Workflow and methods

The workflow is illustrated in Fig. 1. For the training cohort, we identified 340 patients with SOC from the TCGA; for validation cohorts, we identified 79 patients from GSE26193, 70 patients from GSE63885, and 276 patients from GSE140082. Furthermore, we collected 1254 PCD-related genes from previous reports (Tang et al. 2019; Zou et al. 2022) to select the genes needed to build the model. In addition, we conducted gene mutation analysis in 338 patient samples from TCGA-GTEx and ultimately identified 779 differentially expressed PCD-related genes. For the single-cell RNA transcription dataset, we collected seven and one OC samples from GSE184880 (Xu et al. 2022) and the scRNA-seq library of Ren et al. (Ren et al. 2022), respectively. Finally, we screened 14 PCD genes, including 5/12 cell death patterns, via the nomogram model to predict patient prognosis. In addition, we analyzed the association between the CDI and the TME, drug sensitivity, and single-cell sequencing results.

Fig. 1

Flowchart for comprehensive analysis of diverse PCD patterns in patients with SOC

Different expression analysis of PCD-related genes in patients with SOC

We identified PCD-related DEGs in the TCGA-GTEx cohort and analyzed their functions to explore the potential impact of PCD on SOC. First, we observed the variation landscape of DEGs in SOC tissues, identifying 526 upregulated and 253 downregulated genes. We used a heatmap to detect the RNA expression of PCD-related DEGs, with red and blue representing upregulated and downregulated expression levels, respectively (Fig. 2Fig. 2B). Additionally, we analyzed the chromosomal locations, changes in expression, and correlations of the PCD-related DEGs (Fig. 2C). In addition, KEGG enrichment analysis showed that the most notable pathways with enriched DEGs were mainly associated with cell death regulation, including lysosome and necroptosis, with a high number of enriched genes and significant p-values (Fig. 2D). The GO analysis confirmed these findings, indicating that the DEGs were significantly concentrated in the regulation of the apoptotic signaling pathway, the intrinsic apoptotic signaling pathway, and autophagy (Fig. 2E). We also specifically analyzed PCD-related gene variations. The results showed that 97.32% (218/224) of the patients with SOC had mutations; the top 20 mutations of PCD-related genes were displayed, of which TP53 was the gene with the highest frequency of mutations (95%), and the remaining 19 genes had mutation frequencies ranging from 2–4%; missense mutation was the main variant classification, and single nucleotide polymorphism was the main variant type (Fig. 2F and G). The copy number variation (CNV) status analysis results showed that PCD-related genes in patients with SOC had frequent copy number changes, which may help cancer cells escape PCD (Fig. 2H). The above results show the variation landscape of the PCD-related genes in patients with SOC, reveal the biological functions of PCD-related DEGs in SOC processes, and suggest a potential link between PCD and SOC.

Fig. 2

Variant landscape of PCD-related genes in patients with SOC. A Heatmap of PCD-related differentially expressed genes (DEGs) between SOC and normal tissues. B Volcano plot of PCD-related DEGs. Points with labels were part of the obvious DEGs with an adjusted P-value < 0.001 and |log2FC|> 4. C Circos plot of location, fold-change (FC), expression level, and significant correlation with the PCD-related DEGs. D Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of DEGs. E Gene Ontology (GO) enrichment analysis of the DEGs. F Oncoplot of the top 20 most frequently mutated PCD-related genes in the TCGA-OV cohort. G Summary of somatic mutations in the PCD-related genes in the TCGA-OV cohort. H Top 100 PCD-related genes with the most significant copy number variations

Construction and validation of predictive model based on PCD-related genes

We used Lasso regression and Cox analysis to screen the PCD-related DEGs further and build a prognosis gene model to explore the impact of PCD on the prognosis of patients with SOC and to detect the best survival-related genes. We identified the best punishment coefficient among the 272 highly survival-related genes (lambda.min = 0.07488). We narrowed the gene range according to the coefficient, eventually retaining 14 genes (RB1, UBB, TRIM27, FANCD2, CD38, LEPR, CAAP1, FADD, SNCA, BLOC1S1, AP1S2, SLC7A11, DDIT4, and TREM2) for predictive model construction, six being derived from apoptosis, four from autophagy, two from ferroptosis, one from necroptosis, and two from lysosome-dependent cell death (FADD belongs to both apoptosis and necroptosis) (Fig. 3A and B). Pearson's correlation analysis showed the correlation between each model gene (Fig. S1), and KM analysis showed the influence of the expression level of each model gene on the survival rate of patients (Fig. S2). The Wilcoxon test was used to compare the expression differences of each model gene between SOC and normal tissues (Fig. S3A). After testing the effectiveness of the included genes, we calculated the CDI using the following formula:

$$CDIscore = 0.16220*RB1 \,expression - 0.17038 * UBB \,\,expression - 0.41386 * TRIM27 \,\,expression + 0.26017 * FANCD2 \,\,expression + 0.29119 * CD38\,\,expression + 0.25412 * LEP\,}xpression - 0.40824 * CAAP1 \,\,expression + 0.43810 * FADD\,\,expression + 0.20286 * SNCA\,\,expression - 0.22865 * BLOC1S1\,\,expression - 0.26146 * AP1S2\,\,expression - 0.27573 * SLC7A11 \,\,expression + 0.12693 * DDIT4 \,\,expression + 0.09397 * TREM2 \,\,expression$$

Fig. 3

Construction of the PCD-related gene signature for patients with SOC. A LASSO coefficient profiles of 272 PCD genes. B Cross-validation of the gene signature. C Violin plot showing the relationship between the cell death index (CDI) and survival status. D Heatmap of the clinical features and model gene expression in the TCGA-OV cohort. The age cutoff was the median patient age. E The mutation frequency of the top 20 mutable genes between high and low-CDI groups in SOC patients in TCGA-OV

Next, we performed multiple analyses to reveal the relationship between these 14 genes and SOC. We used the TCGA cohort as the training set, calculated the CDI of each patient with SOC, and divided the patients into high and low CDI groups with the median as the boundary. Furthermore, we generated a violin diagram that indicated that the CDI of patients in the death group was significantly higher than that in the survival group to confirm the relationship between the CDI and the survival status of patients. Therefore, suggesting that a high CDI indicates a poor prognosis (Fig. 3C). The heat map further demonstrates the relationship between the CDI and age, survival status, grade, stage, and 14 model genes from the perspective of clinical characteristics. The CDI was significantly correlated with the survival status of the patients; the mortality of patients in the high CDI group was significantly higher than that of the patients in the low CDI group (Fig. 3D). However, the CDI was not associated with patient’s stage, grade, or age (Fig. S3B). These results suggested that CDI may be an independent factor in the prognosis of patients with SOC.

Next, we explored the relationship between CDI and prognosis from the perspective of genetics. The stacked bar chart shows the differences between the top 20 mutated genes of the high and low CDI groups (Fig. 3E). The chart showed that the mutation rate of AHNAK was higher in the low CDI group than in the high CDI group, indicating that the low CDI group has a better prognosis (Ghodke et al. 2021) and that immunotherapy may exert a better effect on the low CDI group (Zhao et al. 2021b). These results indicated that a higher CDI score may associated with a worse prognosis. We successfully constructed a CDI related to patient prognosis and preliminarily validated the predicted model.

Internal training and external validation of prognostic implications of CDI

We conducted internal training and external validation using the three cohorts, TCGA-OV, GSE63885 + GSE26193, and GSE140082. First, we compared the differences between the survival statuses of the high and low CDI groups, which showed significantly lower overall survival (OS) in the patients with a high CDI than in those with a low CDI (Fig. 4A). The PCA results also confirmed significant differences between patients with high and low CDIs, suggesting that CDI may be a good indicator of differentiation (Fig. 4B). Next, we described the relationship between the patient CDI and survival rate using KM analysis, which also showed that the patients in the high CDI group had a shorter OS in general, and that, for the same OS, the survival rate of patients in the high CDI group was significantly lower than that of the patients in the low CDI group (Fig. 4C). Additionally, we used a ROC analysis to validate the sensitivity and specificity of the predictive model to facilitate clinical applications. In the TCGA-OV and GSE26193 + GSE63885, we selected 1 to 9-year survival to validate prognostic performance of the model, and in GSE140082, limited by sample follow-up time, we selected 1 and 2-year survival time points to validate the model's performance (Fig. 4D and Fig. S4A). Our results showed that CDI had the best performance in predicting training cohort survival, as well as competent performance in the validation cohorts. Besides, we further selected the GSE53963 cohort and performed ROC analysis; the results further showed that our model has good predictive performance and reliability (Fig. S4B). To better show the predictive ability of our model, we have also included a comparison of our model with other published signatures, which also illustrates our model has robust predictive ability and performs better than many published biomarkers (Table S3). In all, these results indicated that CDI shows important prognostic significance for patients with SOC and that patients with a high CDI tend to have a worse OS, suggesting that the CDI may be closely associated with other tumor characteristics.

Fig. 4

Prediction model performance evaluation. A Distribution of survival status and time according to the normalized CDI in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts. Dashed lines represent the dividing lines of the median number. B The principal component analysis plot based on the CDI in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts. C Kaplan–Meier (KM) survival curve for the overall survival (OS) of the low and high CDI group patients in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts. D Receiver operating characteristic (ROC) analysis of the model in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts (Only the 3-, 5-, and 7-year survival is shown, and the rest is detailed in supplementary materials)

Establishment and assessment of the nomogram survival model

We performed univariate and multivariate Cox regression analyses on the training set to clarify whether the CDI is an independent prognostic indicator of SOC. Univariate Cox regression analysis indicated that the CDI was significantly associated with OS in patients with SOC (HR = 3.07, 95%: 2.28–4.14, P < 0.05; Fig. 5A, Fig. S5A). Following the removal of other confounding factors, multivariate analysis indicated that the CDI was an independent prognostic predictor of SOC (HR = 89.71, 95% CI 36.12–222.85, P < 0.05; Fig. 5B). Univariate and multivariate Cox regression analyses combining the CDI and clinical parameters were also performed for the GSE63885 + GSE26193 and GSE140082 cohorts and the results were consistent with the TCGA-OV cohort (Figs. S5B and S5C). Based on this result, we constructed a nomogram including clinical parameters (age, stage, and grade) and the CDI to estimate 3-, 6-, and 9- year OS (Fig. 5C). Interrelationships among the nomogram score, age, stage, grade, and CDI score of patients are presented in the alluvial diagram (Fig. 5D). Calibration curves revealed that the nomogram-predicted OS was consistent with the observed OS (Fig. 5E). DCA indicated that the nomogram could accurately predict OS and was effective in clinical practice, outperforming other clinical factors (Fig. 5F). In addition, we further verified the accuracy of the nomogram via ROC analysis results of TCGA-OV, GSE63885 + GSE26193, GSE140082, and GSE53963, which showed that nomogram comprehensively considered more prognostic factor and was a good supplement to the prognostic accuracy of CDI (Fig. 5G; Fig. S4C; Fig. S4D). Overall, these results demonstrated from multiple aspects that the nomogram model showed good ability and reliability for OS prediction in patients with SOC; the predictive model we constructed may be obtained from https://leley.shinyapps.io/SOC-CDI/.

Fig. 5

Establishment and performance evaluation of nomogram model. A Univariate Cox hazard regression for the clinicopathologic characteristics and normalized CDI in the TCGA-OV cohort. B Multivariate Cox hazard regression for the clinicopathologic characteristics and normalized CDI in the TCGA-OV cohort. C Nomogram predicting the 3-, 6-, and 9-year OS of patients with SOC. D An alluvial diagram shows the interrelationship between the grade, stage, nomogram, and CDI groups in patients with SOC. E Calibration plots for 3-, 6-, and 9-year OS probabilities in the TCGA-OV cohort. F Decision curve analysis for predicting the OS. G ROC analysis of the nomogram in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts (only the 3-, 5-, and 7-year survival is shown, and the rest is detailed in supplementary materials)

Functional enrichment analysis of programmed cell death-genes model

We performed a functional enrichment analysis between the high and low CDI groups to elucidate the underlying molecular mechanisms of this PCD-genes model. DEGs between the high and low CDI groups were identified using a volcano plot (fold change > 1.5, P < 0.05; Fig. 6A). The functions of the DEGs were identified via GO and GSEA and further verified using KEGG and Reactome pathway analyses. The cluster-enrichment term network suggested that the DEGs were mainly involved in immune and cancer-related pathways, such as adaptive immune response, immunoglobulin-mediated immune response, NABA ECM REGULATORS, and interferon alpha/beta signaling (Fig. 6B). Additionally, the GSEA analysis revealed that the enriched pathways of the high CDI group included epithelial-mesenchymal transition (EMT), angiogenesis, inflammatory response, and apoptosis, which were largely associated with inflammatory responses, cell invasion, and migration (Fig. 6C and D). Furthermore, KEGG and Reactome pathway analyses revealed that the significantly enriched pathways in the high CDI group were apoptosis, the Wnt signaling pathway, pathways in cancer, the mitogen-activated protein kinase (MAPK) signaling pathway, and extracellular matrix organization, suggesting that the differential expression of these DEGs may be associated with changes in the tumor immune microenvironment (Fig. 6E and F). These results provide evidence that the aberrant expression of PCD genes may be involved in cancer development and illustrate cancer-immunity interaction as a potential mechanism of the PCD-genes model, prompting that relevant studies of this model in TME and immunotherapy are needed.

Fig. 6

Functional enrichment analysis of PCD genes model between low- and high-CDI patients with SOC. A Genes differentially expressed between the high- and low-CDI patients with SOC. Yellow: upregulated genes in the high CDI group; blue: upregulated genes in the low CDI group; B Network of the top 20 clusters with their representative enriched terms (one per cluster): colored by cluster ID, where nodes that share the same cluster-ID are typically close to each other. C Gene Set Enrichment Analysis (GSEA) enrichment plots for the DEGs; the top 25 pathways are displayed based on NES values. D GSEA of the epithelial-mesenchymal transition, angiogenesis, inflammatory response, and apoptosis. E KEGG analysis of genes in patients with SOC with high and low CDI. F Reactome pathway analysis of genes in patients with SOC with high and low CDI

scRNA-seq showed robust tumor signaling pathway activity in tumor cell clusters with high CDI

We analyzed the single-cell RNA transcriptome data from patients with SOC to explore SOC's microenvironment and immune status. We annotated different cell clusters, including NK/T cells, B cells, fibroblasts, epithelial cells, endothelial cells, and myeloid cells (Fig. 7A). The bubble plot showed the expression levels of cell-type marker genes, including myeloid cells (characterized by CD14, C1Q4, AIF1), endothelial cells (PECAM1, CLDN5), epithelial cells (EPCAM, KRT18, KRT19, COL1A1), fibroblasts (DCN, THY1), B cells (IGKC, CD79A, CD79B), and NK/T cells (CD2) (Fig. 7B). Furthermore, we used tSNE to visualize tumor cells and the expression levels of WFDC2 and PAX8 (Fig. 7C). Notably, WFDC2 is widely used as a biomarker in OC, and PAX8 is an essential histological marker in a majority of epithelial OCs, as they are highly expressed in most malignant OCs (Gokulnath et al. 2021; James et al. 2022). Based on the CDI value, the tumor cells were classified into high and low CDI clusters with a relatively clear resolution (Fig. 7D). As shown in Fig. 7E, we compared the scaled score of normalized activators of cancer-related pathways in the two CDI clusters, which showed that 10 activators varied significantly and that pathways such as epidermal growth factor receptor (EGFR), hypoxia, MAPK, transforming growth factor beta (TGF-β), and vascular endothelial growth factor (VEGF) were upregulated in the high CDI clusters. The EGFR, hypoxia, MAPK, p53, phosphoinositide 3-kinase (PI3K), TGF-β, VEGF, and Wnt pathways are involved in the progression, invasion, metastasis, or drug resistance of OC (Basu et al. 2015; Belur Nagaraj et al. 2021; Chen et al. 2018; Dorayappan et al. 2018; Huang et al. 2020; Wang et al. 2023; Zhao et al. 2021a).

Fig. 7

Single-cell transcriptome analysis reveals an association between CDI and malignant tumor cells. A t-distributed stochastic neighbor embedding (tSNE) visualization of the diverse cell types in tumor samples from GSE184880. B Bubble plots of cell-type marker gene expression levels. C tSNE visualization of the tumor cells and WFDC2 and PAX8 expression. D tSNE and violin plots showing the high and low CDI tumor groups and the CDI values of the tumor cells. E Box plot of scaled scores of the cancer-relevant pathways in the tumor cells. F GSEA of the hallmark epithelial-mesenchymal transition gene set and hypoxia. G Chord diagram of the signaling pathways from the high and low CDI tumor groups to other cells. H Heatmaps and Circos plots of the LAMININ, COLLAGEN, and VEGF signaling pathway networks. I Heatmaps and Circos plots of the TIGIT signaling pathway networks

Furthermore, GSEA revealed that the hallmark EMT and hypoxia gene sets were enriched in the cells of high CDI clusters (Fig. 7F). Additionally, we investigated the interactions between tumor cells in the high and low CDI groups and other types of cells in the TME. The chord diagram shows high and low CDI clusters with different cell signaling pathways (Fig. 7G). In the LAMININ, COLLAGEN, and VEGF signaling pathway networks, tumor cells with a high CDI produced more signaling associations with endothelial cells and fibroblasts (Fig. 7H), which may facilitate the migration and metastasis of tumor cells and angiogenesis (Carmeliet 2005; Hao et al. 2022; Song et al. 2022). In the TIGIT signaling pathway network (TIGIT-NECTIN2 axis), tumor cells with a high CDI interacted with NK/T cells more significantly, which was conducive to tumor immunosuppression (Fig. 7I) (Ho et al. 2021; Sim et al. 2022). The above results illustrated that tumor cells with a high CDI in patients with SOC are more inclined toward invasion, metastasis, and angiogenesis and have stronger tumor signaling pathway activities.

We performed scRNA-seq on another cohort, GSE213243, and obtained consistent results (Fig. S6). The high CDI group was correlated with the androgen, EGFR, estrogen, hypoxia, JAK-STAT, MAPK, NF-kB, PI3K, TGF-α, Trail, and VEGF pathways (Fig. S6E) and contained enriched EMT and hypoxia gene sets (Fig. S6F). Expression of the cancer-associated metastasis gene set in high CDI clusters was significantly increased, indicating that patients with SOC with a high CDI may be more prone to cancer metastasis (Fig. S6G). In addition, we explored the relationship between the tumor cells and model genes (Fig. S7). Violin plots showed the magnitude of the CDI values for different cell types, and the CDI was highly composed of epithelial cells (Fig. S7A). The bubble plot showed that different CDI model genes may correspond to different components in the TME, such as TREM2, mainly to myeloid cells, and CD38 to B cells (Fig. S7B). Moreover, we separately calculated the scores of tumor cells in single-cell transcriptome data using different death-type genes and corresponding parameters from the model, which showed that the scores from the apoptosis gene set and the lysosome-dependent cell death gene set differed most significantly between the two groups, suggesting that genes with these two models dominate the CDI division of tumor cells (Fig. S7C). Furthermore, differential analysis of tumor cells in the two single-cell datasets (high CDI vs. low CDI) found that all genes from the lysosome-dependent cell death gene set (BLOC1S1, AP1S2) were stable in both datasets (Fig. S7D).

Dissection of tumor microenvironment based on programmed-cell death signature

PCD may influence cancer progression via the TME. Hence, we conducted landscaping of the TME using the predictive model. We attempted to identify 22 tumor immune cell infiltration landscape using CIBERSORT to determine whether the proportion of immune cells varied between the high and low CDI groups. The TCGA-OV results revealed that M2 macrophages and neutrophils were significantly higher in the TME of the high CDI group than in the low CDI group. In comparison, M1 macrophages, activated memory CD4+ T cells, Tfh cells, and γδT cells were higher in the low CDI group (Fig. 8A). GSE63885 + GSE26193 and GSE140082 also revealed a higher proportion of M1 macrophages in the low CDI group than in the high CDI group (Figs. S8A and S8B). We found that in these three datasets, the high CDI group had a higher M2/M1 macrophage ratio (M2/M1 ratio), and the correlation analysis showed that the M2/M1 ratio is positively correlated with CDI (Fig. 8B).

Fig. 8

Dissection of tumor microenvironment (TME) based on PCD signature. A Box plots of the proportions of 22 types of immune cells between the high and low CDI groups, as predicted by CIBERSORT. B Violin plots of the log-homogenized M2/M1 macrophage ratio (M2/M1 ratio) between the high and low CDI groups in the TCGA-OV, GSE63885 + GSE26193, and GSE140082 cohorts. C tSNE visualization of myeloid cells from five patients with SOC in the GSE184880 cohort. D Bubble plot of the myeloid subset marker gene expression. E tSNE visualization of the CDI cluster group of macrophages. F Lollipop plot of the GSEA results for macrophage M1 and M2 scoring. G Bubble plot of the differentially expressed genes between the high and low CDI clusters of macrophages. H Bar plots showing that macrophages in the high CDI cluster are more likely to originate from high-stage patients

We performed further analyses using myeloid cells from scRNA-seq data (Sun et al. 2022). We showed the main constituent myeloid cells and their marker genes' expression levels (Fig. 8C and D). Next, tSNE analysis was conducted to classify the macrophages into high and low CDI clusters based on their CDI values (Fig. 8E). Subsequently, GSEA revealed that the high CDI group had more significant M2 macrophage characteristics than the low CDI group (Fig. 8F). Meanwhile, we found that secretion of factors associated with anti-inflammatory (MRC1, CXCL8, and MAFB) (Cambier et al. 2023; Kim 2017) in the high CDI group was significantly enhanced, while CXCL9, CXCL10, and CXCL11 expression in the low CDI group was increased (enhanced paracrine CXCL9, CXCL10, and CXCL11 expression indicates anti-tumor activity (Tokunaga et al. 2018); (Fig. 8G). Regarding the relationship between CDI and the tumor stage, we found that myeloid cells with a high CDI were mainly from patients with an advanced stage of cancer. Further, we observed that tumors mostly metastasized to the extra-pelvic involvement of the peritoneum (IIIb) (Fig. 8H). The high CDI group formed a TME dominated by M2 macrophage infiltration, whereas the low CDI group exhibited M1 macrophage infiltration.

Colocalization of CDI and malignant markers identified by spatial transcriptomics

Next, we performed a spatial transcriptional analysis of the OC samples from 10 × Genomics (FFPE samples, HE staining), and pathologists annotated the samples to confirm the spatial relationship between the CDI and malignant tumor cells. Based on the spatial characteristics of the cells, Leuven clustering grouped all the cells into nine clusters (Fig. 9A). Deconvolution of the spatial transcriptome using the scRNA-seq data from Xu et al. (Xu et al. 2022) confirmed the cell composition of the nine clusters; the results showed that fibroblasts dominated clusters three and five, whereas the remaining clusters were mainly epithelial cells (Fig. 9B). The spatial distribution characteristics of each cell cluster are shown (Fig. S9A). The expression distribution of malignant markers of OC (WFDC2 and PAX8) and pathological identification enabled us to determine the aggregated areas of tumor cells (Fig. 9C; Fig. S9B).

Fig. 9

High CDI region overlaps with malignant areas of the tumor. A Projections and tSNE visualization of spot clusters from a patient with SOC. B Bar plots of cell proportions after deconvolution. C tSNE visualization of the main cell type for each spot and the expression of SOC marker genes. D tSNE visualization of the myeloid cell deconvolution and normalized CS values. E tSNE and violin plots showing the CDI group of clusters dominated by tumor cells. F tSNE visualization of angiogenesis-related gene expression. G Bubble plots showing the enriched GO terms (n = 5) of the top 200 upregulated genes in the high CDI cluster dominated by tumor cells

Combining the distribution of myeloid cells, the expression of macrophage markers, CD14, APOE, C1QA, C1QC, and the scaled CXCL9/SPP1 (CS) ratio (Bill et al. 2023), we found macrophage aggregation associated with poor prognostic characteristics in the cluster 2 region (red box) (Fig. 9D; Fig. S9C). Based on the spatial distribution characteristics of the CDI and angiogenesis-related genes, we found that high CDI and angiogenesis characteristics (VEGFA + /PDK1 + /STC1 +) (Claesson-Welsh and Welsh 2013; Law and Wong 2013; Zhou et al. 2021) were co-located in cluster 2 (Fig. 9E and F). Angiogenesis may cause the formation of malignant ascites and metastasis in OC, accelerating its progression (Monk, Minion and Coleman 2016). We performed GO enrichment analysis on the top 200 genes, whose expression in the high CDI group was significantly higher than that in the low CDI group. Genes were significantly enriched in response to hypoxia (Fig. 9G), which possibly activates the expression of hypoxia-inducible factors (HIFs) and multiple angiogenic growth factors, contributing to intratumoral blood vessels (Wicks and Semenza 2022). Therefore, cluster 2 is likely to be a malignant tumor region. In summary, colocalizing the CDI with malignant markers of tumor cells in a space may predict the immune microenvironment and malignant regions of tumors.

Identification of mutation landscape and tumor neoantigens

The efficacy of immunotherapy is affected by various factors, such as immune infiltrates in the TME and the tumor mutational landscape. We speculated that patients with different CDI scores may show differences in tumor progression and immunotherapeutic responses. Furthermore, somatic mutations produce neoantigens that enhance tumor-specific immune responses, making neoantigens an emerging target for personalized immunotherapy (Xie et al. 2023). Therefore, we analyzed the somatic mutation frequency in patients with SOC to discover potential neoantigens. The top 20 most frequently mutated genes are displayed (Fig. S10A). Of these genes, the mutation rates of AHNAK and FAT3 were significantly higher in the low CDI group, whereas the other genes showed no statistical differences (Fig. S10B). Differentially mutated genes (DMGs) were identified and showed a high mutation frequency in the low CDI group (Fig. 10A), suggesting a cumulative effect of low-frequency mutations. Furthermore, we found higher tumor mutational burden (TMB) and neoantigen levels in the low CDI group (PTMB < 0.001, PNeoantigen = 0.015; Fig. 10B and C). Based on the KM analysis, when the CDI integrated with the TP53 mutation status, neoantigens, and TMB, patients with high TMB levels exhibited a longer OS than those with low TMB levels and a low CDI (Low CDI + Low TMB vs. Low CDI + High TMB, P < 0.001; Fig. 10D). Therefore, combining the CDI score with the TMB, we can accurately predict the prognosis of patients with SOC (Low CDI + High TMB vs. High CDI + Low TMB, P < 0.001; Fig. 10D). By contrast, the TP53 mutation status and neoantigens did not reliably predict survival in the patients with SOC (Low CDI + TP53 mutation vs. Low CDI + wild-type TP53, P = 0.77; High CDI + TP53 mutation vs. High CDI + wild-type TP53, P = 0.250; Fig. 10E; Low CDI + Low neoantigen vs. Low CDI + High neoantigen, P = 0.713; High CDI + Low neoantigen vs. High CDI + High neoantigen, P = 0.405; Fig. 10F). Thus, our model based on PCD is more effective in predicting survival than the existing biomarkers (Low CDI + TP53 mutation vs. High CDI + TP53 mutation, P < 0.001; Low CDI + wild TP53 vs. High CDI + wild TP53, P < 0.001; Low CDI + High neoantigen vs. High CDI + High neoantigen, P < 0.001; Low CDI + Low neoanti