We initiated our study by sourcing clinical information and RNA-seq profiles of patients with OC from TCGA database (https://www.cancer.gov/). We retrieved senescence-related genes from the GeneCards database (https://www.genecards.org). Single-cell RNA-seq data were acquired from GSE154600 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE154600). Tumor tissues were procured from patients with OC who underwent primary resection without prior chemotherapy, confirmed through pathology at the Department of Gynecology, Clinical Oncology School of Fujian Medical University, Fujian Cancer Hospital. The study secured approval from the ethical committee of Fujian Cancer Hospital (K2022-052-01). Informed consent was obtained from all participants, apprised of the research objectives.

The analysis encompassed 2740 differentially-expressed genes (DEGs) derived from the examination of raw count RNA-seq data in 197 patients with platinum sensitivity and 90 patients with platinum resistance to OC from TCGA using R package “DEseq2”. Employing Hallmark enrichment analysis, we scrutinized mechanisms or pathways with potential relevance to chemoresistance. After standardization of data by vst function in R package “DEseq2”, 18 prognostic senescence-related DEGs were extracted through univariate Cox analysis, and protein-protein interaction (PPI) was evaluated using the STRING database (https://version-11-5.string-db.org/ ).

Through multivariate Cox regression analysis, we identified three independent prognostic-related genes. The normality of the expression data was verified using Kolmogorov-Smirnov test and visualized by QQ plots. To test the significance of differences between the two groups, we performed Student’s t-test for normally distributed data, and Wilcoxon rank sum test for non-normally distributed data. Subsequently, we constructed a prognosis model following the formula: risk score=SGK1*coef (SGK1)-VEGFA*coef (VEGFA). Patients were categorized into high- and low-risk groups based on the median score value, and the prognosis for these groups was scrutinized. We utilized the “survival” R package to execute Kaplan–Meier (K-M) analysis, probing for the survival differences between OC patients in the low- and high-risk groups. Based on the somatic mutation data from TCGA, we conducted gene mutation through “maftools” package. Single-cell RNA-seq data was integrated by “Seurat” and “SingleR” R packages. We normalized data with Log Normalize method, then utilized t-distributed stochastic neighbor embedding (t-SNE) via the “RunTSNE” function to cluster and visualize cell populations. Finally, we explored the expression of these two genes across various cell types.

To estimate the proportion of 22 immune cell types between the low- and high-risk groups, we employed the CIBERSORT algorithm (13). The R package “ESTIMATE” was employed to evaluate stromal, immune, and tumor purity scores of OC, based on the proportion of immune and stromal cells. We forecasted the chemotherapy response of commonly used drugs for OC patients through the Genomics of Drug Sensitivity in Cancer (GDSC; https://www.cancerrxgene.org/) database. The half-maximal inhibitory concentration (IC₅₀) was determined using the “pRRophetic” package to assess chemotherapy response (14).

Specimens, sized 1- 2 cm³, were minced into 1-mm³ fragments and incubated in a tissue digestion solution (K601003; Tumor Tissue Digestion Solution; bioGenous™) at 37°C for 20-40 minutes. Digestion duration was determined through microscopic observations of cells dissociating into 10- to 20-µm small clusters. The suspension was then filtered through a 100-µm nylon cell strainer and centrifuged at 300 g for 5 minutes. After lysing red blood cells using the Red Blood Cell Lysis Solution (E238010; bioGenous™) and centrifuging again to wash the cell pellet with DMEM (Gibco), a portion of the suspension was mixed with organoid cryopreservation medium (E238023; bioGenous™). The mixture was then placed in a programmed cooling box at -80°C for 24 hours and subsequently stored in liquid nitrogen. Approximately 2000-5000 cells per well were mixed with 35µL basement membrane extract (BME; bioGenous™) and seeded into pre-warmed 24-well plates. Following BME solidification, each well was incubated with 500 µL human OC organoid medium (K2168-OC; bioGenous™) at 37°C. The medium was replenished every 2-3 days. Capture images every 3 days with a microscope (TS2; Nikon) and a camera (SYA-C20). Organoids were passaged approximately every 14 days by dissociation with an organoid dissociation solution (E238001; bioGenous™) for 10 minutes at 37°C. The success rate for establishing OC PDOs was 50%. After organoids were propagated to the first generation, immunohistochemistry was performed to validate whether their origin was consistent with tumor tissue. The process is outlined in Figure 1 .

For organoid construction, the previously mentioned cell suspension was counted using a cell counter. Approximately 2500 cells per well were placed in a 96-wells 3D cell culture plate (Cat.NO: HCKB-1196UA; HonrayMed™) with 200 µL of human OC organoid isolation medium (K2168-OC; bioGenous™) at 37°C for 2-4 days. Subsequently, the medium was replenished with a medium mixture containing varying concentrations of cisplatin (catalog no. C2210000; Sigma): 0, 0.78, 1.56, 3.12, 6.25, 12.5, 25, 50, and 100 µM.) This incubation persisted for 4 days. ATP quantification was performed to determine sensitivity to cisplatin by assessing the inhibition rate of organoid activity. IC₅₀, IC_90, and peak plasma concentration (PPC) were the evaluation metrics used to ascertain whether or not the organoids exhibited resistance to cisplatin. Cases where IC₅₀>25% PPC and IC₉₀>100% PPC were classified as resistant, while those with IC₅₀<25% PPC and IC₉₀< or > 100% PPC, or IC₅₀>25% PPC and IC₉₀<100% PPC, were deemed sensitive (15).

Total RNA extraction for human OC organoids in the sensitive and resistant groups was performed following the instructions of the RNA extraction kit (LS1040; Promega). cDNAs were synthesized through reverse transcription (GoScript™ Reverse Transcription Mix, Oligo(dT) A2790; Promega). Quantitative polymerase chain reaction (Q-PCR) was performed using the SYBR Green Master Kit (Roche), with mRNA expression levels normalized to GAPDH. The primer sequences are provided as follows: SGK1-F: aaacacagctgaaatgtacgac; SGK1-R: ttggttaaaagggggagtaatc; VEGFA-F: atcgagtacatcttcaagccat; VEGFA-R: gtgaggtttgatccgcataatc; GAPDH-F: tgtgggcatcaatggatttgg; GAPDH-R: acaccatgtattccgggtcaat. After culturing P1 generation organoids for 14 days, collected organoids and lysed with RIPA buffer (Epizyme Biotech, PC101), The BCA protein assay kit (Epizyme Biotech, ZJ101) was used to measure protein concentration. The proteins were separated on SDS-PAGE gels and transferred to polyvinylidene fluoride (Millipore, IPVH00010, 0.45µm) and then blocked with 5% milk for 1 h at RT and immunoblotted with primary antibodies at 4°C overnight: GAPDH (cell signaling technology, 2118, 1:1000 dilution), VEGFA (Proteintech, 66828-1-Ig, 1:1000 dilution), SGK1 (Proteintech, 28454-1-AP, 1:1000 dilution). Finally, the membranes were incubated with secondary antibodies (abcam, ab205718, ab205719, 1:10000 dilution) and visualized by imaging systems.

Statistical analyses were carried out using Perl scripts (v5.30.0), R software (v.4.1.0), and GraphPad Prism (v8.0.2). Kaplan-Meier (K-M) analysis was employed to assess survival differences. Differential functions were analyzed using the Wilcoxon rank-sum test between two groups. All experiments were repeated in triplicate. The data were expressed as mean ± standard. P < 0.05 was considered as statistical significance.

Our investigation harnessed clinical data from the TCGA database, yielding 197 cases in the platinum-sensitive group and 90 cases in the drug-resistant group. Clinical pathological parameters were shown in Supplementary Table 1 . Subsequently, RNA-seq data for these cases were acquired and subjected to analysis, resulting in the identification of 2740 DEGs (|log2FC|>0 and P<0.05), including 1373 upregulated and 1367 downregulated genes, which were then visualized through heat-maps and volcano plots ( Figures 2A, C ). We further extracted 43 aging-related DEGs by intersecting 279 aging-associated genes with the 2740 DEGs ( Figure 2B ). Employing univariate Cox analysis, we identified 18 genes exhibiting significant correlations with prognosis, including 9 genes with negative associations and 9 with positive correlations ( Figure 2D ).

These 18 prognosis-linked genes were used to construct PPI networks, from which 12 pivotal genes formed a central network marked by interactions ( Figure 2E ). Functional enrichment analysis indicated their enrichment in pathways associated with hypoxia, inflammation, and glycolysis processes closely intertwined with cellular aging ( Figure 2F ).

Through multivariate Cox regression analysis on these 12 core genes, we obtained 3 genes significantly link to prognosis, depicted in Figure 3A . Distinct survival ability of OC patients with different expression level of IFNG, SGK1 and VEGFA were discerned (P<0.05) ( Supplementary Figure 1A ) ( Figure 3B ). The normality of the expression data was verified using Kolmogorov-Smirnov test and visualized by QQ plots. The expression of SGK1 and VEGFA was verified to be normally distributed ( Figure 3B ), whereas IFNG was not ( Supplementary Figure 1B ). Student’s t-test was used to exam the differences in expression levels between the platinum-resistant and sensitive groups for SGK1 and VEGFA, and IFNG was using the Wilcoxon rank sum test ( Supplementary Figure 1C ). Notably, SGK1 and VEGFA demonstrated significantly different expression (P<0.05), while IFNG presented no significant expression differences between two groups ( Figure 3C ). Therefore, we included SGK1 and VEGFA in our risk model construction following the formula, risk score=SGK1*coef (SGK1)-VEGFA*coef (VEGFA) (coef values were in Supplementary Table 2 ). Segregation of patients based on risk scores yielded the high- and low-risk groups, with those in the high-risk category facing a worse prognosis (P<0.05, Figure 3D ). Gene mutation waterfall plots for both groups are showcased in Figure 3E . In the GSE154600 dataset, there are 9 cell types marked mainly including cancer, immune, and stromal cells, and a t-SNE plot visualized cell subtypes in patients with OC ( Figure 4A ), with subtype annotations provided in Figure 4B . Both SGK1 and VEGFA exhibited increased expression in macrophages, suggesting a correlation between these risk genes and macrophage dysregulation in OC ( Figures 4C, D ).

Based on KEGG database, genes with higher expression levels in resistant group were involved in the peroxisome proliferator-activated receptor (PPAR) signaling pathway, Cytokine Receptor Interaction pathway, and Neuroactive Ligand Receptor Interaction pathway (P<0.05), the same trend was found in Interferon Alpha/Gamma Response pathway based on hallmark gene sets. While elevated involvement of oxidative phosphorylation pathway was found in genes highly expressed in sensitive group (P<0.01; Figures 5A, B ). We carried out the CIBERSORT algorithm to analyze the proportion of the 22 immunocyte. Discrepancies in the microenvironment between high- and low-risk groups primarily centered around CD8⁺ T cell, monocytes, and macrophages ( Figure 6A ). Estimate analysis depicted correlations between immune, stromal and tumor purity scores with high- and low-risk scores ( Figure 6B ). Figure 6C illustrated the heatmap of differences in infiltrating immune cells between the two groups.

We further investigated the correlations between the IC₅₀ of potential target drugs and risk scores. The smaller IC₅₀ presented, the more sensitive patients were to drugs. Patients with high risk appeared to benefit from drugs such as gemcitabine, dabrafenib, epirubicin, oxaliplatin, olaparib, teniposide, ribociclib, topotecan, and venetoclax ( Figure 7 ), while others seemed more beneficial for patients in the low-risk group.

Our study successfully obtained and cultured three OC samples into organoids, visually documented through growth progression as indicated by the red arrows in Figure 8A . The development of organoids from P0 to P2 showed in Figure 8A (Scale bar = 100 µm). Furthermore, patients’ clinical data are summarized in Table 1 . Immunohistochemical testing of p53, WT-1, and MUC16 in the organoids, the routine marker in diagnosing OC, exhibited striking concordance with tissue expression, ( Figure 8B , Scale bar for tissue = 70 µm, scale bar for organoid = 40 µm), affirming the homology between the organoids and the corresponding tumor tissue. We exposed organoids from the P1 generation to cisplatin diluted with medium across varying concentrations and captured their growth state after 72 hours. Figure 9A showed the representative growth status changes under 12.5 µM cisplatin medium in drug-resistant and sensitive PDOs (Scale bar = 100 µm). Notably, the sensitive group’s organoids exhibited shrinkage, rupture, and loss of original structure following cisplatin addition. By contrast, organoids in the resistant group displayed only mild shrinkage. The dose-response curve reflecting drug inhibition rates on the organoids is illustrated in Figure 9B . Organoids were categorized into cisplatin-resistant and -sensitive groups based on IC₅₀ and IC₉₀. Specifically, PDO1121 was recognized as cisplatin-resistant sample, whereas PDO0213, and PDO0315 were designated as the cisplatin-sensitive ones. In these groups, VEGFA exhibited low expression in cisplatin-resistant organoids, while SGK1 was prominently expressed ( Figure 9C ). Figure 9D illustrated the protein expression and the gray scale analysis of SGK1 and VEGFA in OC PDOs. SGK1 were higher expressed in PDO1121 and VEGFA were higher in PDO0213 and PDO0315.