Somatic mutations in cervicovaginal samples: assessing their role in ovarian cancer detection and prognosis.

Beatriz Pelegrina: Writing – review & editing, Visualization, Formal analysis, Data curation. Sonia Paytubi: Writing – review & editing, Methodology, Investigation. Yolanda Benavente: Writing – review & editing, Formal analysis. Fátima Marin: Writing – review & editing, Methodology. Marta López-Querol: Writing – review & editing, Resources, Project administration. Irene Onieva: Writing – review & editing, Investigation. Jon Frias-Gomez: Writing – review & editing, Validation. Claudia Pavon-Diaz: Writing – review & editing, Investigation. José Manuel Martínez: Writing – review & editing, Resources. Sergi Fernandez-Gonzalez: Writing – review & editing, Resources, Conceptualization. Eduard Dorca: Writing – review & editing, Resources. August Vidal: Writing – review & editing, Funding acquisition. Marc Barahona: Writing – review & editing, Resources. Yolanda Pérez-Escanilla: Writing – review & editing, Resources. Joan Brunet: Writing – review & editing, Resources, Conceptualization. Marta Pineda: Writing – review & editing, Methodology. Lara Pijuan: Writing – review & editing, Resources. Jordi Ponce: Writing – review & editing, Resources, Conceptualization. Xavier Matias-Guiu: Writing – review & editing, Resources, Conceptualization. Laia Alemany: Writing – review & editing, Supervision. Laura Costas: Writing – original draft, Supervision, Project administration, Funding acquisition, Formal analysis, Conceptualization.

Not applicable.

This study was funded by competitive grants from 10.13039/501100004587 Instituto de Salud Carlos III through the projects PI23/00790 and FI24/00247 , intramural CIBERESP 2023 (ESP23PI05), CIBERESP CB06/02/0073 and CIBERONC CB16/12/00231, CB16/12/00234 (cofunded by the 10.13039/501100008530 European Regional Development Fund . ERDF: A way to build Europe), GEICO-23 and CaixaImpulse CI24-10681 . Samples and data were provided by Biobank HUB-ICO-IDIBELL, integrated into the Spanish Biobank Network, and funded by 10.13039/501100004587 Instituto de Salud Carlos III ( PT20/00171 ) and by Xarxa de Bancs de Tumors de Catalunya (XBTC) sponsored by Pla Director d’Oncologia de Catalunya. This work was supported in part by AECC , Grupos estables ( GCTRA18014MATI ), Retos Investigación grant CPP2022-009817 . It also includes the support of the Secretariat for Universities and Research of the Department of Business and Knowledge of the Generalitat de Catalunya and grants to support the activities of research groups 2021SGR01354 and 2021SGR1112 .

Supplemental Table 1 shows the clinical and epidemiologic information of 43 patients with ovarian cancer and 99 controls included in this study. Age and body mass index were not significantly different (p = 0.846 and 0.869, respectively) between cases and controls. Most tumors were late-stage (61 % at FIGO stage III-IV), high-grade (86 % were grade 3), and had a serous histology (77 %). The median raw coverage was 20,102 × for tumor samples, 19,217 × for endometrial aspirates, 20,754 × for Pap smears, and 22,639 × for vaginal samples. The median final coverage (after deduplication and filtering) was 2,196 × for tumor samples, 2,044 × for endometrial aspirates, 2,381 × for Pap smears, and 2,592 × for vaginal samples. The final coverage of the samples from these cases is presented in Supplemental Table 2. We estimated that a minimum coverage of 1,000 × is required to detect variants at a frequency of 0.5 % or higher in non-invasive samples. All non-invasive samples had coverage greater than 1,000 ×, except for ten Pap smears from the controls (ranging from 219X to 995X). Pathogenic and likely pathogenic variants were identified in 25.6 % (95 % confidence interval [CI-: 13.5–41.2) of Pap smear samples from patients with ovarian cancer ( Table 1 , Fig. 1 ). The median variant allele frequency (VAF) in the positive Pap smear samples was 2.10 % (IQR: 1.06–5.06, Supplemental Table 2, Fig. 2 ) in cases and 0.99 % (IQR: 0.63–1.57) in controls (p-value = 0.003). The genes most frequently mutated per sample among the cases were TP53, ARHGAP35 , and CTCF , each of which was mutated in 18.2 % (2/11) of the positive samples (Supplemental Figure 2). Somatic variants were also detected in 33.3 % of control samples, leading to a specificity of 66.7 % (95 % CI: 56.5–75.8; Table 1 ). No cancer was detected among the controls with positive results after a median follow-up of 3.4 years (IQR: 2.8-4.2). The specificity was similar when samples from women with bleeding symptoms were excluded (64.6 %, 95 % CI: 53.0–75.0; data not shown) and when samples with <1,000 X coverage were excluded (64.0 %, 95 % CI: 53.2 %–73.9 %; data not shown). Among the controls, the most frequently mutated genes were TP53 (29.6 %, 8/27 positive samples), PIK3CA (22.2 %, 6/27), and JAK1 (22.2 %, 6/27). Table 1 Diagnostic accuracy of the NGS panel in vaginal samples, Pap smears, endometrial aspirates and tumor samples, excluding variants of uncertain significance. Table 1 dummy alt text Test positive Test negative Sensitivity, % (95 % CI) Specificity, % (95 % CI) Pathogenic and likely pathogenic variants Pap smear samples Cases 11 22 25.6 (13.5-41.2) 66.7 (56.5-75.8) Controls 33 66 Paired samples: Tumor samples Cases 9 4 69.2 (39.6-90.9) NA NA Endometrial aspirates Cases 7 2 77.8 (40.0-97.2) 0.0 (0.0-41.0) Controls 7 0 Vaginal samples Cases 2 12 14.3 (1.78-42.8) 84.6 (54.6-98.1) Controls 2 11 NA = not applicable. CI = confidence interval. Fig. 1 Heatmap depicting pathogenic and likely pathogenic variants per gene by variant allele frequency in Pap smear samples from cases. This graph shows the number of mutations in Pap smear samples. The genes are indicated on the y- axis, and the 43 samples assessed from ovarian cancer cases are indicated on the x- axis. Fig 1 dummy alt text Fig. 2 Variant allele frequency distributions of selected variants in vaginal samples, clinician-collected samples, endometrial aspirates and tumor samples. This graph shows the VAF for each of the observed pathogenic and likely pathogenic variants according to sample type. The VAFs are indicated on the y- axis, and the identified variants are indicated on the x- axis. Colors denote individual patients, with each color corresponding to a unique patient ID; multiple points of the same color indicate variants detected across different sample types from the same individual. Grey dots indicate samples without paired specimens or those in which no variants were detected in the corresponding paired samples. Fig 2 dummy alt text Diagnostic accuracy of the NGS panel in vaginal samples, Pap smears, endometrial aspirates and tumor samples, excluding variants of uncertain significance. NA = not applicable. CI = confidence interval. Heatmap depicting pathogenic and likely pathogenic variants per gene by variant allele frequency in Pap smear samples from cases. This graph shows the number of mutations in Pap smear samples. The genes are indicated on the y- axis, and the 43 samples assessed from ovarian cancer cases are indicated on the x- axis. Variant allele frequency distributions of selected variants in vaginal samples, clinician-collected samples, endometrial aspirates and tumor samples. This graph shows the VAF for each of the observed pathogenic and likely pathogenic variants according to sample type. The VAFs are indicated on the y- axis, and the identified variants are indicated on the x- axis. Colors denote individual patients, with each color corresponding to a unique patient ID; multiple points of the same color indicate variants detected across different sample types from the same individual. Grey dots indicate samples without paired specimens or those in which no variants were detected in the corresponding paired samples. Similar results were obtained when only pathogenic variants were considered (overall sensitivity, 20.9 %; 95 % CI: 10.0-36.0; specificity, 72.7 %; 95 % CI: 62.9-81.2; data not shown). In contrast, when variants of uncertain significance were included, the sensitivity increased to 81.4 % (95 % CI: 66.6-91.6), but the specificity decreased to 26.3 % (95 % CI: 17.9-36.1; Supplemental Table 3, Supplemental Figure 3, Supplemental Figure 4). Restricting variants to a VAF of less than 30 % to exclude potential germline mutations resulted in a sensitivity of 74.3 % (95 % CI: 56.7–87.5) and a specificity of 34 % (95 % CI: 24.6–44.5; data not shown). We observed pathogenic and likely pathogenic variants in 30.3 % (10/33) of the patients with serous cancers and 10.0 % (1/10) of the patients with other histologies (Supplemental Table 2). The sensitivity of detecting these variants is greater in advanced-stage and high-grade cancers. Specifically, the sensitivity was 34.6 % for stage III or IV cancer and 29.7 % for grade 3 cancer. In contrast, the sensitivity was much lower for early stage cancers (12.5 % for stages I-II) and lower-grade cancers (0 % for grades 1-2). The median follow-up duration among the cases was 39.98 months (IQR: 34.09–49.81). Among cases with pathogenic and likely pathogenic variants in the Pap smear samples, 72.7 % (8/11) experienced relapse, whereas 46.9 % (15/32) of patients without these variants experienced relapse. Overall survival and disease-free survival were poorer among women who tested positive than those who tested negative (P = 0.0061 for overall survival and P = 0.052 for disease-free survival; Fig. 3 ). According to the Cox regression models adjusted for age and stage, the HR was 1.71 (95 % CI = 0.70–4.20, P = 0.243) for disease-free survival and 4.27 (95 % CI=1.06-17.23, P = 0.042) for overall survival. Fig. 3 Overall and disease-free survival according to the presence of selected variants in Pap smear samples from cases. This graph shows the Kaplan‒Meier curves for disease-free survival (top) and overall survival (bottom) according to the presence of pathogenic and likely pathogenic variants in Pap smear samples from cases. Fig 3 dummy alt text Overall and disease-free survival according to the presence of selected variants in Pap smear samples from cases. This graph shows the Kaplan‒Meier curves for disease-free survival (top) and overall survival (bottom) according to the presence of pathogenic and likely pathogenic variants in Pap smear samples from cases. Pathogenic or likely pathogenic variants were detected in 69.2 % (9 /13) of tumor samples ( Table 1 and Supplemental Figure 5). The median VAF in positive tumor samples was 48.24 % (IQR: 34.30-63.54; Supplemental Table 2, Fig. 2 ). The most frequently mutated genes were TP53 (55.6 % of positive samples, 5/9), ARID1A (33.3 %, 3/9), and KRAS (33.3 %, 3/9; Supplemental Figure 2). Pathogenic or likely pathogenic variants were observed in 77.8 % (7/9) of endometrial aspirate samples ( Table 1 , Supplemental Figure 5). However, all endometrial aspirates from the controls presented at least one pathogenic or likely pathogenic variant, indicating a lack of specificity ( Table 1 , Supplemental Figure 6). The median VAF in positive endometrial aspirates was 0.87 % (IQR: 0.77–1.50, Supplemental Table 2, Fig. 2 ) in cases and 1.26 % (IQR: 0.84–1.36) in controls (p-value = 0.80). Among the cases, the most frequently mutated genes were PIK3CA (85.7 % of the positive samples, 6/7), TP53 (57.1 %, 4/7), and PPP2R1A (42.9 %, 3/7; Supplemental Figure 2), and among the controls, PIK3CA, PPP2R1A, and ARHGAP35 (71.4 % each), while only one sample (14.3 %) harbored mutations in TP53 . Potentially pathogenic variants were detected in 2 of the 13 samples (14.3 %), specifically in the JAK1 and ARID1A genes (Supplemental Figure 5). The median VAF in positive vaginal samples was 1.63 % (IQR: 1.47–1.80; Supplemental Table 2, Fig. 2 ) in cases and 1.05 % (IQR: 0.89–1.21) in controls (p-value = 0.80). Two vaginal samples from the controls also revealed pathogenic variants in CMSD3 and SETD1B , yielding a specificity of 84.6 % (CI: 54.6–98.1; Supplemental Figure 6). Analysis of paired samples revealed some overlapping variants, particularly when variants of uncertain significance were considered (Supplemental Figure 5 and 7). This overlap was substantially reduced when restricting the analysis to pathogenic and likely pathogenic variants, although a few shared alterations persisted in TP53, PIK3R1 , and PTEN between paired tumors and endometrial aspirates (Supplemental Figure 5). . Notably, one woman with low-stage endometrioid ovarian cancer exhibited pathogenic variants in ARID1A across paired tumor, vaginal, and endometrial aspirates, while this mutation was not detected in the corresponding Pap smear (Supplemental Figure 7). Additional pathogenic variants in KRAS, RNF43 , and PIK3R1 were also detected in her paired tumor and endometrial aspirates (Supplemental Figure 7).

Participants were enrolled in a prospective case-control study (Screenwide, 2017–2021) conducted in Spain, which focused on improving early detection methods for endometrial and ovarian cancer [ 21 ]. Consecutive patients with incident ovarian cancer were invited to participate in the study. Controls were enrolled from visits for presumed benign ovarian masses (n = 12), polyps (n = 21), endometriosis (n = 2), myomas (n = 16), abnormal bleeding (n = 17), or regular check-ups (n = 31) and were matched by age in ±5 years groups to cases. The exclusion criteria were pregnancy, puerperium, treatment with chemotherapy or radiotherapy during the previous six months, and communication problems that precluded the provision of informed consent. The inclusion criteria were intact uterus, fallopian tubes, and ovaries and, for cases with an incident diagnosis of ovarian cancer. A predefined form was used to extract clinical data from electronic medical records. The reference standard for diagnosing cases consisted of histological data obtained after biopsy or hysterectomy. Histological confirmation of all cases was performed on hysterectomy-oophorectomy samples, with the exception of two patients who were deemed unsuitable for surgery. In these two cases, the ovarian cancer diagnosis was confirmed via cytological analysis of the ascitic fluid. At enrollment, participants donated vaginal samples, clinician-collected cervical samples (Pap smear samples), endometrial aspirates, and, when available, tumor samples. Vaginal samples were collected during office visits using an Evalyn® Brush device (Rovers Medical Devices) and suspended in 5 mL of ThinPrep liquid-based solution (Hologic). Pap smears were collected using a Cervex brush (Rovers Medical Devices) and suspended in 20 mL ThinPrep liquid-based solution (Hologic). Endometrial aspirates were collected using a pipette. Supplemental Figure 1 shows a graphical summary of the study. Overall, 43 patients with OC and 99 controls were included in this study. We analyzed 198 samples, including one Pap smear per participant (n = 142), 13 ovarian tumor samples from cases, 16 endometrial aspirates (from 9 cases and 7 controls), and 27 vaginal samples (from 14 cases and 13 controls). Pap smears were prioritized as the primary specimen type because they represent the standard approach in cervical cancer screening and offer a practical, minimally invasive method for molecular analysis. Vaginal self-samples were included given their growing relevance in screening programs that are transitioning from clinician-collected Pap smears to self-collection. Endometrial aspirates were incorporated to explore whether sampling closer to the uterine cavity could improve detection. Tumor samples were included to provide a reference for mutation profiles and enable comparison with the rest of samples. Additional samples were randomly selected from participants who consented to provide more than one specimen type during surgery or clinical evaluation. Due to the exploratory nature of these analyses and logistical constraints, we aimed to include approximately 15 samples for each additional specimen type and around 30 vaginal samples to enable meaningful comparisons while maintaining feasibility. We used an NGS custom panel targeting the exonic regions and intron‒exon boundaries of 47 genes frequently mutated in endometrial and ovarian cancers using an algorithm that maximized case detection with the minimum number of selected genes, as previously described [ 12 , 21 ]. Genes predominantly implicated in hereditary cancer syndromes (e.g., BRCA, MMR) were excluded to restrict the analysis to somatic alterations. The panel included TP53, PIK3CA, PIK3R1, ARID1A, KRAS, PPP2R1A, AKT1, APC, PTEN, CDKN2A, BRAF, NRAS, and PAX2, among others. DNA was isolated from all specimens using the automated Maxwell® 16 Instrument (Promega Corporation, Madison, WI, USA), which processes up to 16 samples simultaneously. DNA concentration was measured using the Qubit dsDNA Broad Range Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA), and only samples with ≥250 ng of DNA were included in the analysis. For Pap smears and vaginal samples, DNA extraction was performed using the Maxwell® 16 LEV Blood DNA Kit (Promega Corporation), following the protocol described in Pelegrina et al., 2023 [ 12 ]. Briefly, samples were centrifuged at 11,000 rpm for 20 minutes, and the resulting cell pellets were treated with proteinase K at 56 °C for 20 minutes prior to extraction. DNA was eluted in 50 μL of nuclease-free water. For endometrial aspirates and tumor tissue (FFPE or fresh-frozen), DNA was extracted using the Maxwell® 16 Tissue DNA Purification Kit (Promega Corporation). Frozen aspirate biopsies and surgical resection specimens were processed using the same kit, and DNA was eluted in 300 μL of nuclease-free water. Libraries and target enrichment were prepared following the KAPA HyperCap Workflow v3.0, with modifications to include duplex unique molecular identifiers (UMIs) and to maintain blind case–control status. DNA samples, ranging from 250 ng to 500 ng, were fragmented to 150–350 bp using Frag Enzyme and then end-repaired and A-tailed before ligation to barcoded xGen CS Adapters. Libraries were cleaned, size-selected with AMPure XP beads, and PCR-amplified with UDI Dup Seq primers. For enrichment, libraries were multiplexed in groups of 10, totaling 2 μg of DNA, and hybridized with KAPA HyperChoice MAX probes. Captured DNA was washed, amplified, and purified. The final libraries were quantified using a Qubit fluorometer, and a portion was assessed for quality using an Agilent Bioanalyzer DNA 1000 assay. High-depth sequencing was then performed on an Illumina NovaSeq 6000 platform using an SP flow cell and a 150 bp paired-end protocol. We used a bioinformatics pipeline combining Picard, fgbio, and BWA tools as previously described [ 12 , 21 ]. Variants were filtered based on quality and functional impact criteria. Variants with a population frequency of < 0.1 % were retained to exclude polymorphisms. Each gene in the panel was classified as a tumor suppressor, oncogene, ambiguous, or not driven based on the IntOgen classification [ 22 ] and literature review. All mutations found in the same hotspot codon were selected. For tumor suppressor genes, all nonsense, frameshift, and splice-site variants were selected. For oncogenes, tumor suppressors, and ambiguities, all mutations found in the same codon of a hotspot were selected. Variants were classified into the following categories: 1) benign and likely benign, 2) variants of unknown significance, 3) likely pathogenic, and 4) pathogenic, on the basis of ClinVar [ 23 ] and InterVar [ 24 ]. Variants of unknown significance (<5 %) were excluded from tumor samples. Benign and likely benign variants were excluded and the results were reported according to the remaining three categories. Supplemental Material includes results with variants of unknown significance for comparison with previous studies [ 14 ], while the main results focus mostly on likely pathogenic and pathogenic variants. Samples with at least one selected variant were classified as positive and those without selected variants were classified as negative. Descriptive analyses were performed using medians and interquartile ranges (IQRs) per participant for continuous data, and counts and percentages for categorical data. Fisher’s exact test was used to analyze categorical variables. The test performance was evaluated using sensitivity, specificity, and 95 % confidence intervals, which were calculated using the epiR R package (version 2.0.57). Survival analyses were performed using the survival R package (version 3.5-5) [ 25 ]. In particular, we analyzed survival using Kaplan-Meier curves with log-rank tests and Cox proportional hazards models adjusted for age (tertiles). The proportional hazard assumption was confirmed using the cox.zph function in the survival R package. The start time was defined as the time at which each individual was enrolled and whose samples were collected. For overall survival, the time to death was calculated from the sample collection date to the date of death from any cause. Patients who were still alive on the censoring date were censored at the last follow-up. For disease-free survival, the time to event was measured from sample collection to either the first evidence of recurrent or progressive disease or death due to the disease. Patients who were alive and disease-free at the censoring date or died from unrelated causes were censored at the last follow-up. Censoring included 20 (46.5 %) women for disease-free survival and 34 (79.1 %) for overall survival. We determined that a sample size of 50 cases was necessary to achieve an estimate of 33 % sensitivity [ 13 ], with a 95 % confidence level and 0.20 precision, taking into account a 43 % disease prevalence among women undergoing surgery for ovarian pathology selected with transvaginal ultrasound [ 26 ]. For survival analyses, a sample size of 50 cases was estimated to provide approximately 63 % power to detect a hazard ratio (HR) of 2.0, 93 % power to detect an HR of 3.0, and 98 % power to detect an HR of 4.0, assuming an event rate of 41 % at 3 years of follow-up [ 27 ], using a one-sided test with a 95 % confidence level.

Ovarian cancer is the most lethal type of gynecological cancer [ 1 , 2 ]. Most ovarian cancer patients are diagnosed at a late stage because early stage symptoms are typically vague and nonspecific, and effective screening methods are lacking [ 2 , 3 ]. The five-year relative survival rate for patients with ovarian cancer varies significantly, rising to 93 % when detected early at a localized stage but plummeting to 30 % when diagnosed at a distant stage [ 1 ]. Late-stage diagnosis remains the primary driver of poor survival in ovarian cancer, underscoring the urgent need for novel approaches to early detection and prognostication. Despite substantial efforts, an effective strategy for widespread ovarian cancer screening in the general population is yet to be developed [ 3 ]. Extensive evaluation of serum CA-125 levels and transvaginal ultrasound yielded low accuracy or did not significantly reduce mortality in two large randomized trials [ 3 , 4 ]. The specificity of CA-125 measurements is low, as serum levels are elevated in many benign diseases [ [4] , [5] , [6] , [7] , [8] ]. Similarly, transvaginal ultrasonography is subjective and has poor discriminatory performance [ 4 , 9 ]. Molecular development using non-invasive samples may help develop new detection methods for this lethal disease. Cervical Pap smears have historically been used to detect premalignant lesions of the cervix in population screening programs to reduce the cervical cancer burden [ 10 ]. Similarly, vaginal self-samples are now being used in many cervical cancer screening programs [ 11 ]. These non-invasive samples are now the focus of new molecular approaches for the detection of other gynecological cancers, such as ovarian and endometrial cancers [ [12] , [13] , [14] ]. Some studies have evaluated DNA mutations in Pap smears or vaginal samples to detect ovarian cancer [ [13] , [14] , [15] , [16] , [17] , [18] , [19] ]. The biological plausibility for detecting tumor-derived DNA in the lower genital tract is supported by the anatomical continuity of the female reproductive tract and evidence that exfoliated tumor cells and malignant material can migrate through the fallopian tubes and uterine cavity [ 15 , 20 ]. This provides a rationale for exploring somatic mutations in cervicovaginal samples not only for detection but also for prognostication. Overall, these studies showed limited accuracy in detecting ovarian cancer, which may be attributed to factors such as the small number of mutations or genes analyzed (ranging from 1 to 18 genes) or the absence of control groups. In addition, no study has assessed the potential of evaluating somatic variants in minimally invasive gynecological samples to predict ovarian cancer prognosis. Samples obtained using cervical or vaginal brushings offer a practical and patient-friendly alternative for molecular analysis. These samples may capture tumor-derived DNA shed through the reproductive tract, providing a non-invasive window for tumor biology. Studying somatic variants in these samples could provide earlier prognostic insights, inform treatment decisions, and ultimately improve patient outcomes. There is an urgent clinical need to develop effective, non-invasive methods for early detection and risk stratification of ovarian cancer. In this study, we evaluated 142 Pap smears from 43 patients with ovarian cancer and 99 controls using a sensitive next-generation sequencing (NGS) approach. In addition, we evaluated somatic variants in paired tumor, endometrial, and vaginal samples from a subgroup of patients. Finally, we evaluated survival outcomes according to the sequencing results of non-invasive samples from patients with ovarian cancer.

This study investigated the use of non-invasive samples for detecting ovarian cancer using an NGS-based test. We applied this test to 198 samples, including Pap smear, vaginal self-samples, endometrial aspirates, and tumor samples. Key findings revealed that 26 % of ovarian cancer patients and 33 % of controls had pathogenic or likely pathogenic variants in the Pap smear samples, rendering them ineffective as a distinguishing diagnostic tool. Although still low, the sensitivity was higher in advanced-stage and high-grade cancers. Among the paired samples, 69 % of the tumor samples harbored pathogenic or likely pathogenic variants. Endometrial aspirates revealed somatic variants in 78 % of cases; however, somatic variants were found in all aspirates from controls. Vaginal samples had a much lower mutation detection rate, with only two out of 14 samples from cases positive for somatic variants. Several studies have explored minimally invasive techniques for ovarian cancer detection [ 28 ]. In previous studies, NGS panels have been used to assess the diagnostic potential of detecting ovarian cancer through methods such as uterine lavage, pipelle, Pap smears, Tao brush samples, and cervicovaginal self-samples. Cervicovaginal samples demonstrated sensitivities ranging from 22 % to 38 % [ 13 , 14 , 17 ] and specificities ranging from 94 % to 99 % [ 13 , 14 ], except for a pilot study in which mutations were observed in 100 % of cases (n = 11), but without a control group [ 16 ]. The specificity (66.7 %) and sensitivity (25.6 %) of Pap smears for detecting somatic variants observed in the present study confirm their limited diagnostic utility, consistent with prior evidence that single-modality approaches are insufficient for ovarian cancer detection. In comparison, endometrial samples, including Tao brushes, pipelles, or uterine lavages, showed sensitivities ranging from 30 % to 80 % and specificities ranging from 70 % to 100 % [ 13 , 14 , 20 ]. The outcomes observed with cervicovaginal samples were largely consistent with those of our study, although our specificity was somewhat lower. This difference may be attributed to the broader gene panel used in our study, which included 47 genes, compared to the 8-18 genes analyzed in other studies [ 13 , 14 , 16 , 20 ]. The inclusion of more genes likely increases the probability of detecting mutations in healthy controls. Additionally, most studies included younger controls than cases, who typically had a lower mutation burden in their normal endometrium [ 29 ], possibly contributing to the higher specificity observed in their study. Jiang et al. did not include a control group, which precludes evaluation of the false-positive rate in their analysis [ 16 ]. Interestingly, unlike studies that included healthy controls [ 13 , 19 ], Maritschnegg et al. used controls with benign conditions, such as ovarian masses suspected to be ovarian cancers, which aligns with our approach. They evaluated somatic mutations in uterine lavages from 30 cases and 27 controls and reported a specificity of 70 % and 80 % sensitivity by combining NGS and singleplex analysis restricted to cases with mutations in tumor samples, which probably overestimated the sensitivity [ 20 ]. This underscores the importance of selecting appropriate participants that closely mimic the clinical setting of the target condition and enhances the relevance and applicability of the findings. Our study also revealed that 69 % of tumor samples harbored pathogenic and likely pathogenic variants. Pathogenic TP53 mutations were found in 5 of 9 positive high-grade serous tumors. One additional case carried a TP53 variant of uncertain significance. However, three high-grade serous tumors, including one stage I case, showed no TP53 alteration. Possible reasons include technical factors, biological exceptions, or panel limitations. Two clear cell and two endometrioid tumors were also TP53 -negative, consistent with their lower mutation frequency [ 30 ]. When variants of uncertain significance were included, all tumor samples presented with at least one mutation. Van Bommel et al. also considered such uncertain significance variants and reported a 38 % sensitivity when Pap smear samples were used [ 14 ]. In contrast, we observed a 26 % sensitivity when these variants were excluded, but our sensitivity exceeded 80 % when these variants were included. Notably, the overlapping variants in the paired samples were mostly of uncertain significance; however, some were likely germline variants, as suggested by their high VAF. These differences underscore the variability in detection rates and suggest the need for further refinement of molecular detection techniques for cervicovaginal samples. Similarly, Nair et al. sequenced uterine lavage fluid and found that approximately 50 % of women with benign pathology carried cancer-associated mutations at allele fractions of 1–30 % [ 31 ]. The detection of variants in a substantial proportion of controls highlights the challenge of distinguishing cancer-derived DNA from background alterations. Such background alterations complicate interpretation and could reduce the specificity of any mutation-based screening test. Age-related clonal events and other non-malignant processes, such as clonal hematopoiesis of indeterminate potential (CHIP), may contribute to these findings and introduce overlapping mutations into circulating or exfoliated DNA. Distinguishing truly cancer-derived DNA from this background noise is indeed challenging. Clonal hematopoiesis of indeterminate potential (CHIP) and other non-malignant processes can introduce overlapping mutations into circulating DNA [ 32 ]. These observations underscore the need for caution when interpreting mutation signals from non-invasive samples. While the use of unique molecular identifiers and error-correction methods can mitigate technical artifacts, biological sources of background variation remain an important limitation that future studies must address. Ultimately, integrating multiple lines of evidence may help interpret low-frequency variants in this context. A multimodal approach that combines high-sensitivity sequencing, orthogonal biomarkers such as vaginal or cfDNA methylation biomarkers [ 33 , 34 ], and refined clinical risk models [ 35 ] may improve the accuracy and enhance the early detection of ovarian cancer. The symptoms of ovarian cancer are often nonspecific and contribute to the advanced stage of the disease at the time of diagnosis. Symptoms include abdominal bloating, early satiety, changes in bowel habits, back pain, urinary symptoms, fatigue, and weight loss, which typically appear months prior to diagnosis [ 2 ]. The initial diagnostic steps generally involve measuring CA125 levels and performing pelvic ultrasound, which has low accuracy, especially for early stage tumors. Although previous research has demonstrated low-to-moderate accuracy in detecting this disease, no study has fully leveraged genomic information to predict survival outcomes. We conducted exploratory survival analysis to evaluate whether mutation detection in these samples might also be associated with clinical outcomes. We showed that mutations detected in Pap smears are associated with poorer overall survival, highlighting the potential of using genetic information from these samples for risk assessment and personalized treatment strategies. However, this finding was based on a limited number of events, and may be confounded by unmeasured factors such as treatment modalities and molecular subtypes. Importantly, this analysis was adjusted for stage, and the association between mutation positivity and reduced overall survival remained statistically significant. This suggests that the presence of detectable mutations may provide prognostic information beyond stage alone; however, this observation is exploratory and requires validation in larger, independent cohorts. Nevertheless, most mutation-positive cases had high-grade serous histology and advanced disease, both of which were associated with poor prognosis. While surgical staging remains the clinical gold standard for assessing disease extent, mutation signals from non-invasive samples could potentially support early risk stratification in specific scenarios—such as triaging patients when staging information is delayed or incomplete—and inform treatment prioritization in resource-limited settings. Further studies with more comprehensive clinical annotations are necessary to validate these findings and define the utility of such biomarkers in a prognostic context. We employed rigorous reference tests for ovarian cancer and systematically included all consecutive cases over a specified period to reduce the risk of selection bias. The sample size was moderate, yet it ranked among the largest studies in the field, with only two studies being larger [ 13 , 19 ], as most prior research has been limited to pilot studies with fewer than 30 cases [ 14 , [16] , [17] , [18] , 20 ]. Power calculations for sensitivity were based on a 43 % disease prevalence among women undergoing surgery for ovarian pathology selected using transvaginal ultrasound [ 26 ]. Previous studies have reported ovarian cancer prevalence ranging from 10 % to 55 % [ 35 , 36 ]; however, in primary care, the prevalence has been estimated to be as low as 0.023 % [ 35 ], which would require thousands of participants to adequately assess the sensitivity in this setting. Despite this, the sensitivity was expected to be low, and this study is novel in its focus on survival differences based on the presence of somatic mutations in non-invasive samples, for which it was more powered. However, the sample size was limited to early stage ovarian cancer patients, given that most ovarian cancers are diagnosed at a late stage, and the results have not yet been validated in independent cohorts. We included both cervical clinician-collected and vaginal self-collected samples to increase the generalizability of our findings and to allow for an assessment of the potential of self-sampling. In contrast to previous studies, controls were matched to ovarian cancer cases by age, and the models were adjusted for age, ensuring that any observed differences were more likely attributable to the presence or progression of ovarian cancer rather than to age-related factors.

Our findings suggest that DNA mutations in Pap smear samples may not serve as reliable diagnostic tools for detecting ovarian cancer. Alternative sample types did not outperform Pap smears: endometrial aspirates lacked specificity (0 % vs. 67 % for Pap smears), and vaginal samples lacked sensitivity (14.3 % vs. 25.6 % for Pap smears). However, the presence of mutations in Pap smears was associated with poorer overall survival, indicating potential prognostic value that warrants further investigation. Future research could explore multimodal strategies, such as integrating imaging modalities (transvaginal ultrasound and MRI) with molecular profiling including serum biomarkers (e.g., CA-125, HE4, or other emerging protein signatures), and liquid biopsy approaches including genomic and epigenetic markers, potentially supported by AI-driven risk models, to enhance early detection and improve patient outcomes.

This study received approval from the Ethics Committee for Clinical Research at Bellvitge University Hospital (reference: PR128/16). Prior to any study-related activities, all eligible participants provided written informed consent after being informed about the study. The research adhered to national and international ethical standards and data protection regulations, including the Declaration of Helsinki and its amendments, EU Regulation 2016/679, and Spanish data protection laws (Organic Law 3/2018; Law 14/2007 on biomedical research). The study was registered with the National Register of Biobanks/Collections (C.0004389).

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: L.C. has received a donation from GSK. JHcispoly and IDIBELL signed a service agreement for another project on endometrial cancer. The remaining authors declare no potential conflicts of interest.

The data are available upon reasonable request from the corresponding author at [email protected].