By combining the clinical and transcriptome data from numerous datasets, the molecular landscape of ovarian cancer was examined. Datasets were obtained through the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/). The GSE66957 dataset (GPL15048 platform; Rosetta/Merck Human Rosetta Standard Transcriptome Array Custom Affymetrix 2.0 microarray) comprises 57 ovarian cancer and 12 normal ovarian surface epithelium samples (31) and was selected for differential-expression analysis as it is one of the few publicly available datasets containing both EOC and non-carcinoma ovarian tissues with adequate sample sizes. The GSE140082 dataset (GPL14951 platform; Illumina HumanHT-12 Whole Genome-cDNA-mediated Annealing, Selection, Extension and Ligation V4.0) comprises 380 ovarian cancer samples annotated with OS and PFS data (32) and was therefore selected as an independent cohort for survival validation. The two datasets were generated on different platforms; accordingly, they were normalized and analyzed independently and no cross-platform data merging was performed, thereby avoiding the introduction of batch effects from direct data integration. Probes were mapped to gene symbols using the corresponding platform annotation files; when multiple probes mapped to the same gene, the mean expression across those probes was used.

RNA-sequencing (RNA-seq) data and the corresponding clinicopathological annotations for TCGA-OV (https://portal.gdc.cancer.gov/projects/TCGA-OV) were obtained from the NCI Genomic Data Commons (https://portal.gdc.cancer.gov) and accessed through the cBioPortal for Cancer Genomics Application Programming Interface (https://www.cbioportal.org) in April 2026. Gene-level expression was retrieved as RNA-Seq V2 RNA-seq by expectation-maximization (RSEM) values from the TCGA-OV RNA-seq profile (ov_tcga_rna_seq_v2_mrna) and transformed as log₂ (RSEM + 1) before downstream analysis. Only primary tumor specimens with available KAZALD1 expression data and the transcriptomic features required for downstream analysis were retained, yielding 307 cases for expression profiling. Within this cohort, 29 cases with complete immune-cell marker gene expression profiles (including CD19, CD4, CD8A, Fc γ-receptor IIIb, CD68 and integrin subunit α-X) were used for the immune-infiltration correlation analysis and 303 cases with complete survival follow-up were included in the present survival analysis. The TCGA-OV cohort was not merged with the GEO datasets.

To validate the in-silico findings in a real-world setting, 136 EOC specimens were obtained from 136 female patients (median age: 51 years; range: 20–75 years) who underwent surgical resection between January 2008 and December 2013 at the Obstetrics and Gynecology Hospital of Tongji University (Shanghai, China). All specimens were enrolled according to pre-defined inclusion and exclusion criteria. The inclusion criteria were as follows: i) Histologically determined primary EOC; ii) receipt of primary cytoreductive surgery without prior neoadjuvant chemotherapy, radiotherapy or targeted therapy; iii) availability of adequate formalin-fixed, paraffin-embedded tumor tissues for immunohistochemical staining; and iv) complete clinicopathological and follow-up data for OS and DFS analysis. Patients with non-epithelial ovarian tumors, borderline tumors, ovarian metastatic tumors, insufficient or necrotic non-representative tissues or incomplete clinical and survival information were excluded. The clinicopathological data of patients, including age, FIGO stage, tumor size and TNM classification, were retrieved from medical archives. The OS was calculated from the date of initial surgery to the date of mortality or the final follow-up. The present study was approved by the Ethics Committee of the Obstetrics and Gynecology Hospital of Tongji University (Shanghai, China; approval no. KS25118) with individual patient informed consent exempted.

All ovarian tumor tissues were fixed in 4% paraformaldehyde at room temperature for 24 h, routinely dehydrated, paraffin-embedded and sectioned into 4-µm-thick continuous slices. The sections were dewaxed in xylene and rehydrated through a descending alcohol series. Antigen retrieval was performed using citrate buffer (pH 6.0) at 121°C for 15 min using a pressure cooker. After cooling and washing with PBS, tissue sections were incubated with anti-KAZALD1 primary antibodies (cat. no. 20815-1-AP; 1:100; Proteintech Group, Inc.) overnight at 4°C, followed by incubation with the secondary antibodies [HRP-conjugated Affinipure Goat Anti-Rabbit IgG (H+L); cat. no. SA00001-2; 1:200; Proteintech Group, Inc.] for 2 h at 37°C. The sections were counterstained with hematoxylin for 10 min at room temperature. Immunoreactive signals were developed using DAB solution (cat. no. P0202, Beyotime Biotechnology), according to the manufacturer's instructions. The samples were observed under an Olympus BX53 upright light microscope (Olympus Corporation) equipped with a digital camera.

A total of two independent researchers blinded to the clinical data scored the staining intensity as follows: 0 (negative), 1 (weak), 2 (moderate) and 3 (strong). Furthermore, the positive cell percentage was scored as follows: 1 (1–24%), 2 (25–49%), 3 (50–74%) and 4 (75–100%). The staining intensity and positive cell percentage scores were multiplied to obtain the final score, which was in the range of 0–12. Scores of >6 indicated upregulation, while those of ≤6 indicated downregulation.

Biological functions of KAZALD1 in the TME of EOC were examined using integrated bioinformatic analyses. The present study examined the functional enrichment of the 100 most significant genes co-expressed with KAZALD1 identified in the TCGA-OV cohort (Spearman's r >0.25; P<0.05) using the Metascape platform (https://metascape.org), Gene Ontology biological processes, Kyoto Encyclopedia of Genes and Genomes pathways and Reactome pathways. The hub genes in these enriched clusters were identified by constructing a gene-pathway association matrix.

The correlation of KAZALD1 expression with tumor-infiltrating immune cells (TIICs) was assessed using the Tumor Immune Estimation Resource 2.0 (TIMER 2.0; http://timer.comp-genomics.org/) algorithm. Immune-cell infiltration was first estimated with a marker-gene-based approach within the TIMER 2.0 framework and the results were cross-validated using five complementary deconvolution algorithms, namely Microenvironment Cell Populations-counter (MCP-counter; http://github.com/cit-bioinfo/MCP-counter), EPIC (http://epic.gfellerlab.org/), Cell-type Identification By Estimating Relative Subsets of RNA Transcripts (CIBERSORT; http://cibersort.stanford.edu/), xCell (http://xcell.ucsf.edu/) and quanTIseq (http://icbi.at/quantiseq). Spearman correlation analysis was performed to examine the correlation of KAZALD1 expression with infiltration scores of six main types of immune cells, including APCs. Python (version 3.10; Python Software Foundation) was used for visualization using the ‘matplotlib’ and ‘seaborn’ libraries.

All statistical analyses were performed using SPSS Statistics (version 25.0; IBM Corp.), R software (version 4.2; Posit Software, PBC) and Python (version 3.10; Python Software Foundation). Continuous data were compared using the Mann-Whitney U test. Prior to this, the Shapiro-Wilk test determined non-normal distribution of continuous variables. For categorical data, binary variables including lymph node metastasis (N classification) and recurrence were analyzed using χ² tests, while distant metastasis (M classification) was assessed using Fisher's exact test. Ordered categorical variables (T classification and FIGO stage) were analyzed with Mann-Whitney U tests. The Kaplan-Meier curves for OS and DFS were plotted and compared using the log-rank test. The correlation of KAZALD1 expression with survival was examined using Spearman's rank correlation. The hazard ratio (HR) and 95% CI values were determined by constructing univariate and multivariate Cox proportional hazards models. The clinical variables were screened to reduce multicollinearity and consequently ensure the construction of a robust prognostic model. Two-sided P<0.05 was considered to indicate a statistically significant difference.

GEO transcriptomic data (GSE66957) was first compared between EOC and non-carcinoma ovarian tissues to find genes implicated in EOC development, before the field was narrowed to those meeting strict criteria for both differential expression [log₂ fold change (FC) >1.5; adjusted P<0.001] and survival impact (HR>1.5; P<0.05). In this comparison, KAZALD1 ranked 845th among 2680 significantly upregulated genes by adjusted P-value, falling markedly outside the top tier (Table SI). From this shortlist, KAZALD1 emerged as the primary candidate within the present study, chosen for its established roles in ECM remodeling, EMT and poor prognosis in other cancer types (25,27,28), rather than for topping the differential-expression list.

The KAZALD1 mRNA levels in the GSE66957 dataset were compared between the ovarian cancer and non-carcinoma tissue samples. Compared with those in the non-carcinoma samples, the KAZALD1 mRNA levels were significantly upregulated in the ovarian cancer samples (log₂ FC=1.73; adjusted P=5.42×10⁻⁴; Fig. 1A). This indicates the oncogenic role of KAZALD1.

Subsequently, the correlation between KAZALD1 upregulation and prognosis was evaluated. Patients in the independent GSE140082 cohort were categorized into the high-KAZALD1 and low-KAZALD1 groups based on the median KAZALD1 level. Kaplan-Meir analysis revealed that patients in the high-KAZALD1 group exhibited poor PFS (HR: 1.51; 95% CI: 1.17–1.96; log-rank P=0.0015; Fig. 1B) and OS (HR: 2.21; 95% CI: 1.45–3.37; log-rank P=0.0002; Fig. 1C). Multivariate Cox regression analysis revealed that KAZALD1 was a significant independent predictor of mortality in patients with EOC even after controlling for age and American Joint Committee on Cancer stage (HR: 1.15; 95% CI: 1.04–1.28; P=0.0049; Fig. 1D).

The present study examined the KAZALD1 levels in 136 EOC specimens through IHC analysis. The representative images across the staining intensity range are shown in Fig. 2A-D. The expression of KAZALD1 was primarily localized to the cytoplasm of tumor cells. Based on the 0–12 composite scoring system, 42 (30.9%) and 94 (69.1%) patients exhibited IHC scores of >6 (high expression) and ≤6 (low expression), respectively (Fig. 3A). As shown in Table I, the upregulation of KAZALD1 was associated with advanced FIGO stage (P=0.011; Mann-Whitney U test), distant metastasis (P=0.002; Fisher's exact test) and lymph node metastasis (P=0.025; χ² test). Spearman correlation analysis revealed that KAZALD1 expression was correlated with distant metastasis (r_s=0.224; P=0.009) and tumor size (r_s=0.208; P=0.015; Table II). These findings suggest that KAZALD1 upregulation increases the aggressiveness of EOC.

As KAZALD1 expression was associated with tumor aggressiveness, the present study further assessed the independent prognostic value of KAZALD1. Based on the median IHC score (6.0), patients were categorized into the low-KAZALD1 (n=94) and high-KAZALD1 (n=42) groups. Univariate Cox regression analysis revealed that KAZALD1 expression (HR: 1.17; 95% CI: 1.07–1.28; P=0.0004), FIGO stage (HR: 2.08; 95% CI: 1.50–2.88; P<0.0001), tumor size (HR: 1.17; 95% CI: 1.12–1.22; P<0.0001) and distant metastasis (HR: 5.69; 95% CI: 3.33–9.71; P<0.0001) were associated with poor OS (Fig. 3B). KAZALD1 remained an independent prognostic biomarker for EOC even after adjusting for the confounders age, tumor size and distant metastasis in the multivariate model for limiting collinearity (HR: 1.09; 95% CI: 1.00–1.19; P=0.0417; Fig. 3C).

Subsequently, the clinical outcomes of the two groups were compared. Patients in the high-KAZALD1 group exhibited decreased OS (P=0.0328; Fig. 3D) and DFS (P=0.0809; Fig. 3E) when compared with those in the low-KAZALD1 group. Spearman correlation analysis revealed that KAZALD1 expression was inversely correlated with OS (r=−0.2510) and DFS (r=−0.1720; Fig. 3F). The two subgroups exhibited differential survival outcomes. Kaplan-Meier analysis revealed that patients in the high-KAZALD1 group exhibited poor median OS (45 vs. 77 months; log-rank P=0.0151; Fig. 3G) and median DFS (35 vs. 60 months; log-rank P=0.0372; Fig. 3H). The median OS and DFS of patients in the high-KAZALD1 group were 41.6 and 36.0% lower compared with those of patients in the low-KAZALD1 expression subgroup (Table III). In the univariate hazard model, every single-point increase in IHC score increased the mortality and progression risks by 17.3% (HR: 1.173, 95% CI: 1.068–1.288) and 11.8% (HR=1.118; 95% CI: 1.002–1.248), respectively.

Subsequently, the importance of KAZALD1 expression was determined. KAZALD1 upregulation was associated with disease aggressiveness and poor survival. Thus, the molecular mechanisms underlying these effects were explored. Functional enrichment analysis of the 100 most significant genes co-expressed with KAZALD1 revealed the following two key pathways: ‘Extracellular matrix organization’ (P=1.23×10⁻⁹; 14.2-fold enrichment) and ‘TGF-β signaling’ (P=3.45×10⁻⁶; 9.8-fold enrichment; Fig. 4A). In the gene-pathway map, TGFB1 and IL10 were found to be central components of the ECM/TGF-β pathway for immunomodulation (Fig. 4B). Therefore, KAZALD1 was hypothesized to aid in the establishment of an immunosuppressive TME.

To test this hypothesis, the correlation of KAZALD1 expression with TIICs was examined using the TIMER 2.0 algorithm. KAZALD1 expression was found to be negatively correlated with the infiltration of APCs, especially dendritic cells (r=−0.486; P=0.0075; Fig. 4D) and macrophages (r=−0.431; P=0.0196; Fig. 4E) but not with that of adaptive immune cells, such as B cells or CD8+ T cells (Fig. 4C). Collectively, these correlative findings suggest an association between high KAZALD1 expression and a TME characterized by ECM/TGF-β pathway enrichment and reduced estimated infiltration of dendritic cells and macrophages.