Introduction
Approximately 75% of ovarian cancer cases are not detected until stage III-IV, leading to a 5 year survival rate of less than 30% [
1]. Although the best timing for surgery has been controversial, no residual disease (R0 resection) following primary debulking surgery (PDS) is recognized as the most potent determinant of clinical prognosis in advanced ovarian cancer (AOC, high-grade serous ovarian cancer (HGSOC) with FIGO stage III or IV) [
2,
3]. Two famous randomized clinical trials have confirmed that neoadjuvant chemotherapy followed by interval debulking surgery (NACT-IDS) has similar progression-free survival and overall survival with fewer surgical complications compared to PDS for AOC patients who cannot achieve R0 resection [
4]. In this context, the preoperative identification of patients with unresectable tumors is of utmost importance to optimize the therapeutic choice between PDS and NACT. For commonly used clinical models, CA-125 has no accurate predictive threshold because the preoperative level cannot fully reflect the tumor progression [
5,
6], and the radiological evaluation is subjective to some extent. Moreover, using the classic Fagotti laparoscopy scoring [
7,
8] as a minimally invasive examination, with a 40.5% unsuccessful prediction operations rate, can lead to a delay in starting chemotherapy and may also facilitate tumor implantation metastasis in puncture sites [
9].
At present, biomarkers of liquid biopsies to predict which AOC patients will potentially profit from PDS or NACT therapy are missing. Small extracellular vesicles (sEVs) are membrane vesicles with a diameter of less than 200 nm, which contain specific cargoes that represent selected portions of the source cell's contents, strongly biasing toward microRNAs (miRNAs) [
10]. Mounting evidence suggests that sEVs-derived miRNAs could be used for cancer detection, and prognosis, and to guide therapy [
11,
12]. Circulating concentrations of sEVs miRNAs vary in response to OC stages [
13,
14], and their stability in stored samples makes them plausible candidates as biomarkers.
In this study, we assessed the diagnostic performance of circulating sEV-miRNAs plus CA-125 to distinguish high-risk AOC patients of residual disease for the purpose of treatment stratification into PDS or NACT-IDS.
Materials and methods
Clinical samples and ethics approval
A total of 221 AOC patients (HGSOC with FIGO stage III or IV) who were treated with PDS or NACT-IDS from January 2018 to June 2022 were recruited in this study (R0, n = 99; non-R0, n = 122). In addition, we obtained plasma samples from benign pelvic diseases (n = 21), early-stage ovarian cancer with FIGO stage I or II (n = 20), and advanced colorectal cancer (n = 22) patients. We also collected primary tumor tissues (n = 58) and other site tissues (n = 18) from these AOC patients after PDS. None of the patients involved had infectious diseases. AOC samples with the following characteristics were removed: (1) treatment with surgical operation or chemotherapy before plasma collection; (2) hemolysis. Each participant signed a written informed consent form. The study obtained ethical approval (No. KYJJ-2019-043) from the Hunan Cancer Hospital Institutional Review Board. The CA-125 level of each AOC patient was measured by chemiluminescence assay (Beckman, DXI800, CA, USA) before treatment at four clinical centers. Blood samples (8 mL) from every individual in a fasting state were collected with an EDTA-K2 anticoagulant tube in the early morning, stored at 4 °C, and then processed within 30 min. The samples were extracted via centrifugation at 2000 ×
g for 10 min, and 13,000 ×
g for 10 min at 4 °C to exclude effects from platelet-derived vesicles as described in refs. [
15,
16]. Then the isolated plasma was aliquoted into 2 mL tubes for storage at − 80 °C. Samples from other centers were also processed as described above and then transported via dry ice.
Isolation and purification of sEVs from plasma
sEVs were purified from 2 mL of plasma from AOC patients by differential ultracentrifugation [
16]. After the plasma sample was thawed on ice, centrifuged at 3000 ×
g for 15 min. The supernatant was carefully pipetted into a new tube and diluted with PBS to 23 mL, and centrifuged at 13,000 ×
g for 30 min (Beckman Coulter, Brea, CA, USA). Through a 0.22 μm filter, the supernatant was ultracentrifuged at 100,000 ×
g for 2 h at 4 °C to collect a pellet of sEVs. The pellet was dissolved with PBS, transferred to a new ultracentrifuge tube and centrifuged again at 100,000 ×
g 4 °C for 2 h to eliminate any contaminants from the protein aggregates. Finally, the enriched pellet of sEVs was resuspended twice with 100 µL PBS and then collected into a new tube. The isolation method of plasma sEVs was performed strictly according to the MISEV2018 guidelines [
17].
Isolation and purification of sEVs from tissue
Primary tumor tissue, matched adjacent tissue, metastatic tumor tissue, and distant normal tissue were collected from AOC patients undergoing PDS. After collecting the living tissue, residual blood was washed with PBS, cut into 500 mg per block, and then placed in a frozen tube. The tissue was quickly transferred into liquid nitrogen flash-frozen for 1 h, and then stored at − 80 °C. Tissue sEVs were separated based on the protocol previously established by Vella et al. [
18]. with some modifications [
19]. Tissue dissociation was performed using the Miltenyi Human Tumor Dissociation Kit (Miltenyi Biotec, No. 130-095-929, Germany). Before extraction, it was resuspended by enzymes H, R, and A, according to the instructions, and the dissociated mixture containing 2.2 mL RPMI, 100 μL enzyme H, 50 μL enzyme R, and 12.5 μL enzyme A was prepared. A small piece of tissue (~ 200 mg) was weighed, and ultrathin sections were taken using a frozen slicer (to minimize foreign contamination caused by tissue cell rupturing, thereby enlarging the surface area). Then, the dissociation mixture prepared above was added and incubated at 37 °C for 15 min to enzymatic dissociation and permeabilize the tissue. Halt protease and phosphatase inhibitor single-use cocktail, EDTA-free (100X) (Thermo Scientific, No. 78443, USA) was added into dissociation solution and gently filtered twice through a 70 μm filter to remove residual tissue.
The mixed suspension was centrifuged at 300 ×g for 10 min at 4 °C. The supernatant was transferred to a new centrifuge tube and centrifuged at 2000 ×g for 10 min at 4 °C. The cell-free supernatant was centrifuged at 10,000 ×g for 20 min at 4 °C and then gently passed through a 0.22 μm filter to remove remaining cell debris. The supernatant underwent additional ultracentrifugation at 150,000 ×g for 2 h at 4 °C. The precipitates were collected and resuspended in 1 mL PBS, and further purified by an Exosupur column (Echobiotech, China). Finally, the sEVs-containing fraction was condensed to 200 μL via a 100 kDa Amicon Ultra ultrafiltration centrifuge tube (Merck, Germany).
Separating sEVs by immunoaffinity magnetic bead sorting system (MACS)
Plasma and tissue sEVs were isolated via the above methods. Then, 20 μL of magnetic microbeads with antibodies (EpCAM, FAP, CD31, CD235a, CD45; Miltenyi Biotec) was added to 100 μL of sEVs suspension. After incubation for 60 min at 4 °C, the magnetic immune mixture was resuspended with 1 mL of PBS. The LD column (Miltenyi Biotec) was placed in a magnetic rack, and the column was washed three times with 2 mL PBS. The magnetic immune mixture was added to the LD column and then washed three times with 1 mL PBS to clean unlabeled sEVs. The LD column was removed and placed in a 15 mL collection tube. PBS (3 mL) was added to the column, and the magnetic mixture containing labeled sEVs was pushed into the tube by a plunger. The mixture was centrifuged at 100,000 ×g for 70 min, and the precipitates were resuspended in 100 μL of PBS. Then, 100 μL of elution buffer (0.1 M glycine pH 2.8) was used to dilute the labeled sEV suspension, which was vortexed for 30 s, and combined with 10 μL of renaturation buffer (1 M Tris–HCl, pH 7.4). The mixture was centrifuged at 10,000 ×g for 30 min to separate the sEVs from magnetic beads. The supernatant containing specific sEVs was stored at − 80 °C for RNA extraction.
Transmission electron microscopy (TEM)
sEVs fixed in 1% paraformaldehyde for 10 min and washed with deionized water. The sEV suspension (10 μL) was placed over a Formvar-carbon-coated 300-mesh copper net and incubated for 10 min at room temperature. After washing with deionized water, the sEVs were negatively stained with 2% uranyl oxalate solution for 1 min at room temperature. The grids were dried for 5 min under incandescent light. Images of sEVs were acquired with an FEI Tecnai G2 Spirit Transmission Electron Microscope (TEM) (FEI; Houston, TX, USA).
Nanoparticle tracking analysis (NTA)
Based on the characteristics of the Brownian motion and light scattering, the hydrodynamic diameter and concentration of sEVs were measured by NTA (ZetaView PWX 110, Particle Metrix, Germany). SEVs were diluted to 1 mL with PBS and injected into the cuvette. The hydrodynamic diameter and concentration of particles were calculated from the diffusion coefficient by the Stokes–Einstein equation. Five videos of approximately 10 s duration each were recorded for every sample, and analysis of the data regarding particle movement was by the nanoparticle tracking analysis (NTA) software.
Western blotting
Protein quantification of sEVs samples was conducted using the BCA Protein Assay Kit (Thermo Fisher Scientific, No. 23,225, USA). The primary antibodies used were CD9 (Abcam, ab92726), CD63 (Abcam, ab216130), CD81 (Abcam, ab109201), TSG101 (Abcam, ab125011), HSP70 (Abcam, ab2787), calnexin (Abcam, ab22595), EpCAM (Abcam, ab32392), FAP (Abcam, ab207178), CD31 (Abcam, ab9498), CD45 (Abcam, ab40763), and β-actin (Proteintech, 66,009–-1-lg). For secondary antibodies, goat anti-rabbit (A0208, Beyotime) and goat anti-mouse (A0216, Beyotime) IgG horseradish peroxidase were used.
High-throughput sequencing and differential expression analysis
Total RNA extracted from sEVs of 2 mL plasma was used for miRNA library preparation (Ribobio, China). The clean reads (17–45 nt) were contrasted with human genome databases (Silva, GtRNAdb, Rfam, and Repbase) using the Bowtie software [
20]. The clean reads were further compared with mature miRNAs in the miRDeep2 and miRBase databases. Next, the miRNA expression levels were normalized by the RPM. The significant differential expression analysis of miRNAs was conducted using the edgeR and limma packages.
RNase treatment of sEVs and RNA extraction
The plasma from the same patient was divided into 3 equal parts. One of which extracted sEV fractions was treated with RNase A (10 μg/ML, Tiangen, No. RT405, China), which is used to eliminate the free RNA carried by non-vesicles, for 15 min at 37 °C. The remaining two plasma samples were not digested with RNase, one of which was directly used to extract plasma RNA, and the other was used to extract RNA from sEVs.
According to the standard protocol, total RNA from plasma or sEVs was extracted and purified by the miRNeasy Serum/Plasma Advanced Kit (Qiagen, No. 217,204, USA). Total RNA from tissue was extracted via a miRNeasy Mini Kit (Qiagen, No. 217004, USA). RNA degradation and contamination were detected by gel electrophoresis. The integrity of RNA was assessed on an Agilent 2200 TapeStation (Agilent Technologies, CA, USA). Additionally, RNA samples were quantified by Qubit®2.0 (Life Technologies, USA).
Quantitative real-time PCR (qRT-PCR)
C. elegans cel-miR-39 was used as an external calibration, and U6 was used as an internal reference for tissue and tissue sEV samples. miRNA quantification was performed by a Light Cycler 480 (Roche, Germany) using specific miRNA TaqMan gene expression probes (Synbio Technologies, China) mixed with cDNA templates (Takara, RR037A, Japan). The relative expression levels of candidate miRNAs were normalized to the control group via the Eq. 2-ΔΔCt [
21]. The primer and probe sequences of the miRNAs were listed in Additional file
1: Table S3.
Pathway enrichment analysis
Target genes of miRNAs were screened via TargetScan and miRWalk. The Tothill dataset (GSE61568) [
22] was used to establish the miRNA-gene regulatory network and perform pathway analysis. Tumor cases with low malignant potential, non-serous histology, or low grade from the Tothill dataset were excluded. Samples that received NACT or did not provide residual disease status were excluded. Accordingly, miRNA target genes were matched using only data from primary tumors of patients with HGSOC undergoing surgery in the Tothill dataset. The DAVID website (
https://david.ncifcrf.gov.) was used to conduct functional clustering and enrichment pathway analysis. We classified the GO and KEGG pathways using the two-sided Fisher's exact test and calculated FDR to correct P values. A corrected P value < 0.05 was considered to be significantly enriched in the GO/KEGG pathway. GSEA was performed by GSEA software (
http://software.broadinstitute.org/gsea/).
Statistical analysis and prediction model building
LASSO analysis was performed to select the most relevant predictors via the "glmnet" (strictly set alpha = 1) package. Based on the R0 group as a control, we calculated the relative expression abundance of selected miRNAs in the non-R0 and differential diagnostic groups (BPD, ESOC, ACC). Log2 transformation of 4-miRNA qRT-PCR expression values of all samples was completed using the log (x + 1, 2) formula [
11,
23]. Serum CA-125 expression values were converted via log10. We preliminarily fitted the model using logistic regression, as described in refs. [
11,
24]. Based on this multivariate logistic regression model, we calculated the risk probability of residual disease for each subject through leave-one-out cross-validation. Moreover, the index score of differential diagnostic patients was also calculated by this risk model. We assessed the optimal cutoff value of the above prediction probability (4-miRNA combined CA-125) by receiver operating characteristic (ROC) curves and calculated sensitivity, specificity, AUC, PPV, and NPV in the discovery set [
23]. While using the same cutoff value of risk prediction, the corresponding sensitivity, specificity, AUC, PPV, and NPV were calculated by the ROC curve in the validation set. The NRI and IDI were calculated by R software to quantify the improvement of diagnostic separation [
25]. Furthermore, DCA was conducted by the “rmda” package in R. Gpower software was used to calculate the post hoc power of the prediction model. Multivariate and univariate logistic regression were employed to analyze the influence of various clinicopathological variables on residual disease status, including age, stage, lymph node metastasis, TP53 mutation, serum CA-125 levels, and 4-miRNA panel.
For all cohorts, MedCalc statistical software, version 19.1.0 (Medcalc Software, Ostend, Belgium) was used for the calculation and visualization of ROC curve relative indicators [
24]. Statistical analysis was carried out using SPSS, GraphPad Prism 8, and R software (version 3.6.3). Comparisons between groups were performed using an unpaired t test or Mann–Whitney U test. The limit of significance was defined as a P value < 0.05. The results were visualized via the R packages VennDiagram, pheatmap, GOplot, and ggplot2.
Data availability statement
Discussion
No residual disease after debulking surgery is the most critical independent prognostic factor for AOC. There is an unmet clinical need for selecting primary or interval debulking surgery using existing prediction models. In this study, we discover a distinct sEVs miRNAs profile between R0 and non-R0 AOC patient. And our aim is to find circulating sEVs miRNAs signature with the most stable expressed commonality among different AOC patients to exclude the influence of individual differences. Therefore, we have enrolled 360 patients from four clinical centers in this study. Based on it, we develop and validate a reliable and stable model of circulating sEVs miRNA panel combined with CA-125 for predicting residual disease risk in AOC patients for the first time. And this combined model can also discriminate diseases easily confounded with AOC (e. g., advanced colorectal cancer). On the other hand, Taqman qRT-PCR detection of the prediction panel is a non-invasive, safe, and easy to perform method, which is helpful for clinical application. Our prediction model can be part of a clinical standard monitoring strategy for screening high-risk AOC patients who are allowed for neoadjuvant chemotherapy instead of PDS.
NACT-IDS is the preferred alternative strategy for 30% of AOC patients with a higher disease burden who have entirely no chance for R0 resection [
39]. The platinum-based chemotherapy response rate for HGSOC is as high as 80% [
40]. Taylor et al. discovered that circulating tumor exosomal miRNA signatures paralleled and exhibited a strong correlation with tumor tissue-derived miRNA profiles (ranging from 0.71 to 0.90) [
13]. Takahiro et al. performed the first large-scale comprehensive analysis of circulating miRNAs for the early detection of OC [
14]. Their data showed that miRNA levels in early-stage OC were not significantly different from those of borderline and benign tumors, whereas miRNA levels changed dramatically in AOC. Their work revealed that circulating miRNAs could more clearly reflect AOC progression. In terms of exploring residual disease-related biomarkers, Anil K. Sood [
41] found that FABP4 and ADH1B were highly expressed in the tumor tissue of non-R0 patients. Although this attempt used objective biomarkers, its limitations included not only the difficulty of obtaining tumor tissue before the operation but also the difference in gene expression rate between metastatic and primary tumor sites. There is currently no consensus on the choice of a specific circulating sEVs-derived miRNA signature to non-invasively predict residual disease in AOC.
Therefore, we performed a systemic and comprehensive analysis of the circulating sEVs miRNAs landscape in R0 and non-R0 subjects. Circulating miRNAs previously identified as diagnostic biomarkers for OC (miR-92, miR-21, miR-221, miR-200c, miR-182) were also highly expressed in our plasma sEVs samples, underlining the validity of our analysis [
42]. To maximize the success rate of verification, we narrowed the scope of candidate miRNAs by the following characteristics: (1) high expression abundance; (2) conserved; and (3) targeting HGSOC closely related drive genes. Moreover, patients recruited from four clinical centers enhanced the model’s generalization.
FIGO stage and tumor burden were positively correlated with the residual disease risk of AOC patients [
43]. In addition, oncogenic perturbations of tumor cells are involved in the alteration of specific miRNA species’ content in sEVs, which possibly drives cancer progression and generate novel classes of clinical biomarkers [
44]. Our data showed that EOCCs-derived sEVs captured on EpCAM antibody-coated magnetic beads were the major vehicles affecting circulating 4-miRNA expressions. The model index score was positively correlated with the tumor burden of AOC patients. Although we used a novel and improved experimental method, some sEVs were still lost in the sorting process due to existing technical difficulties. Furthermore, according to the expression of miR-1307-3p and let-7d-3p in Fig.
6E, CAFs might also have some influence on panel-miRNA expression. We could not completely preclude that other cells in the tumor microenvironment or other tissues might express circulating sEVs’ 4-miRNA. During the processing of plasma samples, anticoagulants, storage temperature, and centrifugation time were strictly controlled to avoid the effects of microvesicles derived from platelet, and hemolyzed samples were removed.
A series of studies reported that circulating miR-320a could also be a potential diagnostic factor in OC [
45], prostate cancer [
46], and hepatocellular carcinoma [
47]. Circulating miR-1307-3p is considered as a diagnostic biomarker of OC [
48], breast cancer [
49], and gastric cancer [
50]. Circulating let-7d often emerged as a diagnostic molecular that is down-regulated in pancreatic cancer [
51], hepatocellular carcinoma [
52], and cervical cancer [
53]. Meanwhile, there have been relatively fewer studies on miR-378a-3p as a cancer biomarker. To pin down the mechanisms and pathogenic relevance of 4-miRNA, further studies will be needed in the future.
CA-125 is dramatically elevated in BPDs such as tuberculous peritonitis and theca follicular fibroma [
54]. In addition, ACC with extensive pelvis metastasis is also prone to confusion with AOC; colonoscopy is normally needed to differentiate the two. However, our model could distinguish AOC from these diseases non-invasively (Fig.
6B). Furthermore, a key prerequisite for the molecular signature with a potential clinical application is a fast, robust, and easy-to-perform laboratory assay. Our TaqMan qRT-PCR validation of the circulating sEVs miRNA panel is a vital step in this direction [
11,
55]. This testing panel, as an affordable approach, is expected to provide benefits for AOC patients in developing countries.
Conclusions
Overall, for the first time, we established a reliable and objective model composed of circulating tumor-derived sEVs 4-miRNA and CA-125 for predicting residual disease risk in AOC patients, which could accurately assess the triage of PDS versus NACT-IDS. This model, as a tool for non-invasive liquid biopsy, has great potential to be part of clinical monitoring strategies for screening high-risk patients with residual disease.
Acknowledgements
We thank all study participants and coordinators from the Hunan Cancer Hospital / The Affiliated Cancer Hospital of Xiangya School of Medicine, Central South University, The Central Hospital of Shaoyang, The First People's Hospital of Huaihua / The Affiliated Huaihua Hospital of University of South China, The First People's Hospital of Changde, Northwestern University Feinberg School of Medicine and MD Anderson Cancer Center in the USA, and University College London Hospital in the UK; and thank the nurses as well as colleagues from Gynaecological oncology, and our collaborator team members. We are grateful for the experimental support from the Central Laboratory of Hunan Cancer Hospital and the Cancer Invasion and Metastasis Laboratory of the Cancer Research Institute of Xiangya School of Medicine, Central South University; We particularly thank all patients who participated in the study.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.