivariate model, and a nomogram was constructed using the rms R package [39] to visualize individualized survival predictions. Calibration curves [40] were used to compare predicted versus observed survival probabilities. Decision curve analysis (DCA) [41] was performed to evaluate clinical utility and net benefit across risk thresholds. DCA plots were generated using the ggDCA R package [42].

Clinical sample collection and validation

Fresh CC tissues and matched adjacent normal cervical epithelial tissues were obtained from patients who underwent radical surgery for CC at Zhongshan Hospital and Fudan University Cancer Hospital between November 2018 and August 2019. All patients were pathologically confirmed to have cervical squamous cell carcinoma (SCC) following surgery. Inclusion criteria were: histologically diagnosed cervical SCC, FIGO stage Ib1 or IIa1, age between 18 and 70 years, and no history of neoadjuvant chemotherapy or radiotherapy prior to surgery. Tumor tissues were excised from the central area of the grossly visible lesion. Matched adjacent normal tissues were obtained from cervical epithelium at least 2 cm away from the tumor margin and appeared grossly unremarkable. All adjacent tissues were confirmed to be histologically normal by pathologist review of formalin-fixed, paraffin-embedded sections stained with hematoxylin and eosin (H&E). Only tissues free from tumor infiltration, dysplasia, or other pathological changes were included in the study. All tissue samples were immediately rinsed in cold PBS, snap-frozen in liquid nitrogen within 30 min, and stored at − 80 °C until further analysis. Total RNA was extracted using TRIzol reagent (Invitrogen), and quantitative real-time PCR (qRT-PCR) was performed to assess the expression of selected MHCRGs.

Cell culture and functional assays

Human cervical cancer cell lines (HeLa) were obtained from the Cell Bank of the Chinese Academy of Sciences (Shanghai, China). Cells were cultured in DMEM supplemented with 10% fetal bovine serum at 37 °C in a humidified incubator with 5% CO₂. Cells were transfected with siRNA targeting CANX using Lipofectamine 3000 (Thermo Fisher) per manufacturer instructions. Knockdown efficiencies were confirmed by qRT-PCR and Western blot. All reagents were purchased from commercial suppliers, and detailed information of primers sequences is provided in Supplementary Table S1. Cell migration abilities were evaluated by Transwell assays with Matrigel coating. Apoptosis rates were analyzed by flow cytometry using Annexin V-FITC/PI staining.

Statistical analysis

All statistical analyses were performed using R software (version 4.2.2). For continuous variables with normal distribution, the independent Student’s t-test was used to compare differences between groups. The Mann-Whitney U test (Wilcoxon rank-sum test) was applied for non-normally distributed data. Categorical variables were analyzed using the chi-square test or Fisher’s exact test, as appropriate. Survival analyses were conducted using the survival R package. Kaplan-Meier curves illustrated survival differences between cohorts, and the log-rank test assessed statistical significance. Univariate and multivariate Cox regression analyses were also performed using the survival package. LASSO regression was conducted with the glmnet R package [43] to construct the prognostic model. All P values were two-sided, with P < 0.05 considered statistically significant.

Results

Standardization of CC dataset

The bioinformatics analysis workflow of this study is depicted in Fig. 1. To ensure comparability across datasets, we performed data normalization on the TCGA-CESC, GSE29570, and GSE63514 CC datasets using the R package limma package (Supplementary Fig. 1A–F). The TCGA-CESC dataset comprised 306 CC samples (Supplementary Fig. 1A, B), the GSE29570 dataset included 45 CC samples (Supplementary Fig. 1C, D), and the GSE63514 dataset contained 28 CC samples (Supplementary Fig. 1E, F). Post-normalization, the expression profiles across all three datasets exhibited high consistency, confirming successful data standardization and enabling downstream integrative analyses.

Fig. 1.

Technology roadmap

Mutation analysis of MHCRGs in CC patients

To investigate the somatic mutation profiles of 20 MHCRGs in CC patients, we performed mutation analysis using the TCGA-CESC dataset. Results were computed and visualized using the R package maftools (Fig. 2A–D). Seven primary types of somatic alterations were detected in MHCRGs: Missense Mutation; Nonsense Mutation; Frame_Shift_Del; Splice_Site Mutation; Translation_Start_Site Mutation. Among these, missense mutations were the most prevalent (Fig. 2A). Further analysis revealed that the majority of mutations were SNPs, with a minority comprising deletions. C > T transitions were the most frequent single nucleotide variant (SNV), followed by C > G and C > A changes. Among the MHCRGs, TAP2 exhibited the highest mutation frequency, presenting three different mutation types. The top 10 most frequently mutated MHCRGs in CC patients were: TAP2, TAPBP, B2M, ERAP1, HLA-DMA, PSMB8, PSMB10, ERAP2, CANX, and CALR (Fig. 2B). We further examined copy number alterations in the TCGA-CESC dataset. The top 10 MHCRGs with the highest CNA frequencies were identified (Fig. 2C). CNAs were prevalent across samples and predominantly exhibited copy number gains. Genes such as HLA-DOA, HLA-DOB, HLA-DPA1, HLA-DPB1, HLA-DQB1, and PSMB8 showed amplification rates ranging from 10 to 20%, whereas PSMB6 showed a high frequency of copy number loss (Fig. 2D). These findings highlight PSMB6 as one of the MHCRGs with recurrent CNA loss in CC.

Fig. 2.

Mutation analysis of MHCRGs in CC patients. A Distribution and classification of somatic mutations in MHCRGs. B Frequency ranking of somatic mutations in MHCRGs. C Copy number alteration (CNA) frequencies of MHCRGs. D CNA alteration types (gain/loss) across MHCRGs

Development of the PM and clinical correlation analysis of MHCRGs

To construct a robust PM based on the expression levels of 20 MHCRGs, we applied the LASSO Cox regression using the glmnet R package on the TCGA-CESC cohort. The optimal penalty parameter (λ) was determined through 10-fold cross-validation, which identified the λ value that minimized the partial likelihood deviance. This process ensured that only the most predictive genes were retained, thereby reducing model overfitting and enhancing generalizability. The LASSO coefficient trajectory (Fig. 3A) and cross-validation curve (Fig. 3B) supported the selection of five MHCRGs (CANX, HLA-DMB, HLA-DPB1,PSMB6, and PSMB7) for the final prognostic signature. We further assessed the model’s prognostic performance using the concordance index (C-index) and Akaike Information Criterion, which confirmed its stability and discriminative ability. A RiskScore was computed for each patient based on the linear combination of expression levels of the five genes weighted by their respective LASSO coefficients. Patients were then stratified into high-risk and low-risk groups using the median RiskScore as the threshold, a commonly adopted and data-driven approach in survival modeling. This median-based cutoff was consistently applied to the validation cohorts (GSE29570 and GSE63514), enabling uniform risk stratification across datasets. The classification outcomes are shown in the risk plot (Fig. 3C), which includes (1) risk classification; (2) survival status and duration; and (3) heatmap of gene expression patterns for the five MHCRGs. To explore transcriptional differences between high- and low-risk groups defined by the MHCRGs PM, differentially expression analysis was performed using the limma R package. In the TCGA-CESC dataset, a total of 41,049 genes were analyzed, with 689 DEGs identified under the criteria of |log2FC| >1 and adjusted p-value < 0.05. Among these DEGs, 120 genes were upregulated in the high-risk group, and 569 genes were downregulated in the high-risk group. These findings are illustrated in a volcano plot (Fig. 3D). Additionally, we examined the differential expression of the five MHCRGs between the high- and low-risk groups, with detailed results provided in Supplementary Table 2. Survival differences between the high- and low-risk groups were assessed using Kaplan-Meier (KM) survival analysis. The KM curve demonstrated a statistically significant difference in overall survival (OS) between the two groups (P < 0.05, Fig. 3E), confirming the prognostic value of the MHCRG-based RiskScore. We further evaluated the time-dependent predictive performance of the MHCRG-based PM using receiver operating characteristic (ROC) curves. The area under the curve (AUC) values for 1-year, 3-year, and 5-year OS were 0.668, 0.652, and 0.662, respectively (Fig. 3F). These AUCs indicate moderate predictive accuracy of the model in forecasting patient survival.

Fig. 3.

Construction and validation of the MHCRG prognostic model using LASSO-Cox regression. A Coefficient profiles of 20 MHCRGs from LASSO regression. B Ten-fold cross-validation to select the optimal λ. Five genes were retained at the optimal λ. C Distribution of RiskScore, survival status, and heatmap of gene expression for the five MHCRGs in the TCGA-CESC cohort. D Volcano plot displaying DEGs between high- and low-risk groups. E Kaplan-Meier survival curve comparing OS between high- and low-risk groups (P < 0.05). F Time-dependent ROC curves for 1-, 3-, and 5-year OS based on RiskScore in TCGA-CESC cohort (AUC1 = 0.668, AUC3 = 0.652, AUC5 = 0.662). Note: In ROC analysis, AUC > 0.9 indicates high accuracy, 0.7–0.9 indicates moderate accuracy, and 0.5–0.7 suggests low accuracy. AUC values closer to 1 imply better predictive performance

Differential expression analysis of MHCRGs PM in CC patients

To explore the expression patterns of the five MHCRGs in different risk subgroups of CC patients, we performed DEA across the TCGA-CESC, GSE29570, and GSE63514 datasets. Using the gene expression levels and the LASSO-derived coefficients, we computed the RiskScore for each sample in the GSE29570 and GSE63514 datasets. Patients in each cohort were subsequently stratified into HR and LR groups based on the median RiskScore. We applied the Mann–Whitney U test (Wilcoxon rank-sum test) to compare the expression levels of the five MHCRGs between HR and LR groups across the three datasets. The results, visualized via gene cluster diagrams (GCDs), are presented in Figs. 4A, G, H. In the TCGA-CESC dataset, the results showed that CANX expression was significantly upregulated in the HR group compared to the LR group (P < 0.001), while the expression levels of HLA-DMB, HLA-DPB1, PSMB6, and PSMB7 were significantly downregulated in the HR group (P < 0.001) (Fig. 4A). In the GSE29570 dataset, CANX expression was significantly higher in the HR group (P < 0.05), whereas HLA-DMB and HLA-DPB1 were significantly downregulated (P < 0.001). However, PSMB6 and PSMB7 showed no statistically significant differences between the HR and LR groups (P > 0.05) (Fig. 4G). Similarly, in the GSE63514 dataset, CANX expression was significantly elevated (P < 0.001), and HLA-DMB and HLA-DPB1 were significantly downregulated (P < 0.001) in the HR group, whereas PSMB6 and PSMB7 exhibited no significant expression changes between the groups (P > 0.05) (Fig. 4H). To assess the diagnostic performance of these five genes in distinguishing HR from LR patients, we plotted ROC curves for each gene in each dataset (Figs. 4B–H). In the TCGA-CESC dataset, CANX had moderate diagnostic accuracy (AUC = 0.754, Fig. 4B), while HLA-DMB (AUC = 0.819, Fig. 4C) and HLA-DPB1 (AUC = 0.868, Fig. 4D) exhibited good discriminative ability. PSMB6 and PSMB7 had lower predictive value (AUC = 0.706 and 0.630, respectively, Figs. 4E, F). In the GSE29570 dataset, HLA-DMB (AUC = 0.820) and HLA-DPB1 (AUC = 0.823) again showed strong classification ability, while CANX (AUC = 0.697) had borderline utility, and PSMB6 and PSMB7 displayed poor accuracy (AUC = 0.494 and 0.638, respectively) (Fig. 4G). In the GSE63514 dataset, CANX and HLA-DMB demonstrated high diagnostic accuracy (AUC = 0.893 and 0.888, respectively), HLA-DPB1 showed moderate accuracy (AUC = 0.645), whereas PSMB6 and PSMB7 again exhibited low predictive performance (AUC = 0.546 and 0.485) (Fig. 4H). Overall, CANX, HLA-DMB, and HLA-DPB1 consistently showed differential expression and moderate-to-high classification power across datasets, suggesting their robust association with CC prognosis as part of the MHCRGs-based risk model.

Fig. 4.

Differential Expression and Diagnostic Evaluation of MHCRGs in CC Patients. A Expression comparisons of five MHCRGs between high- and low-risk groups in the TCGA-CESC dataset. B–F ROC curves for CANX, HLA-DMB, HLA-DPB1, PSMB6, and PSMB7 in the TCGA-CESC dataset. G Combined expression analysis and ROC curves of the five MHCRGs in the GSE29570 dataset. H Combined expression analysis and ROC curves of the five MHCRGs in the GSE63514 dataset

 FEA and PEA of MHCRGs

To investigate the biological significance of the five MHCRGs in CC, we conducted GO enrichment analysis (Supplementary Table 3). GO terms significantly enriched (adjusted P < 0.05, FDR < 0.05) were primarily related to MHC class II protein complex assembly, antigen processing and presentation, and proteasome-mediated pathways. These findings indicate that MHCRGs are closely associated with immune-related biological processes, particularly those involved in antigen presentation. Key enrichment results were visualized using bubble plots (Fig. 5A) and network diagrams illustrating biological processes, cellular components, and molecular functions (Figs. 5B–D). Subsequently, we performed KEGG PEA on the five MHCRGs (Supplementary Table 4). The KEGG results revealed significant enrichment in pathways including antigen processing and presentation, phagosome, asthma, human T-cell leukemia virus 1 infection, and allograft rejection. We illustrated the KEGG enrichment results using a bar chart (Fig. 5E) and a pathway network diagram (Fig. 5F), highlighting the involvement of MHCRGs in these key immune and disease-related pathways.

Fig. 5.

FEA and PEA of MHCRGs in CC. A GO enrichment results presented as a bubble plot. B–D Network diagrams displaying GO enrichment terms for biological processes (BP), cellular components (CC), and molecular functions (MF), respectively. E KEGG pathway enrichment visualized as a bar plot. F KEGG pathway-gene interaction network. In the bubble plot (A), the y-axis indicates GO terms, while bubble size and color represent gene count and adjusted P-value, respectively. In the network diagrams (B–D, F), red circles represent enriched GO or KEGG terms, and blue nodes represent MHCRGs. Only terms with adjusted P-value < 0.05 and FDR < 0.05 were retained

GSEA between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset

To evaluate the influence of gene expression differences between HR and LR cohorts defined by the MHCRGs PM on CESC within the TCGA-CESC dataset, GSEA was performed. This analysis examined the association between genome-wide gene expression and enriched BPs, CCs, and MFs. Significant enrichment was determined using adjusted P value < 0.05 and false discovery rate (FDR, q-value) < 0.25 as thresholds. GSEA results demonstrated that genes distinguishing the HR and LR cohorts were significantly enriched in multiple pathways, including KEGG TGF-beta signaling pathway (Figs. 6A–B), Reactome signaling by TGF-beta family members (Fig. 6C), Reactome glycolysis (Fig. 6D), KEGG nitrogen metabolism (Fig. 6E), WikiPathways omega-9 fatty acid synthesis (Fig. 6F), along with other relevant pathways (Fig. 6B and F, Supplementary Table 5). To further explore pathway activity differences between HR and LR groups, GSVA was conducted on 30 Hallmark gene sets (GSs) (Figs. 6G–H, Supplementary Table 6). The analysis revealed significant differential enrichment in multiple Hallmark pathways. Specifically, 16 pathways, including Hallmark Angiogenesis and Hallmark Cholesterol Homeostasis, showed higher enrichment scores in the HR cohort, whereas 14 pathways, such as Hallmark Allograft Rejection, Hallmark Apoptosis, and Hallmark Complement, were more enriched in the LR cohort. These differences were visualized via a heatmap generated using the R package pheatmap (Fig. 6G). The Mann-Whitney U test was applied to assess statistical differences between cohorts, and the results were summarized in a grouped comparison dot plot (GCD) (Fig. 6H), confirming significant variation in Hallmark pathway enrichment between the HR and LR groups.

Fig. 6.

GSEA and GSVA enrichment analyses between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset. A Top five biological attributes enriched in GSEA. B–F Enrichment plots for KEGG TGF-beta signaling (B), Reactome signaling by TGF-beta family members (C), Reactome glycolysis (D), KEGG nitrogen metabolism (E), and WikiPathways omega-9 fatty acid synthesis (F). G Heatmap of GSVA scores across 30 Hallmark gene sets comparing HR and LR cohorts. H Group comparison dot plot showing statistical differences in GSVA scores between HR and LR groups

Disparities in SsGSEA immune characteristics between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset

To investigate immune infiltration differences between the HR and LR cohorts defined by the MHCRGs PM in the TCGA-CESC dataset, we applied the single-sample Gene Set Enrichment Analysis (ssGSEA) algorithm to quantify the infiltration abundance of 28 immune cell (IC) types across samples from both cohorts. Differences in immune cell infiltration between HR and LR groups were assessed using the Mann-Whitney U test, and the results were visualized sssusing a GCD (Fig. 7A). The analysis revealed statistically significant differences in the infiltration levels of 22 immune cell types between the HR and LR cohorts, including: Activated B cells, Activated CD4 + T cells, Activated CD8 + T cells, Activated dendritic cells, CD56^bright NK cells, CD56^dim NK cells, Central memory CD4 + T cells, Effector memory CD4 + T cells, Effector memory CD8 + T cells, Eosinophils, Gamma delta T cells, Immature B cells, Macrophages, Mast cells, Myeloid-derived suppressor cells (MDSCs), Monocytes, Natural killer cells, Natural killer T cells, Regulatory T cells, T follicular helper cells, Type 1 T helper cells, and Type 17 T helper cells. Next, correlation analyses were performed to evaluate the relationships among the infiltration abundances of these 22 immune cell types within the LR and HR cohorts separately (Fig. 7B, C). In both cohorts, most immune cell types showed positive correlations with each other, indicating coordinated infiltration patterns. Furthermore, we investigated the associations between the infiltration levels of the 22 immune cells and the expression of the five MHCRGs in both LR (Fig. 7D) and HR (Fig. 7E) cohorts, visualized by correlation scatter plots. In the LR cohort, Central memory CD4 + T cells were positively correlated with HLA-DPB1 expression; Immature B cells positively correlated with HLA-DMB; Monocytes showed a negative correlation with PSMB7; Activated dendritic cells were negatively associated with PSMB6. In the HR cohort, stronger positive correlations were observed between Central memory CD4 + T cells and HLA-DPB1, as well as between Natural killer cells and both HLA-DMB and HLA-DPB1. Additionally, Immature B cells exhibited negative correlations with PSMB6 and PSMB7. These findings suggest distinct immune infiltration patterns and immune gene expression relationships between risk groups defined by MHCRGs PM in CESC.

Fig. 7.

Analysis of ssGSEA immune infiltration differences between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset. A Comparison of ssGSEA-based immune cell infiltration abundances between HR and LR cohorts. B–C Correlation heatmaps depicting the relationships among immune cell infiltration abundances in the LR (B) and HR (C) cohorts. D–E Heatmaps showing correlations between MHCRGs expression and immune cell infiltration in LR (D) and HR (E) cohorts

ICG analysis, HLA analysis, and immunotherapy evaluation between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset

We retrieved expression data for 12 immune checkpoint genes (BRAF, CD274, CD276, CTLA4, HAVCR2, HLA-G, IDO1, LAG3, PDCD1, TIGIT, TP53, VSIR) and 19 HLA family genes (HLA-A, HLA-B, HLA-C, HLA-DMA, HLA-DMB, HLA-DOA, HLA-DOB, HLA-DPA1, HLA-DPB1, HLA-DQA1, HLA-DQA2, HLA-DQB1, HLA-DQB2, HLA-DRA, HLA-DRB1, HLA-DRB5, HLA-E, HLA-F, HLA-G) from the TCGA-CESC dataset. Using the Mann-Whitney U test (Wilcoxon rank-sum test), we compared the expression levels of these genes between the HR and LR groups defined by the MHCRGs PM. Significant differences were observed in all 12 immune checkpoint genes (Fig. 8A) and all 19 HLA genes (Fig. 8B) between the two cohorts. To further assess tumor immunogenicity, we obtained Immunophenoscores (IPS) from the TCIA database and compared overall IPS as well as PD-1, CTLA-4, and combined PD-1/CTLA-4 scores between HR and LR groups. The Wilcoxon test revealed that the HR cohort had significantly lower IPS scores (P < 0.001, Fig. 8C), suggesting reduced immunogenicity and potentially poorer response to ICB therapy according to IPS metrics.

Fig. 8.

Immune checkpoint gene (ICG) and HLA family gene expression, immunotherapy evaluation, and tumor microenvironment analysis between HR and LR cohorts of MHCRGs PM in the TCGA-CESC dataset. A, B GCD plots showing differential expression of ICGs (A) and HLA genes (B). C Differences in immunogenicity scores (IPS, IPS-PD1, IPS-CTLA4, IPS-PD1-CTLA4) between HR and LR groups. D–E Comparison and correlation of StromalScore, ImmuneScore, ESTIMATEScore, and Tumor Purity between cohorts. F TIDE immunotherapy scores and MSI scores comparisons and their correlations with risk scores

We also applied the ESTIMATE algorithm to infer tumor microenvironment composition, calculating StromalScore, ImmuneScore, ESTIMATEScore, and Tumor Purity for each sample. Significant differences were found between HR and LR groups in StromalScore (P < 0.05), ImmuneScore (P < 0.001), ESTIMATEScore (P < 0.001), and Tumor Purity (P < 0.001) (Figs. 8D, E). Correlation analyses showed that the risk score was negatively correlated with StromalScore (R = − 0.25), ImmuneScore (R = − 0.63), and ESTIMATEScore (R = − 0.52), and positively correlated with Tumor Purity (R = 0.5), all with P < 0.001. Considering immunotherapy response prediction, we evaluated the Tumor Immune Dysfunction and Exclusion (TIDE) score. Interestingly, the HR group showed significantly lower TIDE scores compared to the LR group (P < 0.001, Fig. 8F), implying a better potential response to immune checkpoint inhibitors per TIDE assessment. Moreover, risk score negatively correlated with TIDE score (R = − 0.426, P < 0.001). Additionally, we compared MSI scores and found the HR cohort had higher MSI scores than the LR cohort (Fig. 8F), with a weak positive correlation between risk score and MSI (R = 0.167, P < 0.01).

DST analysis between HR and LR cohorts of the MHCRGs PM in the TCGA-CESC dataset

To explore the appropriate therapeutic strategy of mRNA vaccination for CC, we used the DST data from the GDSC database as the training set to predict the responsiveness of MHCRGs PM to common anticancer agents between HR and LR cohorts in the TCGA-CESC dataset. The Mann-Whitney U test (Wilcoxon rank sum test) was applied to evaluate differences in drug sensitivity between HR and LR cohorts. We identified the top 20 drugs exhibiting significant sensitivity differences: Roscovitine, Rapamycin, LFM.A13, Bortezomib, Sunitinib, Metformin, VX.680, Camptothecin, BMS.536,924, GW843682X, JW.7.52.1, S. trityl-L-cysteine, Paclitaxel, Mitomycin C, Vinblastine, CI.1040, RO.3306, EHT.1864, CGP.60,474, and GNF.2 (Fig. 9A−T). Generally, the LR cohort showed lower sensitivity to these drugs compared to the HR cohort. Based on the expression of five MHCRGs in the TCGA-CESC dataset, a ssGSEA was employed to compute the MHC score for each patient, representing the individual expression level of MHCRGs. Patients were then stratified into high and low MHC score cohorts using the median score as a cutoff. To validate the prognostic value of the MHC scores, univariate Cox regression (UCR) and multivariate Cox regression (MCR) analyses were performed to assess the association between MHC scores, clinical variables (Stage_T, Stage_N, Stage_M, Age), and overall prognosis of CC patients in the TCGA-CESC dataset. All variables significant in the univariate analysis were included in the multivariate model. Results showed that MHC scores were significantly associated with patient prognosis (Supplementary Table 7). The MCR results are summarized in a forest plot (Fig. 10A). We further constructed a RiskScore model incorporating clinical staging and MHC scores as follows: RiskScore=(StageT2 × 1.456)+(StageT3×−0.769)+(StageT4 × 0.904)+(StageN1 × 1.229)+(StageM0×−0.523)+(StageM1×−15.828)+(MHCScores×−0.915). This composite RiskScore integrates tumor size (Stage_T), lymph node status (Stage_N), metastasis (Stage_M), and MHCRGs expression, providing a comprehensive metric to stratify patient risk.

Fig. 9.

Drug Sensitivity Testing (DST) Analysis between HR and LR Cohorts of MHCRGs Prognostic Model in the TCGA-CESC Dataset. A–T illustrate the sensitivity differences of 20 anticancer drugs between HR and LR cohorts based on the GDSC database and MHCRGs prognostic model in TCGA-CESC. The drugs analyzed include Roscovitine (A), Rapamycin (B), LFM.A13 (C), Bortezomib (D), Sunitinib (E), Metformin (F), VX-680 (G), Camptothecin (H), BMS-536,924 (I), GW843682X (J), JW-7.52.1 (K), S-Trityl-L-cysteine (L), Paclitaxel (M), Mitomycin C (N), Vinblastine (O), CI-1040 (P), RO-3306 (Q), EHT-1864 (R), CGP-60,474 (S), and GNF-2 (T)

Fig. 10.

Construction of the Multivariate Cox Regression (MCR) Prognostic Model in the TCGA-CESC Dataset. A Forest plot displaying the hazard ratios of MHC Scores combined with clinical variables based on univariate and multivariate Cox regression analyses in the TCGA-CESC dataset. B Nomogram constructed from the MCR model integrating MHC Scores and clinical variables to predict overall survival probability. C–E Calibration curves for the nomogram model predicting 1-year (C), 3-year (D), and 5-year (E) survival, showing good agreement between predicted and observed outcomes. F–H Decision Curve Analysis (DCA) plots for the MCR model at 1-year (F), 3-year (G), and 5-year (H) follow-up, assessing clinical net benefit of the model

To improve the predictive value of our model, we developed a nomogram based on the MCR model (Fig. 10B). The nomogram assigns weighted points to each variable according to its influence in the MCR, with Stage_T and Stage_M having the greatest impact. Calibration curves for 1-, 3-, and 5-year survival probabilities showed good agreement between predicted and observed outcomes (Fig. 10C−E). DCA further demonstrated the clinical utility of the model, with the highest net benefit observed at the 5-year time point (Fig. 10F−H).

Experimental validation of MHCRGs expression and function in CC

To validate the biological relevance of the five MHCRGs identified through computational analysis, we performed a series of expression and functional experiments using clinical samples and in vitro models. Firstly, we collected CC tissues and matched adjacent normal tissues from clinical samples. Quantitative real-time PCR (qRT-PCR) analysis revealed that the mRNA expression levels of MHCRGs significantly differed between cancerous and adjacent tissues (n = 3, Fig. 11A). Similarly, Western blot analysis confirmed altered protein expression levels of MHCRGs in CC tissues compared with adjacent normal tissues (n = 5, Fig. 11B). Next, we focused on the functional validation of CANX, a key MHCRG. HeLa cells were transfected with CANX-targeting siRNA (si-CANX), and successful knockdown was confirmed at both the mRNA (qRT-PCR, Fig. 11C) and protein (Western blot, Fig. 11D) levels. To investigate the biological role of CANX in CC cell behavior, we conducted Transwell migration assays, which demonstrated that CANX knockdown significantly inhibited HeLa cell migration (Fig. 11E). In addition, apoptosis assays using Annexin V/PI staining demonstrated that CANX knockdown significantly decreased both early apoptotic (Annexin V⁺/PI⁻) and late apoptotic/necrotic (Annexin V⁺/PI⁺) cell populations compared with control (Fig. 11F–G), indicating that silencing CANX reduces apoptosis in HeLa cells. These findings provide experimental validation of MHCRG expression differences in clinical specimens and demonstrate the functional importance of CANX in regulating cervicals cancer cell migration and survival, thereby reinforcing the clinical and biological relevance of our MHCRG-based prognostic model.

Fig. 11.

Experimental validation of MHCRGs expression and function in CC. A qRT-PCR analysis showing differential mRNA expression levels of the five MHCRGs between CC tissues and matched adjacent normal tissues (n = 3). B Western blot analysis confirming altered protein expression levels of MHCRGs in CC tissues compared to adjacent normal tissues (n = 5). C, D Validation of CANX knockdown in HeLa cells by siRNA transfection, demonstrated by qRT-PCR (C) and Western blot (D). E Transwell migration assay showing significant inhibition of HeLa cell migration following CANX silencing. F, G Apoptosis analysis by Annexin V/PI staining indicating reduced apoptosis in HeLa cells after CANX knockdown. Data are presented as mean ± SD. Statistical significance: *P < 0.05, **P < 0.01, ***P < 0.001. These results validate the differential expression of MHCRGs in clinical samples and demonstrate the functional role of CANX in CC cell migration and apoptosis

Discussion

CC remains a significant global health issue, with genetic and immunological factors playing critical roles in its pathogenesis and prognosis. The aim of this research was to shed light on the genetic features, immune infiltration, DST, and PMs of CC using bioinformatics approaches. Key findings include the identification of TAP2 as the most frequently mutated gene in CC, predominantly through missense mutations. A PM featuring five MHCRGs was established as markedly correlated with patient outcomes. FEA highlighted their involvement in MHC class II protein complex assembly processes. GSEA revealed significant pathway enrichments, such as the TGF-beta signaling pathway between risk cohorts. Additionally, differential ICI across risk cohorts was identified in 22 IC types. Our comprehensive bioinformatics analysis offers fresh perspectives on the genetic landscape and immune context of CC.

TAP2, or Transporter 2, ATP-binding cassette sub-family B member, serves a crucial function in the presentation of endogenous peptides to the immune system by translocating them into the lumen of the endoplasmic reticulum for loading onto MHC class I molecules [44]. The high mutation frequency observed in TAP2 within our study suggests its potential involvement in CC pathogenesis. Missense mutations, being the predominant type identified, might lead to alterations in protein function that could disrupt this antigen processing pathway and consequently affect tumor immunogenicity. This disruption might contribute to immune evasion by cancer cells, highlighting the importance of TAP2 as a candidate for further investigation regarding its impact on disease progression and response to immunotherapy. CANX, alternatively referred to as Calnexin, functions as a chaperone that plays a crucial role in facilitating correct folding and maintaining quality assurance of freshly produced glycoproteins inside the endoplasmic reticulum [45]. Its identification as a prognostically significant gene in our LASSO regression analysis underscores its potential role beyond protein-folding; it might influence tumor cell survival and proliferation through interactions with other cellular pathways relevant to cancer biology. Elevated calnexin levels positively correlate with overall survival in CRC patients, suggesting a potential immunotherapeutic target [46]. CANX is also an immune-autophagy-related gene with independent prognostic values in kidney renal clear cell carcinoma [47]. HLA-DMB serves a function in AP by facilitating the removal of an invariant chain fragment from the peptide-binding site and maintaining unoccupied MHC class II molecules in a conformation receptive to peptides [48]. Our findings suggest that HLA-DMB’s expression correlates with CC prognosis. HLA-DPB1, a component of the major histocompatibility complex II (MHCII), is integral in facilitating AP by macrophages and dendritic cells [49]. Variations within HLA-DPB1 have been associated with differential immune responses to various diseases, including cancers [50]. Its presence as a key gene within our PM indicates that it might affect CC prognosis through its pivotal role in initiating adaptive immunity. The overexpression of HLA-DPB1 in breast cancer correlates with a better prognosis, promotes ICI, and suggests that HLA-DPB1 with immune checkpoint inhibitors could optimize breast cancer immunotherapy [51]. PSMB6 and PSMB7 are components of the immunoproteasome, which is integral for generating peptides for presentation by MHC class I molecules. They are involved in proteolytic degradation during antigen processing [52]. The expression of PSMB6 is elevated in various types of cancer [53]. Down-regulation of proteasoma-related genes, especially Psmb4 and Psmb6, promotes HCC progression. In summary, these genes collectively underscore an intricate relationship between antigen processing/presentation machinery and CC prognosis. Their contributions towards modulating both intrinsic cellular processes and extrinsic immune surveillance present them as valuable targets for further research into therapeutic interventions aimed at enhancing anti-tumor immunity.

The FEA of MHCRGs in CC suggested a marked association with BPs like MHC class II protein complex assembly, which is pivotal for antigen processing and presentation. The enrichment within this process suggests that alterations in MHCRGs could impact the efficiency of AP to T cells, thereby influencing the anti-tumor immune response and patient prognosis. Moreover, our GSEA results indicated a distinct enrichment of pathways, encompassing the KEGG TGF-beta signaling pathway between HR and LR cohorts defined by the MHCRGs PM. The TGF-β signaling cascade is recognized for its dual function in cancer, operating as a tumor inhibitor during initial phases but facilitating tumor advancement and metastasis in subsequent stages [54]. Its significant enrichment among different prognostic cohorts might reflect its involvement in CC pathogenesis and could provide insights into potential therapeutic targets. These findings highlight critical pathways and cellular processes that might contribute to CC progression and patient survival. Understanding these mechanisms is essential for developing novel therapeutic approaches aimed at modulating immune responses and enhancing medical results for individuals with CC.

In our study, the ssGSEA immune profiling unveiled notable disparities in the infiltration abundance of 22 types of ICs between HR and LR cohorts in the TCGA-CESC dataset. This finding underscores the complexity of the TME and its profound influence on patient outcomes. ICI within the TME has been acknowledged as a crucial factor in cancer advancement and therapeutic response [55]. For instance, cytotoxic T lymphocytes are pivotal for anti-tumor immunity, while regulatory T cells could suppress immune responses, potentially facilitating tumor escape [56]. MHC Scores in the TCGA-CESC dataset showed marked clinical relevance with CC prognosis, as demonstrated by UCR and MCR analyses. However, it is important to note a discrepancy in our immunotherapy response predictions when comparing different scoring systems. While the TIDE analysis indicated that the HR cohort might respond better to immune checkpoint blockade therapy, as evidenced by significantly lower TIDE scores, the IPS analysis revealed lower IPS-PD1, IPS-CTLA4, and combined IPS scores in the HR group, suggesting reduced immunogenicity and potentially poorer response. This contradiction likely stems from the distinct biological mechanisms and computational frameworks underlying TIDE and IPS algorithms. TIDE primarily assesses tumor immune evasion by modeling T cell dysfunction and exclusion, whereas IPS estimates immunogenicity based on expression profiles of immune-related genes. These differences highlight the complexity of predicting immunotherapy response and suggest that integrating multiple predictive metrics, alongside clinical validation, is essential for accurate assessment. Future studies combining these approaches with clinical trial data will be crucial to resolve these discrepancies and improve personalized immunotherapy strategies for cervical cancer patients.

Despite the comprehensive nature of our study, it is crucial to recognize its constraints. Notably, the predictive power of our MHCRGs-based prognostic model showed moderate accuracy (AUC values around 0.66–0.68), which limits its immediate clinical application. To better contextualized our findings, we compared the model’s performance with existing prognostic models of CC reported in the literature. Our model’s predictive accuracy is comparable to some published models but still leaves room for improvement. Future research will aim to incorporate additional molecular markers and clinical features to enhance predictive performance and clinical utility. Moreover, the lack of experimental validation hinders the direct biological interpretation of our bioinformatics predictions. Secondly, while the sample size from the TCGA-CESC dataset is considerable, it might not fully represent the global heterogeneity of CC populations. Thirdly, the lack of clinical verification studies indicates that our PMs and DST predictions have not been tested in real-world scenarios, potentially impacting their translational utility. Moreover, integrating data from multiple datasets could introduce batch effects that might bias results and interpretations. Addressing these limitations in future research endeavors will be vital for improving the reliability and clinical significance of our findings.

In conclusion, this research effectively established and confirmed a prognostic model grounded in MHCRGs for CC-identified DEGs between HR and LR cohorts, analyzed ICI patterns, estimated immune activity, predicted ICB treatment responses, and explored drug sensitivities. Further research and validation studies are needed to translate these insights into clinical benefits for CC patients.

Abbreviations

CC

Cervical cancer

MHCRGs

Major histocompatibility complex-related genes

PM

Prognostic model

FEA

Functional enrichment analysis

GSEA

Gene set enrichment analysis

GDSC

Genomics of Drug Sensitivity in Cancer

TGF-β

Transforming growth factor beta

ICI

Immune cell infiltration

ICG

Immune checkpoint gene

TME

Tumor microenvironment

ssGSEA

Single-sample gene set enrichment analysis

TCGA-CESC

The cancer genome atlas-cervical squamous cell carcinoma and endocervical adenocarcinoma

LASSO

Least absolute shrinkage and selection operator

AUC

Area under the curve

ROC

Receiver operating characteristic

DEGs

Differentially expressed genes

HPV

Human papillomavirus

DST

Drug sensitivity test

DCA

Decision curve analysis

PMs

Prognostic markers

MHC

Major histocompatibility complex

MSI

Microsatellite instability

AP

Antigen presentation

ICB

Immune checkpoint blockade

TIDE

Tumor immune dysfunction and exclusion

IPS

Immunophenoscore

CRC

Colorectal cancer

HCC

Hepatocellular carcinoma

CIN

Cervical intraepithelial neoplasia

CNV

Copy Number Variations

GEO

Gene expression omnibus

SCC

Squamous cell carcinoma

SNP

Single nucleotide polymorphisms

SNV

Single nucleotide variant

LSIL

Low-grade squamous intraepithelial lesion

HSIL

High-grade squamous intraepithelial lesion

OS

Overall survival

Author contributions

MLX and XQW conceptualized the study design and prepared the initial draft. MLX composed the manuscript. AMR conducted literature searches and gathered relevant data. XX performed data analysis. JRZ revised the manuscript. All contributors participated in drafting the manuscript and approved the ultimate submitted draft.

Funding

This work was supported by the National Natural Science Foundation of China (82203101).

Data availability

The datasets generated and analyzed during the current study are publicly available. The TCGA-CESC dataset is accessible through The Cancer Genome Atlas (https://portal.gdc.cancer.gov/). The GEO datasets GSE29570 and GSE63514 are available from the NCBI Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo/). All other data supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All patients provided written informed consent prior to sample collection. The study involving human participants was conducted in accordance with the Declaration of Helsinki. Ethical approval was obtained from the Ethics Committees of Zhongshan Hospital and Fudan University Cancer Hospital (approval number: NCT02562729).

Competing interests

The authors declare no competing interests.