High-throughput approaches for precision medicine in high-grade serous ovarian cancer

CA125 is a large membrane glycoprotein encoded by the gene MUC16 and was identified as a tumor marker for EOC in 1983 [41]. Significant expression of CA125 is observed in 85% of serous, 65% of endometrioid, 40% of clear cell, 36% of undifferentiated and 12% of mucinous ovarian cancers, highlighting the lack of utility of CA125 in some EOC subtypes [42]. Despite being the most widely used biomarker for EOC, CA125 offers limited value as a diagnostic test. Serum concentrations of CA125 are elevated in 90% of advanced-stage EOCs and less than 50% of early-stage EOCs, resulting in a low sensitivity for detecting early-stage disease [43]. Furthermore, serum CA125 abundance has a low specificity for EOC as levels can be increased due to multiple benign gynecological and medical conditions including endometriosis and pregnancy [44]. The low specificity is especially manifest in premenopausal women who are at an increased risk of many of these other conditions [45]. Given the limitations of CA125 as a stand-alone diagnostic marker, the Risk of Malignancy Index (RMI) [46] and the International Ovarian Tumor Analysis (IOTA) Adnex model [47] were developed to integrate serum CA125 levels, ultrasound criteria and demographics, resulting in improved specificity and sensitivity for differential diagnosis of pelvic masses prior to surgery. When evaluated as a potential screening test, both the UKCTOCS study [9] and the PLCO trial [48] demonstrated that serum testing of CA125 alone or combined with transvaginal ultrasound imaging did not reduce mortality due to EOC and resulted in an increase in unnecessary invasive procedures associated with complications, underlining the clinical consequences of low specificity.

Nevertheless, CA125 offers clinical utility when evaluating treatment response and monitoring remission. A decrease in serum CA125 is indicative of treatment response, whereas a persistence of abnormally elevated CA125 or increases may suggest treatment resistance and/or residual disease [39]. Many post-treatment surveillance protocols include serial measurements of CA125, as rising serum CA125 is strongly predictive of disease recurrence [49‐51]. A rise in CA125 concentration has been shown to precede clinical detection of recurrent disease by at least three to five months [49, 52, 53]. However, up to half of patients within the normal limits of CA125 during remission are found to have small volumes of disease during a second-look surgery [54, 55]. Hence, despite being the earliest sign of recurrence currently available, CA125 is not optimally sensitive for detecting recurrence in all patients.

Human epididymis protein 4 (HE4), encoded by the gene WFDC2, is a secreted glycoprotein that is overexpressed in serous and endometrioid ovarian cancers [56]. Hellstrom et al. [57] initially determined that serum HE4 was comparable to CA125 for distinguishing between patients with advanced-stage disease and healthy controls. Subsequent studies have produced conflicting reports regarding the sensitivity of HE4 compared to CA125 as a diagnostic test, yet there is a consensus that HE4 is more specific than CA125, especially in premenopausal women [58‐61]. This superiority is likely due to serum levels of HE4 being less influenced by other gynecological disorders such as endometriosis [62]. Serum HE4 is also elevated in at least a third of patients who do not demonstrate elevated serum CA125 levels, signifying a role for complementary markers in diagnostics [63]. Serum HE4 is currently approved to be used as a tumor marker for monitoring disease progression and recurrence. A study evaluating serum levels of HE4 and CA125 prior to surgery for suspicious recurrent EOC found HE4 to be more sensitive and specific than CA125 [64]. In a pilot study, Anastasi et al. [65] found that a rise in serum HE4 preceded elevated serum CA125 five to eight months in five out of eight patients with recurrent disease. Preliminary studies have also revealed that HE4 elevation can detect recurrence in a subset of EOC patients that do not present with increased serum CA125 [66, 67]. The combination of both markers resulted in a higher sensitivity and specificity in detecting recurrence than either marker alone [68], but data from larger prospective trials, including potential benefits in survival, are still pending.

In response to the promising data regarding the complementary potential between CA125 and HE4 as diagnostic markers, the Risk of Malignancy Algorithm (ROMA) was developed and is currently approved for differential diagnosis [69]. ROMA combines serum measurements of CA125 and HE4 and uses two different logarithmic regression models based on menopausal status to determine the likelihood of malignancy in women who are having surgery for pelvic masses [70]. A meta-analysis comparing ROMA, HE4 and CA125 revealed that ROMA demonstrated the greatest sensitivity and HE4 exhibited the highest specificity for differential diagnosis, although these differences were not statistically significant [71]. In a direct comparison, ROMA was found to be more sensitive than RMI with similar specificities [72], yet both assays demonstrated low sensitivity for early-stage disease [73]. OVA1 is a biomarker panel of five proteins used for the differential diagnosis of pelvic masses prior to surgery. The test consists of immunoassays for two upregulated proteins (CA125, beta 2 microglobulin) and three down-regulated proteins (transferrin, transthyretin, apolipoprotein A1) in serum [74]. An algorithm is used to integrate the measurements of each marker to generate an ovarian malignancy risk score ranging from 0 – 10. The threshold for risk of malignancy is dependent on menopausal status. Many prospective studies comparing the performance of OVA1 to CA125 have reported higher sensitivity than CA125, especially for early-stage disease, yet lower specificity [75‐77]. There have been no direct comparisons between OVA1 and ROMA to date. Overa is a second-generation multivariate assay which was intended to overcome the low specificity of OVA1 [78]. Overa uses serum measurements of CA125, HE4, apolipoprotein A1, transferrin and follicle-stimulating hormone to assess the likelihood of malignancy in women who will undergo surgery for a pelvic mass. The incorporation of follicle-stimulating hormone eliminates the need for assessing menopausal status as in OVA1. Overa was designed and validated using the same study population as OVA1, thus allowing for direct comparisons between the two assays. Indeed, Overa demonstrated an improved specificity and similar sensitivity to OVA1 [78].

Germline deficiencies in BRCA1/2 are the strongest genetic risk factors for nonmucinous EOC [79]. The cumulative risk of developing ovarian cancer in BRCA1 and BRCA2 carriers ranges from 40 to 59% and 16 to 18%, respectively [80‐82]. As such, genetic counseling and genetic testing can be suggested for patients with familial history of breast, ovarian, pancreatic or prostate cancer to identify those who are at an elevated risk [83] and will likely benefit from preventative measures such as risk-reducing bilateral salpingo-oophorectomies [84, 85]. In addition, germline and somatic BRCA1/2 mutations are considered predictive biomarkers due to strong associations with favorable outcomes following both platinum-based chemotherapy [86, 87] and maintenance therapy with PARP inhibitors [88, 89]. PARP inhibitors have also been approved for usage as monotherapy in BRCA1/2-deficient women with recurrent disease [90, 91]. Hence, genetic testing for BRCA1/2 mutations is recommended for all newly diagnosed ovarian cancer patients to aid in therapy selection and determining cancer risk for family [83].

Circulating cell-free DNA (cfDNA) are short fragments of DNA released into the bloodstream from apoptotic or necrotic cells [93]. The quantification of total cfDNA has revealed that EOC patients have elevated levels of cfDNA compared to healthy controls and patients with benign disease [94‐96]. However, evaluation of cfDNA abundance is not a direct measure of tumor burden as DNA fragments are also released by noncancerous cells. The fraction of cfDNA that originated from tumor cells, termed circulating tumor DNA (ctDNA), can be distinguished by the presence of cancer-specific alterations [97]. Given the low abundance of ctDNA compared to cfDNA, highly sensitive approaches such as digital polymerase chain reaction (PCR) and targeted next-generation sequencing (NGS) are used to detect cancer-specific modifications. Since TP53 mutations are ubiquitous in HGSC, detection of TP53 mutants in cfDNA has been preferentially used when investigating ctDNA as a biomarker [98‐101]. Parkinson et al. [98] analyzed TP53 mutations in longitudinal plasma samples from HGSC patients undergoing treatment to evaluate the value of ctDNA in determining prognosis and response to treatment. This study revealed that the abundance of TP53 mutant ctDNA fractions prior to treatment significantly correlated with volumetric measurements of tumors from CT images, unlike CA125, and that a decrease of > 60% of TP53 mutant ctDNA fractions following treatment was a predictor of time-to-progression. Christie et al. [102] recently investigated whether reversal of germline BRCA1/2 mutations can be detected in ctDNA, as this molecular alteration is known to correspond with acquired chemo-resistance. Reversion mutations were detected in the plasma of three out of five patients with reversion mutations observed in tumor samples, all of whom were resistant to platinum-based therapy or PARP inhibitors. The authors noted that detection of reversal mutations was associated with the fraction of ctDNA out of total cfDNA, measured by the presence of TP53 mutant alleles. Certainly, a limiting factor of utilizing ctDNA as a tumor marker is that current detection strategies may not be sensitive enough to detect rare mutants in early-stage disease where the ctDNA fraction is low [103].

In addition to investigating specific genes, examination of genome-wide chromosomal aberrations in ctDNA can be promising for the discovery of novel tumor markers. Harris et al. [104] used whole-genome sequencing to characterize genomic rearrangements in primary tumors of HGSC patients and investigated whether patient-specific aberrant chromosomal junctions could be detected in plasma. ctDNA with patient-specific chromosomal alterations was detected in pre-surgically drawn plasma samples for eight out of ten patients. Postsurgical detection of ctDNA was specific to the only three patients with clinically documented residual disease, suggesting a potential for personalized markers of tumor burden [104]. As changes in DNA methylation and chromatin remodeling play a role in tumor biology, the rise of epigenetic technologies (e.g., methylation profiling) is also promising for the use of ctDNA as blood-based tumor markers [105]. In the context of HGSC, Widschwednter et al. [106] identified aberrant methylation signatures in tumor tissues and developed a three-marker DNA methylation panel for ctDNA that was able to discriminate patients from healthy women or women with benign masses. The panel was also shown to better distinguish between platinum responders and nonresponders than CA125.

microRNAs (miRNAs) are a class of short (19–25 nucleotides) noncoding RNAs that are involved in gene regulation. miRNAs can function as oncogenes or tumor suppressors depending on cellular context, and expression has been shown to be deregulated in several cancers [107]. miRNAs are actively secreted from cells by binding to protein complexes or by being packaged into extracellular vesicles, thus providing protection from RNAse digestion and degradation in various extreme conditions (e.g., high temperatures, severe pH and multiple freeze–thaw cycles) [108, 109]. The stability of miRNAs in blood renders them as attractive molecules for tumor markers in liquid biopsies. In large-scale miRNA biomarker discovery experiments, high-throughput qRT-PCR panels, microarrays and more recently, NGS, can be used for profiling miRNAs in patient samples [110‐112]. Todeschini et al. [110] used microarrays to profile miRNA expression in sera from HGSC patients and healthy controls. The differentially expressed miRNAs were then quantified in an independent cohort from which a single miRNA that demonstrated the greatest ability in discriminating HGSC patients from controls was identified as a putative diagnostic biomarker. Shah et al. [111] used qRT-PCR panels for serum miRNA profiling and demonstrated that combining measurements of circulating miRNAs and CA125 can be predictive of surgical resection outcomes for women with HGSC, suggesting value for circulating miRNAs as prognostic markers.

Proteins are the primary functional elements of most biological processes, and thus, protein expression is often deregulated in disease states. Mass spectrometry (MS) is a powerful approach for protein measurement as current MS-based proteomic experiments can detect thousands of proteins in a single sample. MS has already proven to be fruitful in EOC biomarker discovery as the four markers in OVA1 (excluding CA125) were discovered using MS-based approaches [74]. In the studies that led to the development of OVA1, seven protein candidates were identified in the original discovery phase, yet verification of candidates was limited to only those proteins which had existing immunoassays [113, 114]. Though this approach is advantageous for faster clinical adoption, antibody availability can pose as a bottleneck for validation of candidate markers in biomarker discovery pipelines. Targeted MS approaches such as multiple reaction monitoring (MRM) and the more recently developed parallel reaction monitoring (PRM) can enable high-throughput robust quantification independent of antibody availability, thus circumventing the need for antibody development during biomarker discovery.

Detection of blood-based protein markers is challenging due to the large dynamic range and high sample complexity of serum/plasma. Considering that the 22 most abundant proteins in plasma account for 99% of the total mass of protein [115], detection of low-abundance proteins, often the most promising proteins for biomarker candidates, is hindered. Several preanalytical workflows have been developed to overcome this complexity, including the depletion of high-abundance proteins, sample fractionation and/or the enrichment for sub-proteomes [116]. N-glycosylation is a posttranslational modification that plays an important role in the stability, solubility and localization of proteins to the cell surface [117]. N-glycosylated proteins can be enriched from biological samples using chemoproteomic- and lectin-based approaches [118, 119]. As N-glycosylation is highly prevalent among extracellular proteins (including secreted proteins) and is not present on several high-abundance blood proteins (i.e., albumin), the N-glycoproteome represents a clinically relevant sub-proteome for liquid biomarker discovery. Sinha et al. [120] recently devised an integrated N-glycoproteomics-based approach for detecting biomarkers of HGSC relapse. N-glycosylated peptides were enriched from the sera and tumors from recurrent HGSC patient-derived xenograft (PDX) mice and from sera of non-engrafted mice. Species mapping was used to distinguish between peptides of human (tumor) and mouse (stroma) origin, and comparative analysis was used to select a set of candidate markers. Subsequently, PRM was used to quantify the candidates in longitudinal HGSC patient serum samples, revealing four candidates that demonstrated an earlier rise between the remission and the recurrence time points than CA125. Although large-scale clinical validation of the markers is warranted, this study is a proof-of-concept for the use of N-glycoproteomics and PDX models in serum protein biomarker discovery for HGSC.

Posttranslational modifications and metabolic processes are important determinants of cellular signaling and modulating phenotypes. As such, these molecular classes also represent promising candidates for blood-based biomarker discovery. Considering that aberrant glycosylation occurs during malignant transformation [121], one such approach consists of profiling differences in glycan structures on glycoproteins. Biskup et al. [122] used MS to compare the serum N-glycome profiles between serous EOC patients and healthy women. This glycomics study revealed a marker panel comprising 11 differentially abundant glycans that demonstrated an improved specificity for distinguishing patients from healthy controls compared to CA125. Moreover, as metabolic alterations have been implicated in tumorigenesis [123], metabolomics and lipidomics have emerged as potential avenues for biomarker discovery. Zhou et al. [124] used MS to examine the metabolite profiles of sera from HGSC patients, women with benign ovarian masses and healthy controls and subsequently developed a machine-learning algorithm for diagnostic classification based on the mass spectrum profiles. Buas et al. [125] performed lipidomics analyses of plasma collected from serous EOC patients and patients with benign ovarian masses. A classification model incorporating CA125 and four lipid metabolites demonstrated an increased diagnostic accuracy compared to CA125 alone. Together, these studies suggest a potential utility for plasma metabolites to aid in the diagnosis of HGSC.

Aside from investigating freely circulating molecules in blood, molecular profiling of extracellular vesicles (EVs) and circulating tumor cells (CTCs) are alternative approaches for blood-based biomarker discovery. EVs, such as exosomes, are secreted from most cell types, play a role in intercellular communication and contain molecular content from the cell-of-origin [126]. In a pilot study, Taylor et al. [127] identified eight exosomal miRNAs that demonstrated significantly distinct expression profiles in the sera of serous EOC patients compared to the sera of women with benign disease. These exosomal miRNAs exhibited a similar expression profile in tumor tissue and were not detected in the sera of healthy controls. Recently, Kobayashi et al. [128] used microarrays to profile miRNAs in exosomes isolated from the conditioned media of HGSC cell lines and immortalized ovarian surface epithelial cells. A single upregulated miRNA was selected for subsequent quantification in EOC patient sera and was found to be differentially expressed between sera of HGSC patients and sera from non-HGSC patients [128]. This illustrates a potential for noninvasive molecular stratification of EOC. Peng et al. [129] compared the proteomes of serum exosomes from serous EOC patients and tumor tissue, revealing 35 proteins commonly upregulated in comparison with normal samples. These findings suggest that exosomes may be of use for noninvasive molecular tumor examinations.

CTCs are tumor cells that are shed into vasculature and play an important role in metastasis [130]. Although CTCs have been explored as noninvasive tumor markers in the general context of EOC, to the best of our knowledge, there are no published investigations specifically focusing on HGSC to date. Studies have primarily focused on the detection and/or the enumeration of CTCs as potential biomarkers in EOC, yet there have been conflicting reports potentially due to differences in isolation strategies [131‐133]. Molecular investigations of CTCs are less prevalent and have traditionally involved the use of qRT-PCR to evaluate the expression of a few specific genes. Zhang et al. [134] examined the expression of six genes that were known to be associated with EOC and demonstrated that EpCAM and ERBB2 expressions in CTCs were correlated with platinum resistance and overall survival. Emerging technologies in microfluidics-based CTC isolation and single-cell molecular analysis present new avenues for high-throughput examinations of individual CTCs. Single-cell RNA sequencing of CTCs has been shown to be promising in understanding clonal resistance and metastasis to potentially inform therapeutic decisions in other cancers [135, 136]. Furthermore, MS-based workflows have recently been developed to profile the proteomes of CTCs, allowing for another layer of molecular characterization [137, 138]. Although single-cell approaches have yet to be applied to CTCs in HGSC, it is proposed that comprehensive molecular characterization of CTCs can provide noninvasive insights regarding intratumor heterogeneity and aid in patient selection for targeted therapies.

Angiogenesis is a rate-limiting step in the evolution of cancer [141] and has therefore been studied as a potential target for systemic treatment. Bevacizumab is a monoclonal antibody that targets vascular endothelial growth factor (VEGF) A which is secreted by tumors to induce the formation of new blood vessels [142]. Early studies of bevacizumab have demonstrated improved progression-free (PFS) and overall survival (OS) in colorectal and renal cancer [143, 144]. Two landmark trials that assessed the role of concurrent and maintenance treatment with bevacizumab in EOC were the GOG-0218 (primary endpoint: PFS) and the ICON7 (primary endpoints: PFS and OS) studies [145, 146]. Both trials have shown significant improvements in PFS in the intention-to-treat populations with bevacizumab compared to chemotherapy alone but have failed to improve OS in the overall study population. The clinical significance of a three-month difference in PFS has been debated, and as such, bevacizumab is not universally used in the first-line treatment of EOC. In the ICON7 trial, women with high-risk features (inoperable stage III, suboptimal debulking and stage IV disease) randomized to bevacizumab had a significant improvement in mean OS of 4.8 months [147]. Similarly, a subanalysis of the GOG-0218 study suggested that patients with International Federation of Gynecology and Obstetrics (FIGO) stage IV disease may have an increased survival benefit from bevacizumab [148].

In addition to its utility as a first-line therapy, bevacizumab has proven to be effective in patients with recurrent disease. Clinical trials have revealed significant improvements in PFS when bevacizumab was added compared to chemotherapy alone in both platinum-sensitive [149] and platinum-resistant patients [150]. Similar to primary disease, the use of bevacizumab for recurrent disease was not associated with significant improvement in OS for all participants [150, 151]. Other anti-angiogenic therapeutics currently under investigation for EOC include pazopanib [152] and nintedanib [153], both of which have shown similar improvements in progression-free survival in clinical trials. Considering the potential severe side effects including hypertension, renal complications, hemorrhage, gastrointestinal perforation and fistula formation, appropriate patient selection and balancing the risks and potential benefits play a pivotal role for anti-angiogenic treatment.

PARP-1 was first described in 1966 [154] but its pivotal role for ovarian cancer was only recently discovered [155]. PARP-1 and PARP-2 are enzymes that play a critical role in base excision repair, a repair mechanism for DNA single-strand breaks [156]. The inhibition of PARP results in an accumulation of single-strand breaks, which can lead to double-strand breaks during replication. The double-strand breaks are normally repaired by a process termed ‘homologous recombination’ [156]. In cancer cells with BRCA1/2 mutations or other HRDs, the inhibition of PARP results in a synthetic lethal interaction as the accumulation of double-strand breaks coupled with inadequate repair mechanisms can lead to chromosomal instability, cell cycle arrest and subsequent apoptosis [157]. Olaparib, niraparib and rucaparib are the three PARP inhibitors that are currently FDA-approved for recurrent ovarian cancer after showing consistent improvement in PFS [88, 158‐160].

In the SOLO-1 trial, olaparib was tested as a frontline maintenance treatment in women with newly diagnosed FIGO stage III–IV ovarian cancer with germline or somatic BRCA1/2 mutation following cytoreductive surgery and platinum-based chemotherapy. PFS was significantly improved in the olaparib arm compared to placebo and the 3-year progression-free survival was 60.4% versus 26.9%, HR 0.3 (p < 0.001) [89]. The FDA has subsequently approved olaparib for frontline maintenance treatment in women with platinum-responsive ovarian cancer and BRCA1/2 mutation. More recently, the PRIMA trial has shown that niraparib is also effective in the overall population regardless of the HRD status, but the post hoc subanalysis has clearly shown that those patients with BRCA1/2 mutations and other HRDs benefited most from maintenance treatment with niraparib [161].

For patients at high risk for recurrence/progression, there is currently a lack of evidence to suggest the superiority of either anti-angiogenic agents or PARP inhibitors over the other. Interim data from the ongoing PAOLA study investigating the combination of PARP inhibitors and bevacizumab suggest a significant benefit in PFS from concurrent use of both agents [153]. Future studies will need to identify patient groups who benefit most from PARP inhibitors, anti-angiogenic treatment, evolving therapeutics such as immunotherapies or a combination of these, while balancing the benefits and added toxicities from combination treatments.

In molecular expression profiling experiments, high-throughput discovery experiments are performed on one or more molecular classes with the underlying assumption that differences in the expression profile of a molecule, or group of molecules (i.e., those associated with a biological pathway), may inform the understanding of disease pathology. As such, the discovery experiments are typically followed by analyses designed to reveal these differences, including differential expression analyses and ontologically informed pathway analyses. Molecular profiling experiments on patient cohorts enable matching clinical data to molecular phenotype which can be especially apt for identification of candidates for targeted therapies. In one such study, Coscia et al. [162] compared the proteomic profiles of platinum-sensitive and -resistant HGSC patient samples revealing cancer/testes antigen 45 (CT45) expression to be predictive of disease-free survival. By establishing a link between demethylating agents and CT45 expression and by linking CT45 to cytotoxic T cell engagement, two potential therapeutic strategies can be devised from these findings.

Molecular expression profiling experiments offers the advantage of not requiring a priori information regarding the pathology of the disease. However, a significant hurdle which lays between molecular expression profiling experiments and novel therapeutics is the potential need to develop novel drug compounds. While there are some measures for druggability which can inform therapeutic target selection, drug development remains a costly process with uncertain success [163]. One strategy to mitigate this challenge is to focus on protein classes for which there is a history of therapeutic intervention. For instance, though kinases are often effective targets for cancer therapeutics, established protein kinase therapies have demonstrated limited utility in HGSC. Recognizing this disconnect, Kurimchak et al. [164] profiled the kinome of primary tumors and PDXs to detect differentially expressed kinases in HSGC revealing a potential therapeutic target, MRCKA.

Surface proteins are another especially useful class of molecules as their accessibility renders them favorable therapeutic targets—evidenced by the fact that over 58% of the known protein targets of FDA-approved drugs are cell surface proteins [165]. However, proteomic workflows which enrich for surface proteins typically require large starting amounts and are therefore not practical or possible for all systems. Despite the challenges associated with surface proteomic workflows, the profiling of the cell surface proteins is well suited for identifying targets for repurposing approved therapeutics and development of new therapeutics. Antibody–drug conjugates (ADC), an emerging class of therapeutics for cancer treatment, enable surface proteins to act as potential therapeutic targets independent of their direct connection to disease pathology [166, 167]. One ADC, IMGN853, which targets folate receptor ɑ, is being evaluated in a phase 3 clinical trial for folate receptor-positive platinum-resistant EOC patients [168]. In this case, although there is some evidence to suggest that targeting folate receptor ɑ alone could have an effect on cancer progression [169], the cytotoxic component of IMGN853 is the maytansinoid compound conjugated to the antibody which targets microtubules.

Phenotypic screening approaches can be used to identify tumor-specific molecular dependencies as putative targets for therapeutic inhibition. Technological advancements have enabled large-scale molecular perturbations which allow for the functional examination of thousands of molecules in a single experiment. Functional genomic screens entail the use of RNA interference (RNAi) or more recently, clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9 systems coupled with NGS to characterize the genes associated with a phenotype of interest [170, 171]. In a pan-cancer comparison, Cheung et al. [172] performed genome-wide shRNA knockdown screens in 102 cell lines, including 25 ovarian cancer cell lines revealing 54 genes exclusively essential for ovarian cancer viability and proliferation, underlining the utility of functional genomic approaches to identify lineage-specific dependencies. Functional genomic screening can also be used to identify concurrent therapeutic targets that improve chemosensitivity of existing therapies. Fang et al. [173] performed a genome-wide CRISPR knockout screen in an HGSC cell line treated with olaparib to identify targets that mimic HRD. Based on this screen, the authors were able to characterize a gene whose knockout increased the cytotoxic effects and can potentially extend the clinical utility of PARP inhibitors in HGSC.

In contrast to genome-wide interrogations, functional genomics screens can also be conducted on a subset of experimentally relevant genes. To identify novel synthetic lethal targets in BRCA2 deficient tumors, Mengwasser et al. [174] conducted targeted CRISPR screens in two pairs of isogenic cell lines; one breast cancer pair and one HGSC pair. The isogenic cell lines differed based on the presence of a functional BRCA2 gene and the screen targeted 380 genes that are involved in DNA damage repair. Interestingly, the authors determined two candidates that not only demonstrated synthetic lethality with BRCA2 deficiencies but also BRCA1, as evident through subsequent investigations. The candidates identified in this study represent potential therapeutic targets for BRCA-deficient tumors that are resistant to PARP inhibitors. Likewise, Baratta et al. [175] designed an in vivo shRNA screen to assess the depletion of ~ 800 genes in xenografts of a human HGSC cell line. The screen revealed several candidates essential for proliferation/survival of HGSC. Through further investigations in patient-derived cell lines, one gene was identified as a potential target for MYCN overexpressing tumors.

Although functional genomics experiments are advantageous approaches for identifying molecular vulnerabilities for potential inhibition, exploiting these candidates as therapeutic targets can be hindered by the druggability of proteins. The use of high-throughput drug screens is an alternative phenotypic screening approach to identify actionable dependencies. In these experiments, numerous small-molecule compounds, typically with known mechanisms of action, are simultaneously tested against cells to identify novel vulnerabilities. Kenny et al. [176] recently performed a fully robotic screen of ~ 45,000 small molecules in an HGSC organotypic model consisting of one of five HGSC cell lines, primary human stromal cells and extracellular matrix components. Subsequent in vitro and in vivo assays identified three compounds that prevent cancer adhesion, proliferation and invasion suggesting that these compounds can be promising therapeutics for ovarian cancer metastasis. Additionally, drug screens can be used to elucidate indirect mechanisms of targeting undruggable cancer proteins. Zeng et al. [177] screened a small molecule library in two HGSC cell lines to identify alternative methods of down-regulating MYC; an oncogene essential when overexpressed in HGSC yet pharmacologically undruggable. This screen revealed a compound that suppressed MYC expression through simultaneous inhibition of three specific cyclin-dependent kinases, hence identifying putative targets for MYC overexpressing HGSC tumors. A caveat associated with drug screens is that target discovery is restricted to those proteins with existing small molecule inhibitors, thus ignoring the potential of other therapeutic classes such as monoclonal antibodies.

The use of the aforementioned high-throughput experiments is beneficial as initial steps in therapeutic target discovery. Both approaches enable the concurrent screening of numerous molecules to select potential candidates for further validation experiments. However, as technological advancements improve the capabilities of these platforms, these large-scale experiments can identify hundreds of hits and given the time-consuming nature of molecular biology interrogations, it is often not feasible to individually follow up on every single hit. The use of bioinformatic tools that prioritize targets (e.g., SurfaceGenie [178]) or mining publicly available data, such as TCGA and the Genotype-Tissue Expression (GTEx) project, can help further narrow down candidates, yet well-annotated complete data are not always available for all diseases. Hence, an emerging workflow for therapeutic target discovery is the integration of both high-throughput approaches for a multifaceted characterization of candidates. By leveraging the advantages of these two orthogonal approaches, an integrated workflow can produce a refined list of candidates that are both actionable and essential. Medrano et al. [179] conducted genome-wide shRNA screens and cell-surface characterizations of 27 HGSC cell lines in parallel, resulting in the identification of CD151 as a cell-surface protein that demonstrated essentiality in a subset of HGSC cell lines. Subsequently, the authors performed RNA sequencing to molecularly characterize the discrepancies in response to knockout of the candidate. This study identified both a novel therapeutic target and a molecular marker of target sensitivity, highlighting the utility of integrated high-throughput workflows for HGSC target discovery. Similar target discovery pipelines have proven promising in other cancer settings as well. Martinko et al. [180] used MS to evaluate changes in cell surface expression associated with oncogenic KRAS and in parallel, conducted a targeted CRISPR knockdown screen of ~ 1600 membrane proteins to functionally characterize the oncogenic KRAS surfaceome. Integration of both datasets resulted in the discovery of CDCP1 as a therapeutic antibody target for KRAS-driven cancer cells. Considering the limited success in translating promising targets into beneficial therapies, comprehensive characterization through both expression profiling and functional analyses early in the discovery process may help ensure the selection of robust targets for therapy development.

Ideal experimental models should accurately reflect tumor biology to ensure maximum translational utility. Given the ease of use and accessibility, immortalized human cancer cell lines are the most widely used models for experimental interrogations of HGSC [181]. However, until recently, the majority of cell lines used in HGSC research were poorly characterized with uncertain histopathological origins. To address these ambiguities, Domcke et al. [182] compared copy-number changes, mutations and mRNA expression profiles of 47 EOC cell lines and 316 HGSC tumor samples examined by TCGA. Strikingly, this extensive evaluation concluded that the most frequently used cell lines in HGSC research poorly recapitulated the genomic and transcriptomic features of HGSC tumors and are likely other EOC histopathologies. The authors recommended an alternative set of cell lines that closely resemble HGSC tumors and thus would be more appropriate as in vitro models. A separate proteomic profiling study of 28 EOC cell lines, two immortalized ovarian surface epithelial cell lines, three primary fallopian tube epithelial cell isolates and eight HGSC tumor tissues revealed distinct groups of cell lines [183]. The majority of cell lines reported to likely represent HGSC as per Domcke et al. [182] clustered with the proteomes of HGSC tumors and fallopian tube samples, further confirming a HGSC histopathology. Additional studies have revealed discrepancies amongst the ability of HGSC cell lines to model tumor metastasis and histopathology in vivo when xenografted [184, 185]. Together, these studies illustrate the disconnect between certain model cell lines and HGSC tumors and highlight the importance of informed cell line selection.

Another caveat of in vitro cell line models is the artificial microenvironment invoked by monolayer growth on plastic and the lack of multicellularity. Three-dimensional (3D) culture has emerged as a step toward bridging the gap between in vitro and in vivo experiments. 3D culture more closely resembles the tumor microenvironment by restoring 3D cell–cell and cell–ECM interactions [186, 187]. Moreover, research groups have successfully demonstrated co-culturing with patient-derived fibroblasts [188] and patient-derived mesothelial cells [189] in HGSC spheroid models to capture the influence of tumor-stromal cross talk on survival and proliferation. Provided that HGSC disseminates through the release of multicellular aggregates into the peritoneal cavity, 3D organotypic in vitro models have been developed to recapitulate significant events in metastasis and gain insights into tumor biology [190]. Nonadherent 3D models have also been used to investigate cancer stem cell populations enriched in disseminated spheroids which are thought to contribute to chemoresistance in HGSC [191]. The use of organ-specific growth factors to model niche environments has enabled the development of ovarian cancer organoid lines that maintain the genomic and histological features of primary tumors and preserve tumor heterogeneity, highlighting their potential utility for precision medicine research [192].

Genetically engineered mouse models (GEMMs) offer the potential for in vivo tumor investigation. Various molecular biology techniques can be used to introduce genetic modifications in a spatial and temporal manner for in vivo modeling of genetic defects contributing to tumorigenesis [193]. However, the preclinical utility of these models is dependent on the accuracy in recapitulating the histology and pathogenesis of human tumors, an element which has historically proven difficult in the context of HGSC. Given the uncertainty in the site of origin for HGSC, early attempts to develop HGSC GEMMs have focused on targeting the ovarian surface epithelium for genetic manipulations [193]. These models failed to replicate the molecular and clinical features observed in human HGSC tumors. In addition to selecting the correct cell of origin, targeting genes that are relevant to HGSC is also an important consideration when generating GEMMs. Indeed, targeting different combinations of oncogenes and tumor suppressors in the same cell of origin has resulted in different HGSC GEMM phenotypes [194, 195]. Fortunately, as molecular understanding of HGSC biology evolves, so does the ability to correctly model the disease. Targeting HGSC relevant genes, such as TP53, BRCA1, RB1 and PTEN, in fallopian tube epithelial cells has resulted in a new generation of HGSC GEMMs [195‐197]. These clinically relevant models reproducibly demonstrated the formation of precursor STICs in fallopian tubes and mirrored the aggressive metastatic patterns observed in human HGSCs. Hence, if modeled correctly, GEMMs represent promising options for interrogations of early-stage disease and identifying new therapeutic targets.

PDXs are an alternate approach for in vivo experimental models, in which minced fragments of patient tumors are transplanted into immunodeficient mice [198]. The primary advantage of this experimental system is the ability to perform in vivo interrogations of human tumors. Although PDXs have been successfully generated through various different engraftment locations, orthotopic engraftment is preferred as it results in a physiologically relevant microenvironment [199]. In the context of HGSC, there are two engraftment sites that are considered orthotopic: intrabursal (IB) engraftment and intraperitoneal (IP) engraftment [181]. IB engraftment refers to the injection of tumor cells into the ovarian bursa, which is the fat pad surrounding a murine ovary [200]. As there are anatomic differences between the reproductive systems of mice and humans, the bursa can often hinder the extensive peritoneal metastasis characteristic of advanced human ovarian tumors [200]. Alternatively, IP engraftment consists of injecting the tumor cell suspension directly into the peritoneal cavity, mirroring human abdominal tumor dissemination [120, 200]. In addition to recapitulating tumor pathology, orthotopic HGSC PDXs have also been shown to maintain molecular profiles highly comparable to patient tumors [201] and emulate patient-specific responses to platinum-based therapy [202]. These studies highlight the advantage in using PDXs for identifying novel precision medicine approaches in HGSC as it provides an in vivo opportunity for investigating tumor heterogeneity. Indeed, Weroha et al. [203] developed an orthotopic PDX bank consisting of 241 EOC models that reflected the molecular diversity observed in patients and can be a promising resource to investigate subtype specific biomarkers and therapies. A limitation of PDX models is the inability to evaluate interactions between the tumor and the immune system, an important facet of the tumor microenvironment [198].