Cervical cancer transcriptomic data, mutation data, and clinical data were retrieved from the TCGA and GEO databases. A total of 299 samples from the TCGA database were included for the analysis of differential gene expression, with matched clinical information. For the construction of the prognostic index, 283 cervical cancer samples were used (16 samples without clinical staging were excluded). The validation set comprised 55 cervical cancer samples from the GSE52904 dataset in the GEO database. Samples lacking matched clinical information in both the TCGA and GEO databases were excluded from the analysis.

Mitochondrial energy metabolism-related genes differentially expressed by DESeq2 were downloaded from the Human.MitoCarta3.0 database. The TCGA data was processed using the easyTCGA R package, and the mutation data was analyzed using the maftools R package to visually assess the mutation frequency of genes in patients. The FDR value ≤ 0.05 was used as the differential gene screening criterion. Mitochondrial metabolism-related differential genes were obtained by intersecting genes obtained by Human.MitoCarta3.0 with differential genes.

Differential gene enrichment analysis was conducted with the R package clusterProfiler to explore the variance in gene enrichment between healthy cervix and cervical cancer. A statistically significant P-value threshold of 0.05 was applied to identify gene sets with less than or equal to 5 genes, and those with more than 5,000 genes.

The differential genes related to mitochondrial metabolism were selected for 10-fold cross-validation using Lasso regression. The lambda value with the smallest mean error, determined from the cross-validation results, was used to select the genes to be included in the prognostic index. The prognostic index was constructed by multiplying the included genes with their regression coefficients and summing them up. Following this, the individuals were categorized into groups with high and low indexes according to the average expression level. Survival analysis was conducted to assess the prognostic significance of the prognostic index. Additionally, the significance of the prognostic index was validated using the GEO database. Furthermore, multifactorial Prognostic indexes were investigated using Cox regression analysis for cervical cancer patients to confirm their independence as prognostic factors.

Utilizing the R software CIBERSORT, we conducted an examination of the relationship between immune cell penetration and immune activity. This assessment enabled us to ascertain the proportion of 22 distinct immune cell categories in the TCGA dataset for individuals with normal health and those diagnosed with cervical cancer, in addition to those classified in high and low index categories. Additionally, we conducted Gene Set Enrichment Analysis using sets of genes related to the immune system to assess differences in immune functions among groups with high and low indices. Furthermore, we compared the activity of immune functions between the two risk groups. To evaluate the immune stroma and immune microenvironment, we calculated immune stroma and immune microenvironment scores using the R package ESTIMATE.

This study analyzed the tumor mutation load (TMB) to evaluate the mutation rate of samples by utilizing mutation data from cervical cancer patients in the TCGA database. We specifically analyzed the mutation differences between the high index and low index groups. Additionally, we categorized individuals with cervical cancer into two cohorts according to the median TMB in order to assess the correlation between the predictive measure of mutation burden and its influence on immunotherapy within the high and low index categories.

The Tumor Immune Dysfunction and Exclusion (TIDE) platform was utilized in cervical cancer patients to forecast the response to immunotherapy directed at PD-1 and CTLA4. The TIDE prediction score was employed to evaluate the likelihood of response to immunotherapy, with a negative correlation indicating a poorer response. In order to delve deeper into the variations in immunotherapy response, we analyzed the TIDE scores of patients in both high-risk and low-risk categories.

Using the R package oncoPredict, drug sensitivity for cervical cancer patients in the TCGA database was predicted in this research. Each patient’s sensitivity score was determined by utilizing the calcPhenotype function. We evaluated the differences in drug sensitivity for various medications by comparing sensitivity scores of the high and low index groups.

Survival analyses were conducted using the R package Survival. To investigate the differential gene expression changes across different clinical stages of cervical cancer, we obtained data from the GEPIA2 online website. Additionally, gene expression data was obtained through a search in the HPA database.

IHC staining was conducted after tissue slides were deparaffinized in xylene and rehydrated in alcohol, followed by 3%H2O2 treatment to inactivate endogenous peroxidase. A microwave-based antigen repair procedure was then performed in 0.1M sodium citrate buffer at pH 6.0. After blocking sections with 5% normal goat serum for 30 minutes at room temperature, primary antibodies were then left to incubate overnight at 4°C. For nuclei visualization, poly-HRP secondary antibody was incubated for 60 minutes at room temperature in the dark, followed by counterstaining with hematoxylin. Our hospital’s pathologists analyzed the staining results from images acquired with an Caseviewer scanner. ISCU expression was scored based on staining intensity as follows: 0 (negative), 1 (weak), 2 (moderate), and 3 (strong). The positive proportion area ranged from 1 (0–25%) to 4 (76%-100%). Scores of 0–12 were defined as ISCU expression results. This research project has obtained approval from the appropriate ethics committee or institution and is being conducted in strict adherence to ethical guidelines. The study upholds the rights and privacy of participants, ensuring the confidentiality of their personal information (KY20222112-C-1).

Inclusion criteria: 1) Patients with cervical cancer who have not received any anti-tumor treatment prior to surgery; 2) All cervical cancer patients have been confirmed by pathological diagnosis, and normal cervical tissues have been confirmed by pathologists to be free of any benign or malignant tumors; 3) Complete clinical and pathological data are available. Pathological diagnoses have been confirmed by two or more senior pathologists.

Exclusion criteria: 1) Patients with a history of previous gynecological tumor surgery; 2) Patients with recurrent cervical squamous cell carcinoma; 3) Patients with concurrent infectious diseases, rheumatic autoimmune diseases, cardiovascular and cerebrovascular diseases, and other major malignant tumors.

The data was analyzed using R version 4.2.2. Statistical differences between the two groups were compared using a two-tailed Student’s t-test. Survival analysis was conducted using the Kaplan-Meier estimate and log-rank test. Statistical significance was established with a p-value below 0.05 (p<0.05).

Data from the TCGA database was collected for cervical cancer patients and normal cervical tissues, including transcriptomic and clinical information. By acquiring the distinct genes and comparing them with genes related to mitochondrial metabolism, a sum of 1,123 differential genes related to metabolism were discovered through the utilization of the DESeq2 software. Among these genes, 754 were up-regulated, and 369 were down-regulated. The expression of the differential genes was visualized in Supplementary Figure S1A using a volcano plot. Additionally, Supplementary Figure S1C illustrated the relationship between the 50 genes with the most significant differences and clinical traits. The analysis revealed that the metabolic alterations in cervical cancer were closely associated with HPV infection status, tumor pathology type, and clinical stage. These findings suggest that metabolic gene alterations may have implications for the prognosis of cervical cancer patients.

The GO enrichment analysis showed that the differentially expressed genes were mainly enriched in the progression of cellular tissues and differentiation of cells. Cell structure enrichment was mainly seen in chromosomes, with molecular function enrichment in signal transduction and cell activation ( Figure 1A ). Analysis of KEGG enrichment revealed that the genes showing differences were primarily associated with interactions between cytokines and cells, binding of neural ligands to receptors, and the PI3K-Akt signaling pathway, among other pathways ( Figure 1B ).

To comprehensively explore the functions of the differential genes and avoid the exclusion of genes with insignificant differences by traditional enrichment algorithms, GSEA enrichment analysis was conducted. The analysis revealed the involvement of differential genes in cell cycle-related functions ( Supplementary Figure S1B ). Metabolism-related genes could potentially influence the outlook of individuals with cervical cancer, paving the way for the creation of a prognostic tool.

Mutation frequency analysis indicated that mitochondrial metabolism-related differential genes exhibited high mutation frequencies. The top 10 genes ranked by differences displayed mutation frequencies above 12%, while the mutation frequencies of the top 20 genes ranked by differences were above 8% ( Figure 1C ). Analyses of enrichment and mutation revealed that genetic mutations in metabolism-related genes could impact physiological functions like tumor metabolism, signal transduction, and tissue progression, ultimately affecting tumor progression, metastasis, and other biological processes.

Based on the aforementioned analysis, we propose a reasonable hypothesis that the survival time of cervical cancer patients can be predicted by constructing a metabolism-related gene index. To validate this hypothesis, we utilized a training set comprising 283 cases of TCGA cervical cancer patients with complete clinical information, and an external validation set consisting of 55 cases of cervical cancer from the GEO database (GSE52904) with complete clinical information. The prognostic index’s effectiveness was evaluated using the validation set.

In order to enhance the clinical value of the index, clinical indicators were also taken into consideration during the construction of the prognostic index. A predictive score consisting of six factors was created using LASSO Cox regression analysis ( Figures 2A, B ). The prognostic index for every individual was forecasted by utilizing the equation established from the LASSO model: Prognostic Index = (Figo*0.205824) -(COX5A*0.000026) -(ISCU *0.000331) + (ACACA*0.003564) + (AIFM3*0.024294) + (PDF*0.273327). The predictive performance of the prognostic index was effectively demonstrated through column line graphs, with an AUC of 0.81 ( Figure 2C ). Evaluation metrics such as the Clinical Impact Curve, DCA Decision Curve, and Calibration Curve indicated that the prognostic model exhibited favorable performance ( Supplementary Figures S2B–D ). Single-gene survival analysis demonstrated significant differences in survival for all prognostic index genes, except for AIFM3 ( Supplementary Figure S3 ).

To further validate the prognostic index, A high index sample and a low index sample were selected from the training set. Survival rates were graphed, showing notable variances in survival rates between the groups with high and low indices ( Figure 3A ). This significant survival difference was also observed in the validation set ( Figure 3B ). Multivariate Cox regression analyses was performed to assess whether the prognostic index could function as an independent prognostic factor, revealing that the clinical data and prognostic index were indeed independent prognostic factors (P=0.045) ( Table 1 ). In the training cohort, we observed that the prognostic index exhibited an AUC of 0.81, 0.70, and 0.74 at 1-, 3-, and 5-year intervals respectively ( Figure 3C ). While the validation set demonstrated an AUC value of 0.80 for a duration of one year ( Figure 3D ), it was not feasible to calculate AUC values for three and five years due to limited data collection time in this set. Furthermore, our findings indicate that the prognostic index effectively distinguishes between populations with varying survival outcomes, thereby reinforcing the significance of our research ( Figures 3E, F ). However, since the dataset with longer survival times lacked essential clinical information (such as FIGO), it could not fulfill our requirements; hence no AUC values were available for longer durations in the validation set.

Approval has been granted for the use of immunotherapy in treating late-stage cervical cancer, but there are still obstacles to overcome. These include determining the best timing for immunotherapy, choosing the right combination of drugs, and addressing resistance to immunotherapy. To address these issues, we investigated immunization-related aspects of cervical cancer.

Through immunological analyses of surviving and deceased cervical cancer patients, we observed significant differences in immune scores, immune microenvironment, and survival outcomes ( Supplementary Figures S4A–C ). These immune phenomena align with current studies highlighting the significant impact of immunotherapy on cervical cancer prognosis. Building upon these findings, we postulate a potential correlation between the mitochondrial metabolic prognostic index and immunotherapy in cervical cancer.

Immunoinfiltration analysis demonstrated significant differences in memory B lymphocytes between normal samples and cervical cancer ( Figure 4A ), while the high and low index groups exhibited substantial disparities in plasma cells. Significant variations were noted between the high and low index groups in relation to immune cells linked to immunotherapy, such as plasma cells and M1 macrophages ( Figure 4B ), prompting speculation that the metabolic index may predict the effectiveness of immunotherapy in patients with cervical cancer.

Several immune cells from normal and cervical cancer samples, as well as from high- and low-index groups, were analyzed to test this hypothesis further. Surprisingly, the groups with high and low indexes showed notable variations in nearly all immune cells, exceeding the impacts seen in both the control and cervical cancer groups ( Figures 4C, D ). Furthermore, the analysis of immune processes revealed significant disparities not only in the proportions of immune cells but also in the underlying immune mechanisms, encompassing nearly all immune processes ( Supplementary Figure S5 ). These findings provide compelling evidence supporting our initial suspicion that the metabolic difference index holds predictive value for the immunotherapeutic efficacy of cervical cancer patients.

Immunoassays have shown that metabolic markers can predict immune cell levels and immune responses in individuals with cervical cancer. Immunosuppressive drugs have been widely used to treat advanced and recurrent cervical cancer, with the goal of determining if the prognostic index can impact outcomes through the immunosuppressive pathway. Gene correlation analysis revealed that among the prognostic index genes, only ISCU exhibited a significant correlation with immunosuppressant genes ( Figure 5 , Supplementary Figures S6 , S7 ). Based on this finding, we hypothesize that ISCU may play a pivotal role in the immune-predictive capacity.

Tumor mutation load (TMB) is closely associated with immunosuppressant response, with a more pronounced effect of inhibitor treatment observed in cases with higher mutation loads. We sorted the samples into high and low mutation groups by determining the median mutation index of each sample in order to study the relationship between index genes and TMB. The findings revealed that 41.4% of the high mutation group exhibited a high index, whereas 68.2% of the low mutation group displayed a low index ( Figure 6F ). We grouped the samples into four categories and performed survival analysis based on the prognostic index and tumor mutational burden (TMB) results. The results of our study showed that patients with a low index and high mutational burden had the best prognosis, while those with a high index and low mutation burden had the worst outcome (see Figure 6E ). These results further support our hypothesis that prognostic scores can serve as predictors of immunotherapy response in patients.

Our model’s predictive outcomes shed light on the reasons why certain patients with a high tumor burden experience poor immunotherapy efficacy. Furthermore, our results suggest that a significant proportion of patients in the low tumor load group have the potential to benefit from immunotherapy.

Immunosuppressant therapy has demonstrated remarkable therapeutic efficacy in select patients with cervical cancer. The TIDE score is employed to assess the probability of tumor immune escape by analyzing the gene expression patterns in tumor specimens. To assess the predictive role of prognostic index-related genes in immunotherapy response, we conducted an analysis to examine the correlation between prognostic scores and TIDE scores.

The examination findings showed notable variances in TIDE rating and exclusion among the high and low index categories. Notably, the high index group exhibited a lower TIDE score, suggesting a potential for improved therapeutic efficacy with immune checkpoint inhibitors ( Figures 6A–C ). Patients were categorized into high TIDE and low TIDE groups using the median value obtained from the TIDE analysis. During the survival analysis, it was discovered that patients with a low predictive score and low TIDE had better results than those with a high score and low TIDE when examined collectively (all patients in the high score group were part of the low TIDE group). The findings suggest that the prognostic score may play a more significant role in forecasting the prognostic outcome ( Figure 6D ).

Additionally, we investigated the correlation between risk assessments and drug responsiveness. Drug sensitivity scores were derived from the gene expression data of cervical cancer patients using the calcPhenotype function within the oncoPredict package. The low-risk group showed more sensitivity to various chemotherapeutic agents, such as IRAK4, Dihydrorotenone, Gallibiscoquinazole, OF-1, JQ1, and more, when compared to the high-risk group ( Supplementary Figure S8 ).

We used the GEPIA database to examine the connection between prognostic index-associated genes and clinical stage. In advanced cervical cancers, ISCU expression was found to be decreased, whereas in early cervical cancers it was observed to be increased according to our analysis. Our observation prompted us to consider the possibility that the ISCU gene could have a protective function in the advancement of cervical cancer ( Figure 6G , Supplementary Figure S8 ).

To evaluate the trustworthiness of the prognostic index model, 40 samples of cervical cancer tissues and 10 samples of normal cervical tissues (obtained from patients who had hysterectomy for non-malignant conditions) were acquired from the Department of Gynecology at XiJing Hospital. Immunohistochemical analysis showed a connection between ICSU protein expression and RNA levels, with the highest expression in healthy tissues decreasing as the stage advanced ( Figure 7 , Table 2 ). Cervical tissues were only available for patients with stages I-III, so the expression in stage IV patients remains unknown. Based on our analysis, we speculate that ISCU could be a useful indicator for predicting outcomes.