Affymetrix miRNA microarray data from OC exosomes and cells were available from Gene Expression Omnibus (GEO; series GSE76449) (Kanlikilicer et al., 2016). Exosomal and original samples of chemo-sensitive OC cells and normal ovarian cells (n = 16) were included to rule out the influence of chemotherapy on prediction results. RNA sequencing (RNA-seq) data from normal ovarian tissues were obtained from the Genotype-Tissue Expression (GTEx) portal. RNA-sequencing (RNA-seq) data and clinical data from OC patients were downloaded from GEO (series GSE9891), Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/), International Cancer Genome Consortium (ICGC), and The Cancer Genome Atlas (TCGA), as detailed below (Wang et al., 2016; Zhang et al., 2019).

Bioinformatics analyses of RNA-seq data from TCGA patients were performed using the UCSC bioinformatic pipeline (TOIL RNA-seq). Each dataset was processed identically after removing unavailable data. Clinical data from TCGA were prepared using the TCGA-GDC (Genomic Data Commons) server for subsequent analysis (Goldman et al., 2020).

Finally, all datasets underwent identical processing to achieve an average expression level of samples from the same patient and unavailable data were deleted (NA).

In this experiment, RNA-seq data were obtained from TCGA and GTEx. Differentially expressed miRNAs (DEMIs) between exosomal and original samples of OC cells were identified with limma R Bioconductor package (version 3.48.3, p < 0.001 and log2-fold change [logFC] > 4, Student’s t-test) (Ritchie et al., 2015). Differentially expressed mRNAs (DEMs) were screened using the same method (adjusted p < 0.001 and logFC > 3). A boxplot was generated with ggplot2 in R.

Five target gene prediction databases were employed to predict the downstream targets of miR-1825, including TargetScan 7.2, miRDB, TarBasev.8, miRmap, and miRwalk 3.0. Intersections among the five databases were regarded as miR-1825 targets (Agarwal et al., 2015; Chen and Wang, 2020; Karagkouni et al., 2018; Sticht et al., 2018; Vejnar et al., 2013).

Survival data (age, clinical stage, histologic grade, survival status, and length of overall survival) of OC patients were collected from TCGA, GEO, and ICGC and merged with gene expression data. We assessed the prognostic ability of miRNAs or CLEC5A in OC using univariate and multivariate Cox regression hazard analyses combined with Kaplan-Meier survival curves.

We performed Gene Ontology Biological Process (GOBP) term and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses with clusterProfiler in R to identify CLEC5A targets (p < 0.05) and explore mechanisms of the role of CLEC5A in OC (Yu et al., 2012). Then, Gene Set Enrichment Analysis (GSEA) was utilized to confirm the selection of target genes by GO or KEGG analyses (p < 0.05 and false discovery rate [FDR] < 0.25) and characterize the activity of KEGG pathways between low and high CLEC5A expression groups in OC, as described in detail elsewhere (Subramanian et al., 2005).

We assessed correlations of CLEC5A gene expression with immune cell infiltrates and the expression of immune suppressors to explore the ability of CLEC5A to predict the outcome characteristics in low and high CLEC5A expression groups in OC. TCGA RNA-Seq expression profiles of immune and stromal cells were utilized. Tumor-infiltrating lymphocytes (TILs) and stromal cells were identified with ESTIMATE (Estimation of Stromal and Immune cells in malignant Tumor tissues using Expression data) in R. Correlations of CLEC5A with immune/stromal cell infiltration were assessed, and the results were validated in a GEO cohort of 276 qualified OC cases whose expression and survival data were available for review (GSE9891). This cohort was also used for subsequent immune cell infiltration analysis (Yoshihara et al., 2013).

The infiltration of 22 immune cell subpopulations in OC tissues was estimated using the CIBERSORT algorithm that can accurately quantify TILs in biopsied tumor tissues and the LM22 signature matrix file containing 22 immune cell reference profiles from TCGA and GEO (Chen et al., 2018). Patients were categorized into the low and high CLEC5A expression groups according to the cutoff of CLEC5A gene expression. The immune cell landscape was compared between the two groups. Sixteen tumor immune suppressors have been shown to exert suppressive effects on solid tumors (Cassetta and Pollard, 2018). Among them, interleukin 6 (IL6), tumor necrosis factor (TNF), interferon gamma (IFNG), epidermal growth factor (EGF), major histocompatibility complex class I (MHC-I), CD96 molecule (CD96), leukocyte immunoglobulin like receptor B1 (LIR1), programmed cell death 1 (PDCD1), programmed cell death 1 ligand 2 (PDCD1LG2), CD80 molecule (CD80), and CD86 molecule (CD86), which were frequently reported in OC were selected and their expression levels were compared between the two groups.

We selected five agents (cisplatin, docetaxel, doxorubicin, gemcitabine, and paclitaxel) from 22 recommended options in the 2020 National Comprehensive Cancer Network (NCCN) clinical practice guidelines in oncology on ovarian cancer (Armstrong et al., 2021). The predicted IC50 was estimated with pRRophetic (version 0.5; https://doi.org/10.1371/journal.pone.0107468) in R and compared between low and high CLEC5A expression OC patient groups from TCGA (Geeleher et al., 2014).

Patients (n = 9) who had pathologically confirmed OC and had undergone radical excision between May 2019 and May 2021 at the Fujian Medical University Cancer Hospital (or Fujian Cancer Hospital) without prior therapy were selected. Specimens (7 OC and 2 peritumoral tissue specimens) were collected, flash-frozen in liquid nitrogen, and reviewed independently by two senior pathologists. The study protocol was approved by the Ethics Committee of Fujian Cancer Hospital.

Immunohistochemistry (IHC) analysis was performed as previously described (Chen et al., 2021). Briefly, formalin-fixed paraffin-embedded sections were dewaxed, dehydrated, and rehydrated. The slides were incubated overnight at 4°C with an anti-CLEC5A antibody (ab203200, Abcam, China, 1:500). They were then incubated at 37°C with a secondary antibody (ab7090, Abcam, China, 1:500). Immunostaining was performed using a SABC(Strept Avidin-Biotin Complex) kit (SA0025, Solarbio, China). Next, the slides were counterstained with Mayer hematoxylin solution (Solarbio, China) for nuclear staining. Images were obtained using a VANOX microscope (Olympus, Japan) and scored. The scoring was as follows: 0 points (no staining), 1 point (light yellow), 2 points (brownish yellow), and 3 points (brown). For each specimen, fields were randomly selected, and the proportion of CLEC5A-positive (CLEC5A+) cells was calculated and scored as follows (1 point, 0%–25% of CLEC5A+ cells; 2 points, 26%–50% of CLEC5A+ cells; 3 points, 51%–75% of CLEC5A+ cells; 4 points, 76%–100% of CLEC5A+ cells). Specimens with a pathological IHC score greater than the cutoff were identified as high CLEC5A+ OC and low CLEC5A+ OC if the cutoff was otherwise.

Categorical variables were expressed as numbers (percentages) and compared using Spearman’s test. Overall survival data were analyzed using Kaplan-Meier survival curves and compared using the log-rank test. Continuous variables following normal distribution were presented in mean ± standard error of measurement (SEM) and compared using the Student’s t-test. Correlations of CLEC5A gene expression with immune/stromal cell infiltration were assessed using Spearman’s correlation test. All statistical analyses were performed using R 4.1.0, and a two-tailed p-value of < 0.05 was deemed statistically significant.

We explored the expression patterns in four groups of samples using the GSE76449 dataset. These were: (1) normal ovarian cells; (2) exosomes released from normal ovarian cells; (3) ovarian cancer cells and (4) exosomes released from OC cells. We then identified DEMIs among these four groups. The result indicated that microRNAs exhibited a specific distribution pattern in exosomes and their source cells. Interestingly, the expression levels of miR-1281, miR-1825, miR-6877-3p, and miR-3921 were upregulated in exosomes released from OC cells while they were significantly downregulated in the other three sample groups (Figs. 1A–1E). These observations were also supported by a previous study (Rashed et al., 2017). These miRNAs may be secreted from OC cells after being wrapped in exosomes. Further, the prognostic significance of miR-1281,miR-1825, and miR-6877-3p was evaluated in the high vs. low expression groups using TCGA survival data. MiR-3921 was ruled out as its expression was absent in most OC samples (Fig. 2A). In the Kaplan-Meier survival analysis, miR-1825 upregulation in OC cells was correlated with improved overall survival (p < 0.05) (Figs. 2B–2D). This suggests that it might act as a tumor suppressor in OC.

We performed target gene predictions using TargetScan7.2, miRDB, TarBase v.8, miRmap, and miRwalk3.0 to identify miR-1825 targets. The results displayed 52 genes (Fig. 2E). Differential expression analysis of these 52 target genes using the limma method revealed that the mRNA expression levels of calcium voltage-gated channel auxiliary subunit beta 2 (CACNB2) and KIT proto-oncogene receptor tyrosine kinase (KIT) were downregulated and those of NEDD4 binding protein 3 (N4BP3), CLEC5A, NOTUM palmitoleoyl-protein carboxylesterase (NOTUM), and doublecortin domain containing 2 (DCDC2) were upregulated in the tumor group from the TCGA database in comparison with normal ovarian samples from the GTEx database (|logFC| >3 and adjusted p-value < 0.001) (Fig. 2F). Given the tumor-suppressive effect of miR-1825, the prognostic potential of the four genes with higher expression (N4BP3, CLEC5A, NOTUM, and DCDC2) was assessed. The Kaplan-Meier curves showed that among these four genes, CLEC5A appeared to be a prognostic potential marker for OC, as validated using the validation cohort (GSE9891 and ICGC dataset OV-AU) (Figs. 3A–3C), and was significantly associated with worse overall survival of OC patients (p < 0.05). CLEC5A was a poor prognostic factor in ovarian cancer, cervical cancer, and uterine sarcoma (Fig. 4A). Moreover, CLEC5A could serve as a prognostic biomarker independent of clinicopathological parameters (age, clinical stage, and histologic grade) in OC, as validated using GSE9891 and OV-AU validation cohorts (Fig. 4B). The IHC assays revealed that CLEC5A protein expression was significantly increased in OC vs. normal ovarian tissues (Fig. 4C).

To clarify the functions of CLEC5A expression signatures associated with OC, we assayed biological processes of GO and KEGG pathways enrichment analyses of genes highly correlated to CLEC5A based on the TCGA dataset. As shown in Figs. 5A and 5B, the intersection of GO terms and KEGG pathways showed that a proportion of highly correlated genes was associated with the immune response that plays an important role in tumor inhibition and promotion. These results suggested that immune alterations might contribute to OC occurrence and development. However, the exact mechanism that CLEC5A plays in immunity in OC is not yet clear.

To further verify the results of GO/KEGG enrichment analyses, we conducted a GSEA with the low and high CLEC5A expression datasets. GSEA analyses showed significant differences (p < 0.05) in the enrichment of immune-related pathways and biological processes (Figs. 6A and 6B). Interestingly, CLEC5A also seemed to activate the PI3K/Akt pathway in OC. This activation has been confirmed by experiments in brain glioblastoma by Fan et al. (2019). Given the importance of the PI3K/Akt signaling pathway and the involvement of the immune system on OC, we speculate that the activated PI3K/Akt signaling pathway may contribute in part to the development of the immune environment in OC.

To further investigate the interaction between CLEC5A and the immune microenvironment in OC, the R package ESTIMATE was used to examine correlations between immune cell and stromal cell infiltration levels and CLEC5A expression. Notably, we observed a positive correlation between CLEC5A expression and immune cell/stromal cell infiltration. A similar result was obtained based on the GSE9891 dataset (Figs. 7A and 7B). These results indicated that CLEC5A might play a role in mediating the immune response and immune cell and stromal cell infiltration in ovarian tumors. To further explore which immune cell subtype correlated with the CLEC5A expression, we utilized CIBERSORT to calculate the infiltration levels of 22 immune cell subtypes. These included B cells (naive B cell, memory B cell, and plasma cell), T cells [CD8 T cell, naive CD4 T cell, memory resting CD4 T cell, memory activated CD4 T cell, follicular helper T cell (Tfh), regulatory T cells (Tregs), and gamma delta T cell], natural killer (NK) cells (resting NK T cell and activated NK cells), and myeloid subsets [monocytes, MO macrophages, M1 macrophages, M2 macrophages, and resting dendritic cells (DC), activated DCs, resting mast cells, activated mast cells, eosinophils, and neutrophils]. After integrating the results from CIBERSORT (p < 0.05) based on the TCGA and GEO databases, CLEC5A was found to be associated with immune cell infiltration in gynecologic tumors. Further, in OC, CLEC5A expression was positively correlated with M2 macrophages and negatively correlated with plasma cells and Tfh cells (Fig. 7C). We found that the proportion of M2 macrophages was higher in the CLEC5A high expression group than that in the CLEC5A low expression group. In contrast, the plasma cells and Tfh cells were lower in the CLEC5A high expression group (Fig. 8).

The accumulation of M2 macrophages has been reported as a pivotal part of promoting tumor progression(Yang and Zhang, 2017). As seen above, the CLEC5A expression level may affect the function of M2 tumor-associated macrophages (TAMs). We thus analyzed the correlations between CLEC5A expression level and 16 factors involved in TAMs- associated with solid tumors. Following our hypothesis, CLEC5A was found to be positively correlated with inhibiting factors based on the TCGA and GEO databases (Figs. 9A and 9B). These results suggest that an increased CLEC5A expression can promote tumors via TAMs.

To summarize, CLEC5A was strongly correlated with the immune environment in OC and its expression level could affect the type of immune cells that infiltrated the tumor. An example is the significant accumulation of M2 macrophages and increased CLEC5A expression. One possible explanation for this accumulation of M2 macrophages might be that the activated PI3K/Akt signaling pathway could cause the macrophage M2 polarization(Vergadi et al., 2017).

We then attempted to offer patients more personalized chemotherapy drug options based on CLEC5A expression. This entailed the evaluation of the responses of the CLEC5A high expression group and low expression group to five chemotherapeutic drugs that were recommended chemotherapeutic regimens. These were cisplatin, docetaxel, doxorubicin, gemcitabine, and paclitaxel. Interestingly, the high CLEC5A expression group showed increased sensitivity to paclitaxel, gemcitabine, and docetaxel with lower half maximal inhibitory concentration (IC50) values. On the contrary, the low CLEC5A expression group was more sensitive to cisplatin and doxorubicin (Fig. 10).