Potential clinical utility of liquid biopsies in ovarian cancer

CTCs are cancer cells found in peripheral blood that intravasate or are passively shed from a primary or metastatic solid tumour site. Several analytic methods for CTC isolation have been developed and validated for ovarian cancer that are based on various biological (i.e., positive epithelial markers, negative hematopoietic markers) or physical properties (i.e., size, density, deformability, electric charges, and invasive capacity) [14‐29]. The ability to detect CTCs in the bloodstream has important prognostic implications in ovarian cancer for identifying potential micrometastasis, pre-neoplastic lesions, tumour heterogeneity, and tumour evolution over time [30‐33].

Isolation of CTCs from peripheral blood samples is technically challenging given the low concentration with approximately 1 CTC in 1,000,000 circulating cells [34‐36]. Following release from the primary tumour, CTCs overcome several obstacles to survive in the systemic vasculature and spread to distant organs [37]. First, tumour cells shed from solid tumours often traverse the endothelium to enter the circulation by undergoing the epithelial-to-mesenchymal transition (EMT) process. EMT is a phenotypic transformation of epithelial cells with loss of polarity, morphology, and cell markers such as the epithelial cell adhesion molecule (EpCAM), to gain the migratory and invasive properties of mesenchymal cells [38]. After dissociation from a primary model V₂ carcinoma (established from skin carcinoma of cottontail rabbits) site of 1 cm, approximately one million CTCs intravasate via dermal invasion into the peripheral circulation each day, of which < 1% typically remain viable for metastasis [39]. Most CTCs undergo apoptosis or necrosis due to the profound environmental challenges in the bloodstream such as starvation, shear stress, and immunological detection [40]. Only a small proportion of CTCs can survive through upregulation of several signalling pathways, including increased secretion of growth factors, downregulation of death receptors, and over-expression of anti-apoptotic ligands [39]. CTCs must also evade natural immune system defences and natural killer (NK) cell recognition. According to the adaptive immune resistance theory, tumour cells avoid activation of the antitumor response from NK cells and T cells by upregulating the naturally occurring programmed death-ligand 1 (PD-L1). They also avoid phagocytosis by macrophages through the upregulation of CD47 [41‐43].

Currently, the CellSearch detection system is the most widely used isolation strategy with US Food and Drug Administration (FDA) approval. CellSearch uses an immunoaffinity-based isolation strategy to identify CTCs based on positive EpCAM expression [44]. However, the application of CellSearch in ovarian cancer may be limited in the setting of low EpCAM expression. In a study with newly diagnosed or recurrent ovarian cancer patients, Liu et al. found no correlation between serial CTC enumeration by the CellSearch system and clinical outcomes [15]. One possible explanation is the downregulation of EpCAM during EMT or the heterogeneous expression of cell surface markers in ovarian cancer [15]. Obermayr et al. also reported EpCAM expression in a small proportion of EOC patients [21]. The researchers used RT-qPCR to analyze EpCAM expression of CTCs isolated using density gradient centrifugation from 216 EOC patients before and after primary treatment compared to 39 healthy controls. The researchers reported EpCAM expression in only 8% of patients at baseline before treatment and 4% for patients after six months of adjuvant chemotherapy.

To overcome these limitations, alternative approaches targeting various biophysical properties of CTCs such as cell size and invasive potential have been developed [14, 29]. For example, Yang et al. recently used a technique called CanPatrol enrichment, in which CTCs were filtered through 8-μm porous membranes and detected using RNA in situ hybridization (RNA-ISH). CTC subpopulations were identified using epithelial (EpCAM and CK8/18/19), mesenchymal (vimentin and Twist), and epithelial-mesenchymal hybrid markers [14]. The researchers used hybrid markers and found a mean CTC count of 8.70 ± 5.69 detected in 5 mL of blood among stage 1-IV EOC patients that was significantly higher compared to controls with benign gynecologic diseases (1.04 ± 0.73). Multivariate analyses demonstrated both higher CTC counts and higher percentage of mesenchymal CTC were independent prognostic factors for significantly lower OS (p = 0.012 and p = 0.009 respectively). Fan et al. proposed another novel enrichment method utilizing the unique property that blood containing CTCs will invade and ingest Cell Adhesion Matrix (CAM) while non-tumour and dead tumour cells do not [29]. The researchers used a cell invasion assay that enriches and identifies tumour cells based on CAM invasion (CAM +) and expression of standard epithelial markers (Epi +) to analyze peripheral blood samples of 71 suspected ovarian cancer patients. The study found a significantly higher mean CTC count in stage III/IV patients at 41.3 CTCs/ml compared to 6.0 CTCs/ml in stage I/II patients and 0 CTCs/ml in benign patients (p-value = 0.001). Kaplan Meier analysis showed a significantly lower disease-free survival in patients with detectable CTCs with a median survival of 15.0 months compared to 35.0 months in those without detected CTCs (p= 0.042). Other novel techniques include modified immunoaffinity-based strategies targeting several ligands at once (e.g. EpCAM, folate receptor alpha, Human epidermal growth factor receptor 2) and nanoparticles conjugated with the antibody against Mucin 1 (MUC1) [45, 46]. Given the rarity of CTCs in peripheral circulation despite their prognostication potential, further studies are required to optimize the detection and isolation of CTCs in ovarian cancer.

Normally, plasma contains cfDNA that is passively released from necrotic or apoptotic cells, while ctDNA is the cfDNA secreted from cancer cells. In healthy individuals, cfDNA concentrations are elevated following tissue damage such as intense exercise, inflammation, sepsis, surgery, radiotherapy, trauma, or during pregnancy [47‐49].

Compared to CTCs, cfDNA concentrations are higher in blood, making them suitable targets for liquid biopsy [50]. Tumours harbour unique somatic genetic alterations that help in distinguishing ctDNA from noncancerous cfDNA [47, 48]. The majority of cfDNA is expected to originate from healthy cells, while a variable amount of cfDNA (0.01–93%, depending on the tumour size) can originate from cancer cells (ctDNA) [51, 52]. However, a popular hypothesis posits that a large fraction of cfDNA is released from cells in the tumour microenvironment that were destroyed due to hypoxia or the antitumour response [53]. Recent studies confirmed that cfDNA levels in the blood are higher among ovarian cancer patients with an average of 180 ng/mL compared to 30 ng/mL in healthy controls or individuals with benign ovarian pathologies [54‐56]. Therefore, increased amounts of cfDNA may serve as a diagnostic tool for ovarian cancer, while genomic analysis of ctDNA may provide valuable prognostic and predictive information [57, 58].

The mechanism of cfDNA released from cells into the circulation remains unclear, although apoptosis and necrosis are the most widely accepted hypotheses based on cfDNA properties. Previous studies have estimated the size of cfDNA to vary from ~ 40–200 base pairs (bp), with a peak at around 166 bp [53, 59‐61]. Agarose gel electrophoresis to separate extracted cfDNA has found fragment ladders ranging from 160 bp up to 21 kbp [61, 62]. The size of these fragments corresponds primarily to mono- and oligonucleosomes that are characteristic of caspase-dependent cleavage during apoptosis [59, 61‐63]. In contrast, DNA fragments larger than 10 kb are thought to be a result of necrotic cell death in tumours with different kinetics and amount of cfDNA released from different necrosis-inducing agents [64‐66]. However, this theory has been called into question following studies reporting that radiation therapy, which typically induces tissue necrosis, results in a reduction of cfDNA levels by up to 90% in the plasma of cancer patients [67, 68]. Other proposed cfDNA release mechanisms include active secretion in living cells with the expulsion of nuclei, phagocytosis, neutrophil extracellular trap release (NETosis), and excision repair [69‐74].

Once released into the bloodstream, the size, integrity, and half-life of cfDNA have important clinical implications in diagnosis and tumour detection. One challenge currently is the small amount of ctDNA in the blood compared to cfDNA released from normal cells, particularly when the tumour size is small. Since a significant proportion of ctDNA is released from necrotic cancer cells, the cfDNA size in cancer patients is generally longer than those of healthy individuals. However, the length of ctDNA released from apoptotic cancer cells is shorter than cfDNA released from the normal cells due to apoptosis, with a mean of 133–144 bp [75]. ctDNA enrichment may therefore be possible based on a size selection approach. Selecting shorter DNA fragments between 90–150 bp improved the detection of ctDNA with up to 11-fold enrichment of mutation allele fraction [61, 75, 76]. In addition, the distribution of differently sized DNA fragments has implications for disease staging as an indicator of cfDNA Integrity (cfDI). The cfDI is defined as the ratio of long (released from necrotic cells) to short (released from apoptotic cells) cfDNA fragments. cfDI is calculated by measuring long and small ALU sequence fragments (ALU₂₄₇ and ALU₁₁₅respectively) using qPCR [77]. Studies have shown that cancer patients have a higher cfDI compared to healthy controls or individuals with benign disease [78, 79]. Higher integrity is associated with increased levels of necrotic cell death in advanced disease with larger and more aggressive tumours [78, 80].

The short half-life of cfDNA present in the bloodstream allows for real-time analyses of the tumour mutational profile. The level of cfDNA in circulation at any given time is determined by the net amount of DNA released minus DNA clearance. cfDNA clearance may occur in organs including the liver, spleen, kidney, or lymph nodes [81]. In the bloodstream, circulating enzymes such as DNase I, plasma factor VII–activating protease (FSAP), and factor H are responsible for cfDNA breakdown [82, 83]. Rapid clearance of apoptotic cells and cfDNA normally allow for healthy individuals to have low levels of cfDNA. In cancer patients, cfDNA accumulates due to impaired clearance that is currently poorly understood. Using fetal DNA in postpartum maternal circulation, Lo et al. estimated the half-life of cfDNA to be approximately 4 to 30 min, which has been consistent across other studies [84‐87]. However, the half-life of cfDNA may vary depending on several factors, including interactions with molecular complexes that interfere with cfDNA degradation, tumour stage and subtype, and treatment [69, 81]. Interestingly, one study used next-generation sequencing (NGS) technology to examine the kinetics of cfDNA and found that the clearance of cfDNA may occur in a bi-phasic manner. The first rapid phase has a mean half-life of an hour, followed by a second slow phase with a mean half-life of 13 h [88].

Several technologies have been developed for ctDNA detection in blood, including quantitative PCR, digital droplet PCR (ddPCR), and NGS for targeted sequencing or whole-genome sequencing (WGS). In addition to quantitative changes, these technologies detect qualitative changes in ctDNA, which include tumour-specific variants (TSVs), gene fusion, copy number variations, aberrant DNA methylation, and chromosomal instability. The development of NGS and digital polymerase chain reaction (dPCR) has improved the sensitivity and specificity of ctDNA detection. To date, most ctDNA detection methods have focused on high-grade serous ovarian cancer (HGSOC) patients with targeting TP53 mutations [45, 89‐91]. One study used targeted error correction sequencing (TEC-Seq) to examine 58 cancer-related genes encompassing 81 kb and reported the highest sensitivity and specificity at 75–100% and > 80%, respectively [89]. In stage I-II disease, the highest detection rate was 68% with a specificity of 100% that was achieved using TEC-Seq and ddPCR combined. The high specificity achieved in this study may be attributable to TEC-Seq advantages for using deep sequencing for more direct evaluation of sequence changes. In fact, deep sequencing using random unique molecular barcodes annealed to each DNA template fragment has been the preferable method for detecting low-level signatures of TSVs in liquid biopsies [92]. Duplex sequencing using molecular barcodes on both DNA strands for removing sequencing errors that are in one strand only has improved variant detection accuracy by > 10,000 times compared to conventional NGS [93, 94].

Although ctDNA analysis with plasma samples is currently the preferred method, alternative approaches have utilized different sources. In 2013, Kinde et al. examined the ability of the liquid Pap test with uterine cervix sampling to detect ovarian and uterine cancers. The researchers used massively parallel sequencing for TSVs using a 12-gene panel and found ctDNA in 41% (9 of 22) of ovarian cancer patients [95]. Similarly in 2018, Wang et al. analyzed Pap brush samples from 245 ovarian cancer patients using PapSEEK with an assay for mutation in 18 genes and reported a limited detection sensitivity of 33%, including 34% for patients with stage I–II disease [96]. Likewise, Maritschnegg et al. conducted a study with uterine cavity lavage samples from EOC patients and benign gynecologic patients [97]. The researchers used NGS for sequencing AKT1, APC, BRAF, CDKN2A, CTNNB1, EGFR, FBXW7, FGFR2, KRAS, NRAS, PIK3CA, PIK3R1, POLE, PPP2R1A, PTEN, and TP53 genes to analyze lavage samples. Using NGS, the researchers reported detectable mutations, mainly in TP53, in 60% of ovarian cancer patients and an improved ctDNA detection rate of up to 80% with more sensitive methods of digital droplet polymerase chain reaction (ddPCR) and the Safe-sequencing system (SafeSeqS). Interestingly, the study also found TP53 mutations in lavage samples of all 5 patients with stage IA disease. Building on these results, the same team conducted a study in 2018 demonstrating the feasibility of this technique that found a median absolute amount of 2.23 μg cfDNA in uterine and tubal lavage samples and TSVs using deep-sequencing in 80% (24 of 30) of ovarian cancer patients [27]. Given these findings, the molecular analysis of uterine lavage samples may be a potential technique for the early diagnosis of ovarian cancer. Other novel techniques including peritoneal washing, urine sampling, and vaginal sampling have been utilized for ctDNA profiling. However, such methods require more research to elucidate their diagnostic utility in ovarian cancer [98‐101].

The rapid turnover of tumours results in high gene transcription and shedding of high amounts of cfRNA consisting of mRNA and microRNA (miRNA) into the circulation [102]. Normal and tumour cells secrete miRNAs into various body fluids, including plasma, urine and vaginal discharge, and breast milk [103]. In the blood, mRNA and miRNA are bound to specific ribonucleoprotein complexes, high-density lipoproteins, platelets, or packaged in extracellular vesicles (EV) such as exosomes to avoid degradation and acquire more stability [103, 104]. Several studies have suggested the role of miRNAs in tumorigenesis, cell differentiation, proliferation, inhibition of angiogenesis, metastasis, and apoptosis. Importantly, the biogenesis and activation of miRNAs are faster with longer half-lives compared to mRNA and proteins, which may make miRNAs more suitable for earlier diagnosis of ovarian cancer [105‐108].

The diagnostic, prognostic, and therapeutic potential of circulating miRNAs in ovarian cancer have been explored in many studies. In 2008, Taylor et al. first reported that higher levels of 8 exosomal miRNAs (miR-21/141/200a/200b/200c/203/205/214) were found in the serum of ovarian patients compared to healthy controls, although there was no significant difference in early versus late-stage ovarian cancer [109]. These findings were subsequently supported by several other studies reporting that serum miRNAs (miRNA-141/200a/200b/200c) were upregulated in ovarian cancer patients compared to normal or benign tumour controls [110, 111]. Gao et al. also found that different miRNA-200c expression levels may correlate with ovarian cancer staging, with more advanced tumours having lower miRNA-200c levels and higher miRNA-141 [110]. However, Kim S. et al. analyzed seven serum exosomal miRNAs and concluded that the expression of miRNA-141, 200a, and 200b were too low to be an appropriate serologic biomarker [112].

Although miRNA-145 was identified as the best-performing single marker with a sensitivity of 91.7% and accuracy of 86.8%, similar changes in miRNA-145 levels were observed in other malignancies besides ovarian cancer [112, 113]. The lack of discrimination between cancer types suggests that single miRNAs are unlikely to be a reliable biomarker. To overcome these challenges, a recent study by Elias et al. was the first to combine NGS analysis of serum circulating miRNA with a machine learning technique called a neural network model and developed a diagnostic algorithm for EOC. The study authors reported an AUC value of 0.90 for this model, which was significantly higher compared to CA-125. This study suggests the potential for a new era of machine-learning application in biomarker discovery [114].

In 2017, Yokoi et al. performed miRNA sequencing to identify the optimal combination of candidate circulating miRNAs for the early detection of ovarian cancer [115]. This study identified eight miRNAs with RT-qPCR validation and statistical cross-validation with a large research cohort. The predictive model using a combination of 8 circulating serum miRNAs was able to differentiate early-stage ovarian cancer from benign tumours with 86% sensitivity and 83% specificity, and from healthy controls with 92% sensitivity and 91% specificity [115]. In a later study, the same research team analyzed 4,046 serum samples from 333 ovarian patients, 95 benign or borderline ovarian tumours, 2,759 healthy controls and 859 other solid cancers using miRNA microarray [116]. The study found that combined miRNAs can successfully discriminate ovarian from lung, gastric, breast, hepatic, colorectal, and pancreatic cancers, but not sarcoma or esophageal cancer. In this study, utilization of circulating miRNA yielded a sensitivity of 99% and a specificity of 100% for discriminating between ovarian cancer and healthy controls. This was the first large-scale comprehensive study examining circulating miRNAs in ovarian cancer and reported promising miRNA combinations for the detection of early-stage disease.

In addition to miRNA, circular RNAs (circRNAs) and long non-coding RNAs (lncRNAs) also demonstrated potential utility as biomarkers for liquid biopsy in ovarian cancer. circRNAs have a covalently closed loop structure and lncRNAs have transcript sizes of > 200 nucleotides, which allow increased stability and resistance against RNase degradation in the peripheral circulation. circRNAs are abundant and diverse, with a half-life > 48 h, that facilitate easier detection [117‐119]. circRNA expression differs between primary and metastatic sites and is thought to play a role in regulation of ovarian cancer. A recent study found that the expression levels of circular RNAs are inversely associated with activating many signalling pathways involved in tumour metastasis (i.e., NF-κB, PI3k, AKT, and TGF-β) [120]. Using RT-qPCR in a sample of 83 EOC patients compared to 166 benign or healthy controls, Hu and colleagues found that CircBNC2 was associated with histological grade, serous subtype, and distant metastasis [121]. Similarly, lncRNAs were found to contribute to the early pathogenesis, progression, metastasis and chemoresistance of recurrent ovarian cancer [122‐124]. Although there is emerging evidence suggesting an association between differing expression levels of lncRNAs (H19, LSINCT5, XIST, CCAT2, HOTAIR, AB073614, and ANRIL) and clinical progression or treatment response of ovarian cancer, the diagnostic sensitivity and specificity of lncRNAs remain to be fully elucidated [125‐130]. To date, no lncRNA has been approved for clinical utility and further research is required to identify the most clinically relevant candidates with cancer-enriched or specific signatures in ovarian cancer.

Tumour-educated platelets (TEPs) play an important role in local and systemic responses to tumour growth. Platelets are normally anucleate, although they may contain residual mRNA and miRNA derived from their megakaryocyte precursors or captured from intercellular interactions in the circulation. Platelet education denotes the transfer and sequestration of biomolecules from tumour cells into platelets [131, 132]. External factors in the tumour microenvironment such as stromal and immune cell signals may activate platelet surface receptors to induce specific splice events of pre-messenger RNAs (pre-mRNAs) in circulating platelets [133, 134]. Key advantages of TEPs include their high abundance, easy isolation, and high-quality RNA that may be processed according to external signals. Therefore, TEPs have a dynamic mRNA repertoire with both specific splice events in response to external signals and direct ingestion of spliced circulating mRNA that may provide useful diagnostic information in ovarian cancer. Best et al. first studied the diagnostic potential of TEPs by mRNA sequencing in patients with various cancers [135]. This study found that TEPs were able to distinguish cancer patients from healthy controls with a high accuracy of up to 96% and detect the primary tumour location with 71% accuracy. Later, Piek et al. concluded that TEPs can differentiate early stage ovarian cancer from benign pathologies with 80% accuracy [136]. An ongoing clinical trial (NCT04022863) may further build upon these results by examining the accuracy of TEPs and ctDNA in determining the nature of ovarian tumours and provide information on its diagnostic potential [137]. Interestingly, a recent retrospective cohort study by Giannakeas et al. examined the association between thrombocytosis (platelet count greater than 450 × 109/L) and cancer. By studying 53,339 adults aged 40–75 years who developed thrombocytosis with normal platelet count in the previous 2 years and no malignancy history, they estimated the risk of cancer in a 10-year follow-up period [138]. The authors reported that the 2-year relative risk (RR) was highest for ovarian cancer (RR = 7.11; 95% CI, 5.59–9.03), while the 6-month RR for developing ovarian cancer was even higher (RR = 23.33; 95% CI, 15.73–34.61). In the future, TEPs profiling with complementary ctDNA/CTC analysis and platelets quantification may potentially be a blood-based method for cancer diagnostics.

Interest in the diagnostic and prognostic potential of exosomes has increased in recent years. Exosomes are extracellular vesicles (EVs) typically 30–100 nm in diameter. Such vesicles are extremely stable and can survive under extreme conditions. Exosomes are released from both normal and tumour cells. Likewise, they are found in various body fluids, such as saliva, plasma, urine, ascites, and cerebrospinal fluid [139]. Exosomes can participate in local and distant signalling by fusing with the membrane of the recipient cell or attaching to receptors on the cell’s surface. In cancer, exosomes have the ability to enhance tumorigenesis [140], help tumour cells escape the immune system [141], and cause treatment resistance [142]. Likewise, exosomes can enter the circulation and increase the likelihood of metastasis by preparing distant tumour microenvironments [143, 144]. Exosomes have been also used as therapeutics to successfully eliminate tumour cells [145]. Furthermore, exosomes contain tumour-specific proteins, lipids, DNA, and RNA making them potential diagnostic biomarkers in cancer. For example, exosomes with heat shock protein (HSP70) expressed on their membrane, are observed more in ovarian, breast, and lung cancer samples compared to healthy controls [146]. Additionally, studies have shown increased total exosome concentrations in serum samples of EOC patients [109, 111].

Exosomes can carry significant quantities of miRNAs. Multiple studies have observed differences between the miRNA profiles of exosomes in EOC patients and healthy controls. Meng et al. showed that the concentrations of miR-200b and miR-200c are higher in exosomes obtained from patients with stage III–IV EOC and are associated with significantly shorter OS [111]. Another study showed that the miRNA profiling of circulating exosomes using a modified magnetic-activated cell sorting (MACS) technique can differentiate between benign and malignant ovarian tumours [109]. Additionally, exosomes derived from EOC patients have higher concentrations of TGFB1 and melanoma-associated antigen 3 (MAGE3) and MAGE6 [147]. EOC exosomes also have a higher concentration of Claudin 4 that is associated with tumour stage and CA125 levels [148]. CD24 and EpCAM were also shown to be elevated in exosomes isolated from EOC plasma samples [149]. Furthermore, Liang et al. identified 2,230 proteins in exosomes secreted from OVCAR-3 and IGROV1 ovarian cancer cell lines. Many of these identified proteins were involved in tumorigenesis and metastasis, indicating the prognostic potential of exosomal profiling [150]. Overall, exosomal profiling can act as a cancer-specific diagnostic and prognostic biomarker and replace invasive cell biopsies. However, more comprehensive clinical studies are required to determine the clinical value of this approach.

The early detection of ovarian cancer is critical in reducing mortality and morbidity. Staging is the most important prognostic factor in determining recurrence-free survival (RFS), with stage I or II diseases having significantly longer RFS and OS compared to more advanced stages [151]. Therefore, biomarkers that allow for the diagnosis of OC in stages I-II may extend survival and improve patient outcomes.

Two-thirds of EOC cases are diagnosed at advanced stages, with a significantly worse prognosis [152]. Interestingly, several studies have suggested the diagnostic potential of promoter methylation that leads to epigenetic inactivation of tumour suppressor genes as an early event during ovarian cancer pathogenesis [153‐160]. Analysis of ctDNA methylation status in pre-operative plasma samples from ovarian cancer patients with both early (stage I-II) and advanced-stage disease (stage III-IV) demonstrated a significant association with abnormal methylation of tumour suppressor genes compared to healthy controls [154, 157, 158]. Specifically, the detection of hypermethylation in promoter regions of tumour suppressor genes RUNX3, TFPI2, RASSF1 and RASSF2 from plasma samples has been reported for the diagnosis of ovarian cancer, although these markers are also implicated in other cancers [153‐160]. Despite the potential for identifying aberrant gene promoter methylation to detect malignancies, its utility remains limited by the concentration of extracted ctDNA available for analysis. Earlier stages of EOC are often asymptomatic and are correlated with lower concentrations of ctDNA [159, 160]. There is currently limited evidence regarding the application of liquid biopsy for the early detection of ovarian cancer at pre-symptomatic stages. Only one study by Widschwendter et al. conducted in the early screening setting reported that methylation in a set of epigenetic markers including COL23A1, C2CD4D and WNT6is able to detect EOC up to 2 years before clinical diagnosis with a sensitivity of 23% and specificity of 97% [161]. This study used reduced representation bisulfite sequencing (RRBS) to analyze samples collected from 100,000 asymptomatic women and 43 cases of ovarian cancer. It remains difficult to conduct screening studies for EOC due to the relatively low incidence and large sample sizes required.

Other approach is looking for gene mutations rather than methylation changes. The sensitivity and specificity of mutation detection are important considerations when applying liquid biopsies for the early diagnosis of ovarian cancer. ctDNA has better diagnostic performance compared to traditional CA-125 with several studies demonstrating that quantitative analysis of ctDNA has a relatively high specificity of 88% and sensitivity ranging from 27–100% [92, 162‐168]. A recent systematic review of 23 studies evaluating ctDNA for the diagnosis of EOC in symptomatic patients preoperatively yielded similar results [89, 96, 153, 154, 156‐158, 160‐162, 166, 167, 169‐179]. Other studies that evaluated ctDNA analysis in HGSOC patients for TP53mutation detection, reported high sensitivity (75–100%) and specificity (> 80%) [45, 89‐91]. For stage I-II disease, Phallen et al. observed a sensitivity of 68% and a specificity of 100%. The researchers used TEC-Seq and ddPCR for ctDNA detection and found detectable mutations in driver genes for over two-thirds of stage I-II ovarian cancer cases [89]. More recently, Barbosa et. al used NGS and a custom panel of 27 genes to analyze tumour and matched plasma samples from 96 ovarian cancer patients and detected tumour somatic variants in 75% of patients with stage I disease [180]. Cohen et al. analyzed circulating protein biomarkers and genetic alterations in cfDNA using a commercial blood test called CancerSEEK [177]. This test used 61 amplicons for massively parallel sequencing to increase sensitivity while minimizing any changes in specificity. The authors reported that ctDNA detected in ovarian cancer with a sensitivity of 98% and specificity of 99%, however, the early-stage detection rate was only 38% [177].

CTC is another liquid biopsy component that has been studied for early EOC detection. Zhang et. al used immunomagnetic bead screening, targeting epithelial markers EpCAM, HER2, and MUC1 on ovarian cancer cells, combined with multiplex RT-qPCR analysis of isolated mRNA from CTCs and tumour tissues for detecting CTCs in 109 EOC patients. This study showed that patients with stage IA-IB disease have a much higher CTC positive rate (93%) compared to the CA-125 positive rate (64%) in the same patients [181]. In 2018, Guo et al. prospectively enrolled 61 women with suspicious ovarian masses to investigate the diagnostic value of CTCs. The researchers used microfluidic isolation and immunofluorescent staining to identify and quantify CTCs and determined the sensitivity of CTCs using receiver operating characteristic (ROC) curve analysis to be 73.3% with a specificity of 86.7%, which was superior to CA-125 (sensitivity = 56.7%) [182]. The main challenge facing CTC-based diagnostics is the reduced amount of CTCs detectable in circulation with lower tumour burdens at early disease stages [15, 182, 183]. In contrast, advanced stage tumours release more CTCs that often travel with metastatic colonies to distant sites allowing for easier detection. As such, detection of CTCs is significantly associated with advanced stages (III and IV), where the diagnostic performance of CTCs through immunocytochemistry (ICC) has been reported to have a sensitivity of 76–83% and specificity of 55–95% [21, 25, 181, 184, 185]. Compared to benign controls, early-stage (I-II) and late-stage (III-IV) EOC samples are 8.4 and 16.9 times more likely to have CTCs, respectively. Likewise, significantly lower levels of CTCs are found in stage I patients compared to advanced stages [181]. Current studies on utilizing CTC for early diagnosis of ovarian cancer are limited by small sample sizes and future well-powered studies are warranted to confirm these findings.

Although some studies have suggested the diagnostic potential of miRNA expression profiling and exosome analysis in ovarian cancer, further research is required to determine their diagnostic sensitivity and specificity in early-stage disease. For example, Todeschini et al. analyzed two cohorts consisting of 168 stage III-IV HGSOC patients and 65 healthy controls and demonstrated the clinical potential of miR-1246 as a diagnostic biomarker for HGSOC with a significant increase in the expression of miR-1246 in post-operative serum samples of HGSOC compared to healthy individuals. This study used a novel microarray data normalization to identify candidate diagnostic miRNAs followed by signature validation with RT-qPCR. The study reported a significant over-expression of miR-1246, miR-595, and miR-2278 in HGSOC patients and the highest detection performance for miR-1246, with a diagnostic sensitivity of 87%, specificity of 77%, accuracy of 84%, and AUC of 0.89 [184]. Similarly, the clinical potential of exosomes have also been suggested in ovarian cancer. Zhang et. al analyzed plasma samples from 40 stage III or IV EOC patients versus 40 healthy controls to investigate the role of four exosome proteins including Lipopolysaccharide Binding Protein (LPB), Fibrinogen Gamma Chain (FGG), Fibrinogen Alpha Chain (FGA) and Gelsolin (GSN) as diagnostic biomarkers [186]. This study reported significantly elevated FGA and GSN levels and significantly downregulated FGG and LBP levels in the ovarian cancer group. FGA conferred the highest diagnostic sensitivity among the 4 candidates with an AUC of 0.8459. In another study with 78 EOC patients (63 stage III-IV and 7 stage I-II) and 30 healthy controls, Schwich et al. reported a seven-fold increase in HLA-G levels in plasma circulating exosomes of EOC patients (mean 14.3 ng/mL) compared to healthy controls (1.9 ng/mL) [187]. Therefore, studies to date on miRNA and exosome analysis have been conducted mainly in patients with advanced ovarian cancer, and future research is needed to elucidate their diagnostic utility for early-stage disease. Table 1 summarizes the studies investigating the clinical application of liquid biopsy analytes in ovarian cancer diagnosis.

There is ongoing research examining the clinical application of liquid biopsy to identify microscopic residual disease following primary debulking surgery as a prognostic indicator, predict survival outcomes, and for earlier detection of disease recurrence. Clinically, the implementation of liquid biopsy may aid in selecting individuals at greater risk of relapse for consideration of alternative management approaches and potential inclusion in clinical trials.

The strongest evidence to date on the prognostic utility of liquid biopsy is for ctDNA [188, 189]. Quantitative analysis of ctDNA demonstrate that ctDNA concentrations are positively correlated with advanced stages of EOC and may indicate response to treatment [188‐192]. ctDNA concentrations are more associated with earlier progression and decreased response to treatment than CA-125 or imaging [162‐164, 193]. Pereira et al. used qPCR and targeted sequencing to quantify ctDNA levels immediately following surgery in 22 EOC patients. This study found that undetectable levels of ctDNA at 6 months postoperatively were associated with significantly improved PFS (P = 0.001) and OS (P< 0.05) [163]. Another study examining TP53 mutations in ctDNA of relapsed HGSOC patients highlighted the prognostic ability of TP53, with a less than 60% decrease in TP53 mutant allele fraction after one cycle of chemotherapy associated with poor response and shorter PFS compared to a decrease of more than 60% [164]. Similarly, a recent study with 48 HGSOC patients found that approximately 80% of patients classified as having no surgical residual disease had detectable ctDNA. This study reported that these patients had a higher mortality risk compared to those who did not have detectable post-surgery ctDNA with a 5-year survival rate of 58.3% for those with detectable ctDNA, compared to 85.7% for those with undetectable ctDNA [193]. Although most studies were limited by small sample sizes, all authors concluded that analyzing ctDNA through liquid biopsy has potential as a prognostic biomarker in clinical settings.

Furthermore, there is also evidence supporting the role of ctDNA in detecting recurrent disease. Up to 85% of EOC patients can experience recurrence following first-line therapy. Recurrence of EOC limits the survival of patients and is often referred to as incurable. CA-125 and imaging techniques such as CT and PET-scans are used as traditional recurrence markers [194]. However, recent studies have shown that ctDNA quantification can potentially improve the detection of relapse compared to traditional imaging techniques and CA-125 [188, 190]. Similarly, Parkinson et al. examined TP53 mutations in ctDNA of relapsed HGSOC patients and reported that ctDNA was detected at ≥ 20 amplifiable copies/mL of plasma in nearly all relapsed patients with disease volume > 32 cm³ [164]. Likewise, Minato et al. detected ctDNA in all patients with recurrent EOC using droplet digital PCR, while no ctDNA was detected in recurrence-free patients. In the majority of cases, ctDNA was detected before CA-125 levels indicated recurrence [195]. These results are consistent with the study by Pereira et al. that reported a mean predictive lead time of 7 months for ctDNA over CT imaging for the detection of recurrence [163]. As a result, ctDNA has the potential to be used as an early detection tool for EOC recurrence.

CTCs have also demonstrated prognostic potential in ovarian cancer. In a study by Marth et al., immunobeads coated with MOC-31 antibodies were used to isolate CTCs from blood samples of 90 EOC patients. The authors did not find an association between the detection of CTCs and poor prognosis [18]. Using the CellSearch system, Poveda et al. found that among 216 ovarian cancer patients after primary debulking surgery and diagnosis of recurrence with failed first-line chemotherapy, those with higher CTC levels, defined as ≥ 2 CTCs per 7.5 mL blood, had a 2.06-fold (p = 0.003) higher overall mortality risk and 1.89-fold (p= 0.003) higher risk of progression on doxorubicin treatment [183]. On the contrary, a multivariate analysis conducted by Marth et. al reported no statistically significant correlation between the presence of CTCs in the bloodstream before surgery and PFS or OS. This study reported a mean overall survival of 25 and 28 months for patients with and without detected CTCs, respectively [18]. Similarly, Judson et al. analyzed pre-operative blood samples from 64 EOC patients and reported no significant difference in the OS (p = 0.96) or PFS (p= 0.72) between patients with and without detectable CTCs at a mean follow-up of 18.7 months [196]. Given this controversy, Huang et al. subsequently conducted a meta-analysis including 1285 patients from 2 clinical trials and 13 retrospective studies that demonstrated a significant association between the presence of CTCs before treatment with surgery or chemotherapy and both OS (HR = 1.79, 95% CI:1.43–2.24, p < 0.00001) and PFS (HR = 1.59, 95%CI:1.30–1.94, p < 0.00001) [197]. Other studies have suggested that the number of CTCs may be a potential prognostic factor for EOC patients, but the results are limited by small sample sizes and contradictory findings [21, 181‐183, 198‐201]. For example, some studies highlighted the prognostic role of CTCs, showing that the amount of pre-operative or post-adjuvant chemotherapy CTCs was associated with poor prognosis reflected in both PFS or OS [21, 181, 183, 199‐201]. However, the results of these studies are inconsistent with other investigations that failed to find similar associations [22, 182, 198].

Currently, the prognostic value of other liquid biopsy components including cell-free miRNAs and exosomes lacks enough evidence for clinical applications. Several limiting factors in cell-free miRNAs studies, including the lack of standardized experimental procedures, varied normalization processes, and the inadequately powered sample sizes for statistical analysis have contributed to the controversy in this domain [110, 111, 186, 202, 203]. Several studies have supported the prognostic role of cell-free miRNAs such as miR-200 family, particularly miR-200a, miR-200b, and miR-200c [110, 111, 202, 203]. In another study with 40 EOC patients, Zhang et al. used Western blot analysis and enzyme-linked immunosorbent assay (ELISA) to analyze exosomal protein markers [186]. The study reported high levels of fibrinogen gamma chain (FGG) or lipopolysaccharide binding protein (LBP) mRNA expression were associated with worse prognosis and shorter PFS and OS for patients with EOC (FGG: OS HR = 0.79, P = 0.001 and PFS HR = 0.77, P < 0.001 for LBP: OS HR = 0.81, P = 0.003 and PFS HR = 0.77, P< 0.001) [186]. Since EVs contain several different tumour-derived components, EVs may be a promising all-in-one prognostic biomarker, providing information on the tumour and its microenvironment. However, the absence of a standardized approach for cell-free miRNA and EVs isolation and the small sample sizes of current studies limit the ability to draw definitive conclusions and require future validation in larger cohorts. Table 2 summarizes the studies investigating the clinical application of liquid biopsy analytes as prognostic biomarkers in ovarian cancer.

Although most ovarian cancer patients achieve complete remission after primary debulking surgery and adjuvant chemotherapy, up to 70% of patients develop recurrence due to chemoresistance. Intra-tumour heterogeneity has been proposed as the main cause of drug resistance and treatment failure in ovarian cancer [205]. Intra-tumour heterogeneity refers to the genomic variations within a lesion that arise from tumour cell evolution during the multistep tumorigenesis process. Clonal development from a single malignant cell into a functionally heterogeneous tumour mass is shaped by the tumour microenvironment and adaptation to various external selection pressures (e.g. evasion of apoptosis, self-sufficient cell proliferation, acquisition of replicative immortality). Subclones may evolve and expand in a sequentially linear fashion, or they may follow branched trajectories by continuing to diverge during their evolution trajectory (Fig. 5) [206, 207]. The molecular characterization of all ovarian cancer subclones is crucial for selecting appropriate targeted therapy and identifying acquired resistance in tumour cell clones over time. Liquid biopsy can potentially offer more comprehensive analysis of tumour heterogeneity and allow longitudinal monitoring of tumour evolution over the course of treatment.

The ability of ctDNA to reflect tumour heterogeneity is useful for predicting resistance to platinum-based chemotherapy, in addition to both primary and acquired resistance to PARPi. Approximately half of HGSOC patients have a Homologous Recombination Repair (HRR) deficiency leading to an impaired ability to repair double-stranded DNA breaks and increased sensitivity to the alkylating action of platinum-based chemotherapies that induce DNA damage [208]. BRCA mutated cells also depend on single-stranded DNA repair mechanisms involving Poly ADP-ribose Polymerase (PARP), which results in susceptibility to the synthetic lethality of PARPi [209]. Although ovarian tumours with pathogenic BRCA1/2 variants have demonstrated increased sensitivity to platinum-based chemotherapy and PARPi, the majority will subsequently develop treatment resistance. Mechanisms leading to treatment resistance may be categorized broadly into two types. One of these is the small subclonal tumour cells that do not have the BRCA1/2 mutations, and after starting treatment, become the main clone that does not respond to targeted treatment. The second mechanism is through acquired reversion mutations located close to the initial loss-of-function variant that effectively restore functional protein production [210, 211].

Longitudinal monitoring of BRCA mutation evolution in HGSOC patients during PARPi therapy may be achieved with serial cfDNA sampling. Previous studies have demonstrated that both germline and somatic BRCA reversion mutations can be detected in the cfDNA of patients with HGSOC [212, 213]. In the ARIEL2 trial, Lin et. al assessed the association between BRCA reversion mutations and clinical efficacy of adjuvant PARPi treatment. This study included 112 HGSOC patients with germline or somatic BRCA1/2 mutations and used NGS to analyze plasma cfDNA collected prior to PARPi treatment with rucaparib after disease progression on platinum-based chemotherapy. The researchers found that the absence of reverse BRCA mutations in pre-treatment cfDNA was associated with a longer PFS on rucaparib with a median PFS of 9.0 months, compared to 1.8 months in patients with detectable BRCA reversion mutations in their cfDNA (HR 0.12; p < 0.0001). In addition, this study also sequenced 78 post-progression cfDNA samples to examine acquired resistance and identified an additional 8 patients with novel BRCA reversion mutations not initially found in pre-treatment cfDNA, suggesting the ability of cfDNA to monitor dynamic changes in BRCA mutational status over time [212]. Other studies have reported a similar association between detection of BRCA1/2 reversion mutations in plasma ctDNA and resistance to adjuvant platinum and PARPi therapy (Table 3) [213, 214]. Christie et al. conducted a prospective study with 30 HGSOC patients with a germline BRCA1/2 mutation to determine whether cfDNA analysis could predict chemotherapy response in both the neoadjuvant and adjuvant setting. In this study, paired tumour and plasma samples were collected for two cohorts of patients: one cohort comprised 14 patients receiving treatment prior to primary debulking surgery and the second cohort included 16 patients receiving treatment after recurrence with previous adjuvant platinum-based chemotherapy. The researchers used targeted NGS to first screen tumour samples for germline BRCA1/2 mutation or reversion status before analyzing cfDNA. This study only detected BRCA1/2 reversion mutations in tumour samples of 31.3% (5 of 16) patients in the second cohort with recurrent disease, among which 18.8% (3 of 5 patients) also had detectable reversions in cfDNA. Clinically, all cases with detectable reversions in cfDNA subsequently became resistant to platinum or PARPi therapy. Routine monitoring through liquid biopsy may therefore allow for earlier detection of resistance and selection of more personalized combination therapies (i.e. alternative chemotherapies, targeted therapies or immunotherapy) targeting different oncogenic drivers to reduce the risk of resistance.

Other somatic mutations seen more frequently in ovarian cancer may be used for monitoring response to treatment. As the main driver of clonal pathogenic lesions, TP53 mutations are in nearly all primary ovarian tumour cells, synchronous tumours, and metachronous lesions. Detection of TP53 mutations was therefore hypothesized to be a suitable biomarker for disease monitoring. Since the feasibility of sequencing cfDNA fragments to detect tumour-associated TP53 mutation was first reported more than 15 years ago, studies have demonstrated that TP53 mutation detection can evaluate tumour burden, predict time to progression (TTP) and detect tumour recurrence [58, 91, 162‐165, 189, 229, 230]. Similarly, another study that used whole-exome sequencing of plasma cfDNA to serially quantify allele fractions (AF) in plasma identified an increased representation of mutant alleles associated with the development of treatment resistance in three ovarian cancer patients. The researchers found that increased mutant AF of RB1 and ZEB2 was associated with cisplatin resistance, while MTOR, CES4A and BUB1 mutations were associated with paclitaxel and carboplatin resistance, and PARP8 mutation was associated with liposomal doxorubicin resistance [168].

In terms of other mutations, Forshew et. al reported the emergence of a de novo EGFR tyrosine kinase mutation in an HGSOC patient at disease relapse that was not present 15 months earlier in the initial tumour sample [162]. This patient was identified to be eligible for treatment with gefitinib or erlotinib, however, the treatment efficacy and final clinical outcome were not reported. The study authors used tagged-amplicon deep sequencing (TAm-Seq) and screened 5995 genomic bases for low-frequency mutations to identify cancer mutations present in cfDNA at allele frequencies as low as 2%. More recently, Oikkonen et al. conducted a prospective cohort study with 78 cfDNA samples collected from 12 HGSOC patients collected before, during and after platinum-based chemotherapy, and detected clinically actionable mutations in 7 (58%) of patients. In one patient with ERBB2 amplification detected in ctDNA following the development of resistance to platinum chemotherapy, treatment was changed to trastuzumab with reduced dose carboplatin and dose-dense paclitaxel, with significant tumour shrinkage and complete normalization of CA-125 achieved [224]. Table 3 summarizes the studies investigating the clinical application of ctDNA for predicting and monitoring response to treatment in ovarian cancer.