Early Detection of Ovarian Cancer.

Hypermethylation of tumor suppressor promoters and hypomethylation of oncogenes are frequent genetic events 40 – 42 . Methylation-specific PCR (MSP) is very sensitive, able to identify 1 methylated allele in 1000 unmethylated alleles 43 . The frequency of promoter hypermethylation increases with advancing disease stage 44 . Using multiplexed MSP to examine cfDNA for seven candidate genes ( APC , RASSF1A , CHDH1 , RUNX3 , TFP12 , SFRP5 , and OPCML ), Zhang, et al , reported 85% sensitivity at 91% specificity for early stage ovarian cancer compared to a single CA125 value, which produced a sensitivity of 56% at 64% specificity. However, this was based on only 17 early-stage patients 45 . More recently, Widshwendter¸ et al , described a three-DNA-methylation-serum-marker panel developed from 699 cancerous and non-cancerous tissue samples 46 . They used targeted ultra-high coverage bisulfite sequencing in 151 women and validated in 250 women with various conditions, including those associated with high CA125 levels (endometriosis and other benign pelvic masses), serial samples from 25 patients undergoing neoadjuvant chemotherapy, and a nested case control study of 172 UKCTOCS control arm participants. The marker panel discriminated high grade serous ovarian cancer patients from healthy women or patients with a benign pelvic mass with 41.4% sensitivity at 90.7% specificity. When applied to serum samples collected 1–2 years before an ovarian cancer diagnosis, the methylation panel had 16.7% sensitivity at 96.9% specificity.

TVS is the preferred clinical modality for imaging the adnexa. TVS can be delivered at most centers, at low cost, without radiation, and with minimal discomfort to patients 56 . Among women with an adnexal mass, morphology indexing has a high NPV of 0.997 for excluding malignancy 57 . Doppler flow studies have improved the specificity of TVS in experienced hands. However, one drawback of TVS is that the resolution of sonographic imaging is insufficient to diagnose very small invasive or pre-invasive lesions. Failure to image fallopian tubes is a particularly important limitation in that many high grade serous ovarian cancers are believed to arise from epithelial cells on the fimbriae of the fallopian tubes 38 . Microbubble contrast holds promise and has improved the ability to distinguish benign from malignant adnexal masses but is not likely to improve detection of fallopian tube lesions. Hyperpolarized 13 C MRI has shown a unique signature in prostate cancers and might prove useful in ovarian cancer 58 . Magnetic relaxometry (MRX) is another modality that might substantially enhance sensitivity by two orders of magnitude 59 . Superconducting Quantum Interference Detection (SQUID) can measure delays in magnetic relaxation of antibody-coated iron oxide nanoparticles. Such delays are observed when nanoparticles bind to cancer cells, but not when they are free in the blood or peritoneal cavity. This modality has been applied to detecting breast cancer cells in murine xenografts, minimal residual disease in leukemic bone marrow biopsies, and measuring nanoparticle accumulation in biological samples 60 , 61 . Studies are currently being conducted with human ovarian cancer xenografts, but clinical studies have not yet been performed.

CA125 remains the most sensitive and specific protein biomarker for detecting early stage disease in apparently healthy populations. CA125 is a high molecular weight (~5 MDa) heavily glycosylated membrane-spanning mucin (MUC16) glycoprotein. The extracellular domain of MUC16 is cleaved near the ovarian cancer cell surface, releasing CA125 into the peri-cellular space and ultimately into the blood where it can be measured with an immunoassay. CA125 levels are elevated in blood from >90% of patients with advanced stage (III-IV) and in 50–60% with stage I ovarian cancer 17 . More than 110 potential protein biomarkers have been evaluated individually and in combination with CA125 18 . Other top candidates include HE4, transthyretin, CA15.3, and CA72.4 19 . HE4 (human epididymal protein 4) is a 124-amino acid glycosylated whey protein that is elevated in sera from approximately 60–75% of ovarian cancer patients and that detects a small fraction of cases missed by CA125. CA15.3 and CA72.4 are distinct epitopes on the MUC1 mucin. Terry, et al , measured CA125, HE4, CA72.4, and CA15.3 in 810 invasive epithelial ovarian cancer cases and 1,939 controls from Phase III specimens from the European Prospective Investigation into Cancer and Nutrition study 20 . All the markers performed best within 6 months of diagnosis, but the capacity to discriminate between future case patients and non-cases dropped rapidly with increasing time from blood collection to clinical diagnosis. Successive additions of CA125, HE4, CA72.4 and CA15.3 as pre-diagnostic predictors of future ovarian cancer diagnosis could improve the model C-statistic, but only minimally compared to a model based on CA125 alone (0.70 to 0.71). Our own studies indicate that the addition of HE4 and CA72.4 detects 18% of cases missed by CA125, but does not provide diagnostic lead time in specimens from the UKCTOCS trial 18 .

Because there is continuity between the distal fallopian tube and the vagina, the use of body fluids more proximate to the ovary as screening tools is of interest. Somatic mutations in TP53 have been isolated from tampons of women with ovarian cancer 53 . This has also been achieved through uterine lavage 54 . Using multiplexed PCR to detect 18 mutations or aneuploidy in endocervical brushings from 656 patients with endometrial or ovarian cancers and 1002 healthy controls, Wang, et al¸ showed that their test, called PapSEEK, had 33% sensitivity at 99% specificity for ovarian cancer. This improved to 45% sensitivity and 100% specificity in a smaller cohort of 299 women assessed with an intrauterine brushing 55 .

Early detection of ovarian cancer remains an important but, to date, an elusive goal. Efforts to develop efficient and cost-effective ovarian cancer screening have been hampered by the low prevalence of this cancer. A common theme from the clinical trials testing various screening methods is that no single marker has the test characteristics necessary to be a standalone screening test, and to date, no effective strategy exists. Rather, multimodal assessments based on dynamic and algorithmic models are more likely to produce the specificity required for clinical development. Two stage strategies where rising values for blood tests trigger imaging have attained adequate specificity, but not at acceptable sensitivity. Circulating protein biomarkers, autoantibodies, ctDNA and miRNA and proximate fluid collection all deserve further evaluation to enhance the sensitivity of the initial screening stage. In the end, cost effective screening is likely to depend upon an extremely high specificity first screen, followed by a more sensitive secondary imaging strategy.

miRNAs are short (18–24 nucleotide) non-coding RNAs that regulate gene expression through post-transcriptional modification of mRNA transcripts 47 . An individual miRNA may regulate several different genes within a pathway; thus, knowing information about a relatively small number of miRNAs can convey information about thousands of target genes 48 . miRNAs can circulate either bound to the chaperone protein Argonaute 2 (Ago2) or contained within extracellular vesicles (EVs) 49 . They are highly stable in circulation and resistant to ribonucleases 50 . An important property of miRNAs is that they act in a coordinated fashion. Thus, any single miRNA is unlikely to be a reliable biomarker, as compared to a miRNA panel. Using 8 miRNAs, Yokoi et al were able to distinguish early stage ovarian cancers from benign tumors with 86% sensitivity and 83% specificity. Additionally, miRNAs were detectable in EVs collected from cultured ovarian cancer cell lines 51 . Our group has shown that the specificity of miRNA prediction models can be improved by combining next generation sequencing technology with machine learning algorithms 52 . A neural network prediction model was derived using serum miRNA-seq from 98 incident cases of invasive ovarian cancer, including 53 cases of Stage I or II disease, and applied to an independent 454-patient sample set with a disease prevalence of 3.3%. At a sensitivity of 75% and specificity of 100%, the model had an AUC of 0.92 (95% CI 0.82–1.00). Among samples where CA-125 data were available, neither the miRNA signature nor any individual miRNA correlated with CA-125 levels, suggesting miRNAs as an independent disease marker.

The rationale for early detection of ovarian cancer is compelling. Ovarian cancer confined to the ovaries (stage I) can be cured in up to 90% of patients, and disease confined to the pelvis (Stage II) is associated with a 5-year survival of 70%. However, disease that has spread beyond the pelvis (stage III-IV) has a long-term survival rate of 20% or less. Only 20% of ovarian cancers are currently diagnosed in stage I-II 1 . Computer simulations suggest that detection of preclinical disease at an earlier stage could improve survival by 10–30% and would be cost-effective 2 , 3 . The clinical requirements for early detection are stringent. Given the postmenopausal prevalence of 1:2,500, effective screening requires not only high sensitivity for pre-clinical disease of ≥ 75%, but also very high specificity of ≥ 99.7% to achieve a positive predictive value (PPV) ( Table 1 ) of 10% (i.e., 10 operations for each case of ovarian cancer detected). Increasing specificity, rather than improving sensitivity alone or screening only high-risk patient subsets, will have the greatest impact on the positive likelihood ratio (LR+) of a test result ( Table 2 ). Successful early detection strategies for ovarian cancer should diagnose more high grade epithelial ovarian cancers at an early stage and improve outcomes, i.e. overall survival 4 . However, this relies on two basic assumptions 5 . First, one assumes that high grade epithelial ovarian cancers currently diagnosed at an advanced stage, if detected earlier, will have the same favorable prognosis as Stage I cancers, which are heterogeneous and include low grade neoplasms and non-serous histotypes. Secondly, determinations of screening efficacy must control for lead-time bias. Any earlier diagnosis, irrespective of whether it impacts overall survival, will add to a patient’s total survival time from diagnosis. While the survival proportion at any given time point is increased, there is no true improvement in the patient’s life expectancy 6 . Several trials have examined different screening strategies, most notably the Prostate, Lung, Colorectal and Ovarian (PLCO) Screening Trial, the Normal Risk Ovarian Screening Study (NROSS), and the United Kingdom Collaborative Trial of Ovarian Cancer Screening (UKCTOCS). Transvaginal sonography (TVS) and the protein biomarker CA125 are the two best studied screening tools for ovarian cancer. However, neither is sufficient for general screening, exemplified by PLCO Screening Trial 7 . PLCO enrolled more than 70,000 post-menopausal women between 1993 and 2001 to receive usual care or in 39,110 to undergo annual screening with CA125 and TVS, or, in the later rounds of the study, CA125 alone. Overall, 388 ovarian cancers were diagnosed, but 1080 women underwent surgery for a false positive result, with 15% experiencing major complications. Moreover, screening failed to improve survival. With a median follow-up of 14.7 years in each arm, the ovarian cancer death risk ratio was 1.06 (95% CI: 0.87–1.30) between the two groups 8 . The failure of PLCO highlighted the need for a two-tiered screening strategy. Ovarian cancer is associated with rising CA125, and benign disease is not 9 . The Risk of Ovarian Cancer Algorithm (ROCA) relies on each woman’s own baseline CA125 value to determine whether there has been a significant increase 10 . Over the last 17 years in NROSS, 34,637 samples have been obtained from 5,729 postmenopausal women at conventional risk 11 . Rising CA125 judged by the ROCA prompted TVS, and sonographic findings compatible with cancer led to exploratory surgery. Less than 0.9% of these women have been referred for ultrasound after each annual screening, and 2.6% over multiple years on study. In UKCTOCS >200,000 postmenopausal women at average risk were randomized to three groups: control (101,359), annual TVS (50,639) and annual CA125 with ROCA prompting TVS (50,640) 12 . With the algorithm, only 3–4 operations were required to detect each case of ovarian cancer. Excluding prevalent cases and primary peritoneal disease, a 20% reduction in mortality was observed (P=0.021). The statistical bounds around the estimate of reduction in mortality were, however, large and additional follow-up will be required to validate this estimate. In the absence of definitive data, early detection of ovarian cancer remains a critical unmet public health need. In 2018, updated recommendations from the US Preventive Services Task Force (USPSTF) conclude with “moderate certainty” that “the net balance of the benefit and harms of screening is negative” and recommend against screening in average risk asymptomatic women 13 . For women at increased genetic risk (i.e., BRCA1/2 mutation carriers) who delay prophylactic bilateral salpingo-oophorectomy, screening with semi-annual CA125 and TVS is recommended, but there is not yet definitive evidence that this strategy detects fallopian or ovarian cancer earlier or improves outcomes 14 . Two ongoing clinical trials will hopefully shed light on whether use of the ROCA algorithm for screening triage will impact mortality in this population 15 , 16 . Regardless, there remain opportunities to improve the two-tiered screening approach. The discussion below focuses on additional screening strategies beyond CA125 and TVS.

Autoantibodies to mutant proteins can be stimulated by small volumes of cancer in the ovary or fallopian tube, providing greater sensitivity and earlier detection than CA125 or other assays that detect shed biomarkers. Autologous antibodies can be produced against mutant TP53 protein. Alteration in TP53 is the most common genetic mutation among ovarian cancers, seen in up to 96% of high-grade serous carcinomas 21 . At a specificity of 97%, autoantibodies could be detected in 21–30% of serum samples from ovarian cancer patients from MD Anderson, the Australian Ovarian Cancer Study and the UKCTOCS 22 . Among 164 cases with rising CA125 detected in serial preclinical serum samples with the ROCA, 20.7% had elevated TP53 autoantibody. Of the 34 ovarian cancer cases detected with the ROCA, TP53 autoantibody titers were elevated 8 months before CA125. In the 9 cases missed by the ROCA, TP53 autoantibody was elevated 22.9 months before cancer diagnosis. Consequently, TP53 autoantibody levels provide the first bio-marker with clinically significant lead time over elevation of CA125 or an elevated ROCA value. Kaaks, et al , performed a prospective analysis on a selected panel of four autoantibodies—against TP53, CTAG1A, CTAG2 and NUDT11— using serum samples collected up to 36 months before diagnosis from 194 ovarian cancer patients and 705 matched control participants 23 . With lead times less than or equal to 6 months, sensitivity for early detection ranged from 19–23% for the four autoantibodies at 98% specificity, but with lead time of greater than 1-year, sensitivity ranged from only 1–11%. Addition of the four autoantibodies to CA125 did not improve sensitivity for detection at 98% specificity, although serial preclinical specimens were not analyzed. A recent review of the world literature has reported that 6 individual autoantibodies against EpCAM, IL-8, PLAT, MDM2, c-Myc and HOXA7 provide 39–67% sensitivity at 98–100% specificity for detecting ovarian cancer at all stages 24 . These and other candidates are being evaluated in combination with protein biomarkers.