Abstract
Biomarker detection in urine offers a potential solution to increase effectiveness of cervical cancer screening programs by attracting nonresponders. In this prospective study, the presence of high-risk human papillomavirus (hrHPV) DNA and the performance of DNA methylation analysis was determined for the detection of cervical cancer and high-grade cervical intraepithelial neoplasia (CIN2/3) in urine, and compared with paired cervicovaginal self-samples and clinician-taken cervical scrapes.
A total of 587 samples were included from 113 women with cervical cancer, 92 women with CIN2/3, and 64 controls. Samples were tested for hrHPV DNA and five methylation markers. Univariate and multivariate logistic regression and leave-one-out cross-validation were used to determine the methylation marker performance for CIN3 and cervical cancer (CIN3+) detection in urine. Agreement between samples was determined using Cohen kappa statistics and the Spearman correlation coefficients.
HrHPV presence was high in all sample types, 79% to 92%. Methylation levels of all markers in urine significantly increased with increasing severity of disease. The optimal marker panel (ASCL1/LHX8) resulted in an AUC of 0.84 for CIN3+ detection in urine, corresponding to an 86% sensitivity at a 70% predefined specificity. At this threshold 96% (109/113) of cervical cancers, 68% (46/64) of CIN3, and 58% (14/24) of CIN2 were detected. Between paired samples, a strong agreement for HPV16/18 genotyping and a fair to strong correlation for methylation was found.
HrHPV DNA and DNA methylation testing in urine offers a promising solution to detect cervical cancer and CIN2/3 lesions, especially for women currently unreached by conventional screening methods.
Cervical cancer is still the fourth most common cancer in women worldwide, despite widespread implementation of screening programs. A major reason for suboptimal screening effectiveness is low participation rates. Urine sampling, a noninvasive liquid biopsy, offers a promising solution to reach nonresponders. This study is not only the first to determine the performance of high-risk human papillomavirus (hrHPV) DNA and DNA methylation testing in urine for the detection of cervical cancer and high-grade cervical intraepithelial neoplasia (CIN2/3) in a clinical setting, but also the first to make a direct comparison to cervicovaginal self-samples and cervical scrapes. A total of 587 samples of 269 women were tested for hrHPV DNA and five methylation markers. We defined an optimal marker panel for CIN3 and cervical cancer detection in urine, with a cross-validated AUC of 0.84. These findings could greatly improve cervical cancer prevention strategies for women currently unreached by conventional screening methods.
Introduction
Cervical cancer is the fourth most common cancer in women worldwide (1, 2). It develops through precancerous lesions [squamous intraepithelial lesions (SIL) also known as cervical intraepithelial neoplasia (CIN), graded from low-grade (LSIL or CIN1) to high-grade (HSIL or CIN2-3)] and is caused by a persistent infection with hrHPV. While implementation of screening for cervical cancer resulted in a decrease in incidence and mortality, effectiveness of screening is still suboptimal, amongst others due to low participation rates (3, 4). Offering self-sample devices for the collection of a cervicovaginal self-sample or a urine sample, is known to increase the participation rate among nonresponders (5, 6).
Primary hrHPV DNA testing is recommended by several (inter)national cervical cancer screening guidelines, which can be applied on both clinician-taken cervical scrapes as well as on cervicovaginal self-samples (7, 8). Additional triage testing, using for example cytology, HPV16/18 genotyping, or DNA methylation analysis, is necessary to prevent unneeded referral and overtreatment of women with a transient and therefore clinically irrelevant hrHPV infection (9, 10). Known disadvantages of cytology triage are the subjective interpretation, the suboptimal sensitivity, and the disability to apply on self-sampled material. Due to the objective outcome and good sensitivity and specificity for CIN3 and cervical cancer detection, DNA methylation analysis is considered to be a promising triage alternative to cytology (11, 12). DNA methylation is the covalent binding of a methylgroup on a cytosine-guanine dinucleotide, which can result in silencing of tumor suppressor genes and thereby induce cancer development. Levels of DNA methylation of specific genes increase with increasing severity of the cervical lesion and are highest in cervical cancer (13). Combined with hrHPV DNA testing, DNA methylation analysis enables a fully molecular and objective workflow for the detection of cervical (pre)cancerous lesions that can be applied on both clinician-taken cervical scrapes and self-samples (14–16).
Urine for cervical cancer detection has increasingly received attention because it can be easily and noninvasively obtained and is preferred by women over other sampling methods (17–20). Urine collection can be done in a home-based setting, and therefore offers a solution to reach nonresponders within cervical screening programs. HrHPV DNA detection in urine is extensively studied and is considered adequate for detection of high-grade CIN lesions, albeit the sensitivity is slightly lower compared with clinician-taken cervical scrapes (19, 21–26). More recently, our and other research groups have shown the feasibility of DNA methylation analysis in urine in small sample series (27–30). Further comparative analysis of different urine fractions showed most favorable performance of urine sediment to detect CIN3 (29).
This study aimed to test hrHPV DNA and to determine the clinical performance of methylation analysis in urine for the detection of cervical cancer and high-grade CIN lesions, and to make a comparison with conventional samples used for cervical cancer screening (i.e., cervicovaginal self-samples and cervical scrapes). For this purpose, paired urine samples, cervicovaginal self-samples and clinician-taken cervical scrapes of women with cervical cancer, and paired urine samples and cervicovaginal self-samples of women with a CIN2 or CIN3 lesion were collected prior to treatment. Urine samples were compared with urine of healthy female controls.
Materials and Methods
Study population and sample collection
Women with cervical cancer (n = 168) were included between October 2016 and August 2020 (SOLUTION 1 study) in tertiary care cancer centers. A urine sample, a cervicovaginal self-sample, and a clinician-taken cervical scrape were collected after histologic confirmation of cervical cancer and prior to primary cancer treatment. Women diagnosed with a CIN lesion, scheduled for a large loop excision of the transformation zone (LLETZ) procedure (n = 121), were included between October 2017 and February 2021 (SOLUTION 2 study) in colposcopy centers. A urine sample and a cervicovaginal self-sample were collected before the LLETZ procedure. A small subset of urine samples (n = 30) from women with a histologically proven CIN3 lesion, have also been used for technical comparison of the optimal urine fraction for DNA methylation analysis on which we reported earlier (29).
Women with cervical cancer or CIN were instructed to self-collect a complete urine void, irrespective of time of collection and personal hygiene, prior to collecting the cervicovaginal self-sample. In women with cervical cancer, clinician-taken cervical scrapes were collected during gynecologic examination or surgery. If not available, the clinician-taken cervical scrape taken for clinical diagnostics or screening was used. Only complete sample-sets from women with cervical cancer or a CIN2 or CIN3 lesion were included for current analysis.
Urine samples, used as controls, were collected on a voluntary basis by healthy women without a medical history in oncology between April 2019 and July 2020 [Urine Controls (URIC) biobank]. A simulation study was performed to investigate the sufficiency and adequacy of the sample size of controls (Supplemental File 1).
Ethics approval and consent to participate
This study is performed in accordance with the Declaration of Helsinki, and ethical approval was provided by the Medical Ethical Committee of the VU University Medical Centre (Amsterdam, the Netherlands) for the use of samples collected from women diagnosed with cervical cancer (SOLUTION 1 study reference number 2016.213; Trial registration ID: NL56664.029.16), from women diagnosed with a CIN lesion (SOLUTION 2 study reference number 2017.112), and healthy female controls (URIC biobank reference number 2018.657). All women were 18 years or older, and provided written informed consent.
Sample collection and processing
Urine samples and cervicovaginal self-samples were sent to the Pathology department of Amsterdam University Medical Center (UMC), location VUmc (Amsterdam, the Netherlands), and processed within 24 to 72 hours after collection. Urine samples were collected in three 30-mL collection tubes, each prefilled with 2 mL 0.6M Ethylenediaminetetraacetic acid (EDTA), in a final concentration of 40 mmol/L to maintain DNA quality during transport (31). Urine sediment was obtained by centrifugation of 15 mL of urine at 3,000 × g for 15 minutes and stored at −20°C (29). For collecting the cervicovaginal self-sample, a dry brush device (Evalyn Brush, Rovers Medical Devices) was used. After the brush arrived at the laboratory, it was resuspended in 1.5 mL Thinprep Preservcyt medium (Hologic), vortexed, and stored at 4°C.
The clinician-taken cervical scrapes were collected using a Cervex-Brush (Rovers Medical Devices), and directly placed in Thinprep Preservcyt medium (Hologic, Marlborough) and stored at 4°C.
DNA isolation and DNA modification
DNA was isolated from 200 μL urine sediment (15 mL original volume) using the DNA mini and blood mini kit (Qiagen). DNA from cervicovaginal self-samples (13.3% of the original sample) and clinician-taken cervical scrapes (15.0% of the original sample) was isolated using the Microlab Star robotic system. A NanoDrop 1000 (Thermo Fisher Scientific) was used for DNA concentration measurements. Isolated DNA was bisulphite converted using the EZ DNA Methylation Kit (Zymo Research). All procedures were performed according to the recommendations of the manufacturer.
HrHPV DNA testing
Isolated DNA was tested for hrHPV DNA using the HPV Risk assay/QIAscreen HPV PCR Test (Self-screen B.V.). This is an in vitro real-time PCR–based DNA assay for the qualitative detection of 15 hrHPV genotypes (i.e., 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 67, and 68) and partial genotyping for HPV16 and HPV18. The HPV Risk assay is validated on clinician-taken cervical scrapes and cervicovaginal self-samples using a predefined threshold for the detection of clinically relevant infections (32). No threshold was used for the urine samples (33).
Host cell gene DNA methylation analysis by quantitative methylation-specific PCR
Methylation analysis of ASCL1, GHSR, LHX8, SST, and ZIC1 was performed as described before (34, 35) using 50 ng of bisulphite-converted DNA. For quantification and quality control ACTB was used as a reference gene. To assure a good sample quality, samples with a Cq value for ACTB > 32 were excluded for methylation analysis. The methylation levels of all markers were normalized to reference gene ACTB using the comparative Cq method (2-ΔCq × 100) to obtain Cq ratios.
Data analysis
Women without a complete sample-set, with no residual disease in the surgery specimen, or women of which samples were collected at different time points (e.g., before LLETZ and after LLETZ) were excluded from the analyses.
Differences in DNA methylation levels between healthy female controls, and in order of increasing severity of disease, CIN2, CIN3, and cervical cancer were visualized using boxplots and tested for statistical significance using the Kruskal–Wallis test (significance: P < 0.05), followed by pairwise Mann–Whitney U test with Bonferroni correction for three simultaneous comparisons (α = 0.05/3; significance: P < 0.017).
Evaluation of the diagnostic performance of ASCL1, GHSR, LHX8, SST, and ZIC1 methylation in urine to discriminate between controls and women with VIN3 and cervical cancer (CIN3+) was performed using univariate logistic regression. To obtain an optimal marker panel, multivariable logistic regression with stepwise selection based on the lowest Akaike Information Criterion (AIC) was done. Women diagnosed with a CIN2 lesion were excluded for logistic regression analysis, because the morphological CIN2 diagnosis reflects a heterogeneous group with less reproducibility (36). Age of the patient was not included as a covariate in this model, since previous data showed no or slight improvement of performance (35, 37). Leave-one-out cross-validation (LOOCV) was performed to evaluate predictive performance. The predicted probability obtained from LOOCV was used to classify samples according to a predefined specificity of 70% and 80%, similar to previous studies on HPV-positive cervicovaginal self-samples and cervical scrapes (37, 38). ROC curves were made of all individual markers and the marker panel and were quantified as AUC.
The hrHPV-agreement between urine samples, cervicovaginal self-samples, and clinician-taken cervical scrapes was analyzed using Cohen kappa statistics, and was defined as poor (κ ≤ 0.19), fair (κ = 0.20–0.39), moderate (κ = 0.40–0.59), strong (κ = 0.60–0.79), and very strong (κ ≥ 0.80).
Albeit this study was not performed on HPV-positive urine samples, for comparison of the DNA methylation results in urine with the paired self-sample and cervical scrape, predefined and validated thresholds of marker panel ASCL1/LHX8 for CIN3 detection were used. The threshold for HPV-positive self-samples corresponded to a sensitivity of 88% and a specificity of 81% (35), and for HPV-positive clinician-taken cervical scrapes to a sensitivity of 83% and a specificity of 82% (37).
The correlation of DNA methylation levels between paired sample types, and the correlation of DNA methylation levels within sample types was assessed with Spearman rank correlation. The Spearman correlation coefficient (r) was defined as poor (r ≤ 0.19), fair (r = 0.20–0.39), moderate (r = 0.40–0.59), strong (r = 0.60–0.79), and very strong (r ≥ 0.80).
Data was collected using Castor EDC (39), and R Statistical Software (version 3.5.1) with packages pROC and GLM, IBM SPSS statistics software (version 26) and GraphPad Prism (version 8.2.1) were used for statistical analyses and production of graphs. Reported P values are two-sided, and unless reported otherwise P < 0.05 was used as the significance threshold.
The data generated in this study are available upon request from the corresponding author.
Results
Study population and characteristics
Study samples were collected from 168 women diagnosed with cervical cancer (SOLUTION 1 study), of which 55 were excluded for current analysis for various reasons (e.g., incomplete sample-set, no residual cancer at time of sample collection; Supplementary Fig. S1). Ultimately, 113 women with cervical cancer were included in present study with a median age of 47 years [interquartile range (IQR): 38–57 years] with stages ranging from FIGO (2018) IB1 to IVB. Histotypes included squamous cell carcinomas (n = 82), adenocarcinomas (n = 21), adenosquameus carcinomas (n = 6), clear cell carcinomas (n = 2), and neuroendocrine carcinomas (n = 2).
Study samples were collected from 121 women scheduled for a LLETZ procedure (SOLUTION 2 study), of which 29 were excluded for current analysis for various reasons (e.g., incomplete sample-set, no residual dysplasia at time of sample collection; Supplementary Fig. S1). The remaining 92 women diagnosed with a CIN2 or CIN3 lesion included in this study had a median age of 39 years (IQR: 31–46 years). In four cases an adenocarcinoma in situ (ACIS) was diagnosed, three of which cooccurred with a CIN3 lesion. For the analysis all four ACIS cases were analyzed together with the CIN3 cases.
Urine samples, used as controls (URIC biobank), were collected from 64 women with a median age of 52 years (IQR: 39–58 years).
A summary of the study population is shown in Table 1.
Study population and characteristics.
Cervical cancer cases | |
n | 113 |
Median age (IQR) | 47 (38–57) |
Histology | n (%) |
Squamous cell carcinoma | 82 (72) |
Adenocarcinoma, usual type | 21 (19) |
Clear cell adenocarcinoma | 2 (2) |
Neuroendocrine carcinoma | 2 (2) |
Adenosquamous carcinoma | 6 (5) |
2018 FIGO stage | n (%) |
IB1 | 23 (20) |
IB2 | 32 (28) |
IB3 | 26 (23) |
IIA1 | 4 (4) |
IIA2 | 3 (3) |
IIB | 17 (15) |
IIIA | 1 (1) |
IIIB | 4 (3) |
IIIC1 | 1 (1) |
IVA | 1 (1) |
IVB | 1 (1) |
LLETZ before primary treatment | n (%) |
Yes | 25 (22) |
No | 88 (78) |
CIN cases | |
n | 92 |
Median age (IQR) | 39 (31–46) |
Histology | n (%) |
CIN2 | 24 (26) |
CIN3 | 64 (70) |
CIN3 and ACIS | 3 (3) |
ACIS | 1 (1) |
Biopsy before LLETZ | n (%) |
Yes | 71 (77) |
No | 21 (23) |
Healthy female controls | |
n | 64 |
Median age (IQR) | 52 (39–58) |
Cervical cancer cases | |
n | 113 |
Median age (IQR) | 47 (38–57) |
Histology | n (%) |
Squamous cell carcinoma | 82 (72) |
Adenocarcinoma, usual type | 21 (19) |
Clear cell adenocarcinoma | 2 (2) |
Neuroendocrine carcinoma | 2 (2) |
Adenosquamous carcinoma | 6 (5) |
2018 FIGO stage | n (%) |
IB1 | 23 (20) |
IB2 | 32 (28) |
IB3 | 26 (23) |
IIA1 | 4 (4) |
IIA2 | 3 (3) |
IIB | 17 (15) |
IIIA | 1 (1) |
IIIB | 4 (3) |
IIIC1 | 1 (1) |
IVA | 1 (1) |
IVB | 1 (1) |
LLETZ before primary treatment | n (%) |
Yes | 25 (22) |
No | 88 (78) |
CIN cases | |
n | 92 |
Median age (IQR) | 39 (31–46) |
Histology | n (%) |
CIN2 | 24 (26) |
CIN3 | 64 (70) |
CIN3 and ACIS | 3 (3) |
ACIS | 1 (1) |
Biopsy before LLETZ | n (%) |
Yes | 71 (77) |
No | 21 (23) |
Healthy female controls | |
n | 64 |
Median age (IQR) | 52 (39–58) |
Abbreviation: FIGO, International Federation of Gynecology and Obstetrics.
DNA methylation levels in urine
Methylation levels in urine samples are plotted in Fig. 1. From controls, through CIN2 and CIN3 to cervical cancer, all markers showed an increase in methylation levels with increasing severity of disease. Methylation levels were significantly different among the different groups (Kruskall–Wallis omnibus test; all: P < 0.001). Compared with controls all markers were significantly increased in cervical cancer (Mann–Whitney U test; all: P < 0.001), with ASCL1 and LHX8 also in CIN3 (Mann–Whitney U test: P < 0.001).
Levels of DNA methylation in urine. DNA methylation levels of ASCL1, GHSR, LHX8, SST, and ZIC1 in urine samples from healthy female controls, women diagnosed with a CIN2 lesion, CIN3 lesion and cervical cancer. DNA methylation levels are shown by the log2-transformed Cq ratios. Boxplots show medians with lower and upper quartile and range whiskers. The green, orange, and red dots represent the DNA methylation level of all individual cases. A P value, after Bonferroni correction for multiple testing, of 0.017 was considered to be significant. Abbreviation: CNTRL, healthy female controls; CC, cervical cancer.
Levels of DNA methylation in urine. DNA methylation levels of ASCL1, GHSR, LHX8, SST, and ZIC1 in urine samples from healthy female controls, women diagnosed with a CIN2 lesion, CIN3 lesion and cervical cancer. DNA methylation levels are shown by the log2-transformed Cq ratios. Boxplots show medians with lower and upper quartile and range whiskers. The green, orange, and red dots represent the DNA methylation level of all individual cases. A P value, after Bonferroni correction for multiple testing, of 0.017 was considered to be significant. Abbreviation: CNTRL, healthy female controls; CC, cervical cancer.
Performance of DNA methylation analysis in urine for cervical cancer and CIN detection
The ability of the individual markers to distinguish CIN3+ (n = 181) from controls (n = 64) in urine was assessed by univariate logistic regression analysis and resulted in AUCs between 0.73 to 0.85. Upon LOOCV, AUCs ranging from 0.72 to 0.83 were obtained (Fig. 2). To evaluate if a combination of markers could further improve CIN3+ detection in urine, multivariable logistic regression analysis was applied. This resulted in an optimal marker panel of ASCL1 and LHX8, which upon LOOCV yielded an AUC of 0.84 [95% confidence interval (CI), 0.78–0.89) for CIN3+ detection (Fig. 2). Predefined specificities of 70% and 80% corresponded to CIN3+ detection sensitivities of 86% and 77% (Table 2).
Performance of DNA methylation in urine. Performance of ASCL1, GHSR, LHX8, SST, ZIC1, and marker panel ASCL1/LHX8 for CIN3+ detection in urine. Cross-validated ROC curves and corresponding AUCs and corresponding 95% CIs are shown.
Performance of DNA methylation in urine. Performance of ASCL1, GHSR, LHX8, SST, ZIC1, and marker panel ASCL1/LHX8 for CIN3+ detection in urine. Cross-validated ROC curves and corresponding AUCs and corresponding 95% CIs are shown.
Sensitivities at predefined specificities of marker panel ASCL1/LHX8 in urine.
. | Predefined specificitya 70% . | Predefined specificitya 80% . | ||
---|---|---|---|---|
. | 86% . | 77% . | ||
Sensitivity . | ASCL1/LHX8-positive (%) . | ASCL1/LHX8-negative (%) . | ASCL1/LHX8-positive (%) . | ASCL1/LHX8-negative (%) . |
Cervical cancer (n = 113) | 109 (96) | 4 (4) | 106 (94) | 7 (6) |
CIN3 (n = 68) | 46 (68) | 22 (32) | 33 (49) | 35 (51) |
CIN2 (n = 24) | 14 (58) | 10 (42) | 11 (46) | 13 (54) |
Controls (n = 64) | 19 (30) | 45 (70) | 13 (20) | 51 (80) |
. | Predefined specificitya 70% . | Predefined specificitya 80% . | ||
---|---|---|---|---|
. | 86% . | 77% . | ||
Sensitivity . | ASCL1/LHX8-positive (%) . | ASCL1/LHX8-negative (%) . | ASCL1/LHX8-positive (%) . | ASCL1/LHX8-negative (%) . |
Cervical cancer (n = 113) | 109 (96) | 4 (4) | 106 (94) | 7 (6) |
CIN3 (n = 68) | 46 (68) | 22 (32) | 33 (49) | 35 (51) |
CIN2 (n = 24) | 14 (58) | 10 (42) | 11 (46) | 13 (54) |
Controls (n = 64) | 19 (30) | 45 (70) | 13 (20) | 51 (80) |
Note: Corresponding sensitivities and cervical cancer and CIN detection rates and percentages at predefined specificities of 70% and 80% of DNA methylation marker panel ASCL1/LHX8 in urine.
aSensitivities and specificities for CIN3+ detection, based on multivariable logistic regression and LOOCV analysis.
Comparison of hrHPV DNA and DNA methylation results in paired sample types
In Table 3 the counts and percentages of hrHPV DNA and ASCL1/LHX8 methylation positivity as detected in the different sample types are summarized. Since the ASCL1/LHX8 methylation thresholds for self-samples and cervical scrapes previously corresponded to an 80% specificity in HPV-positive women (34, 37), the ASCL1/LHX8 methylation thresholds for urine samples were also set at 80% specificity to compare DNA methylation results among sample types. In women with cervical cancer, 74% (14/19) of urine samples, 69% (9/13) of cervicovaginal self-samples, and 78% (7/9) of clinician-taken cervical scrapes who tested hrHPV-negative, were positive for ASCL1/LHX8 methylation (Supplementary Table S1). In Fig. 3 the ASCL1/LHX8 methylation results per sample are visualized for all women individually. Supplementary Figure S2 provides further clinical information on women with aberrant results.
HPV DNA and DNA methylation results in paired samples.
. | Urine sample . | Cervicovaginal self-sample . | Clinician-taken cervical scrape . |
---|---|---|---|
hrHPV-positive (%) | hrHPV-positive (%) | hrHPV-positive (%) | |
Cervical cancer (n = 113) | 94 (83) | 99a (88) | 104 (92) |
CIN3 (n = 68) | 59 (87) | 57 (84) | – |
CIN2 (n = 24) | 19 (79) | 21 (88) | – |
Controls (n = 64) | 4 (6) | N/A | – |
ASCL1/LHX8-positive (%) | ASCL1/LHX8-positive (%) | ASCL1/LHX8-positive (%) | |
Cervical cancer (n = 113) | 106 (94) | 106 (94) | 111 (98) |
CIN3 (n = 68) | 33 (49) | 36b (53) | – |
CIN2 (n = 24) | 11 (46) | 6 (25) | – |
Controls (n = 64) | 13 (20) | – | – |
. | Urine sample . | Cervicovaginal self-sample . | Clinician-taken cervical scrape . |
---|---|---|---|
hrHPV-positive (%) | hrHPV-positive (%) | hrHPV-positive (%) | |
Cervical cancer (n = 113) | 94 (83) | 99a (88) | 104 (92) |
CIN3 (n = 68) | 59 (87) | 57 (84) | – |
CIN2 (n = 24) | 19 (79) | 21 (88) | – |
Controls (n = 64) | 4 (6) | N/A | – |
ASCL1/LHX8-positive (%) | ASCL1/LHX8-positive (%) | ASCL1/LHX8-positive (%) | |
Cervical cancer (n = 113) | 106 (94) | 106 (94) | 111 (98) |
CIN3 (n = 68) | 33 (49) | 36b (53) | – |
CIN2 (n = 24) | 11 (46) | 6 (25) | – |
Controls (n = 64) | 13 (20) | – | – |
Note: hrHPV- and ASCL1/LHX8-positive result counts and percentages from women diagnosed with cervical cancer, CIN3, CIN2, and controls in paired urine samples, cervicovaginal self-samples, and clinician-taken cervical scrapes. ASCL1/LHX8 threshold corresponds to a specificity of 80% for CIN3+ detection.
aOne cervicovaginal self-sample collected by a woman with cervical cancer tested invalid for hrHPV DNA analysis.
bTwo cervicovaginal self-samples collected by women with a CIN3 lesion tested invalid for DNA methylation analysis.
hrHPV DNA and ASCL1/LHX8 results in paired samples. hrHPV DNA and ASCL1/LHX8 DNA methylation in paired urine samples, cervicovaginal self-samples, and clinician-taken cervical scrapes. hrHPV-positive, -negative, and -invalid samples are shown in black, white, and gray, respectively. ASCL1/LHX8-positive, -negative and -invalid samples are shown in green, red, and gray, respectively. In the “Other” category, the first two columns are clear cell carcinomas, and second two columns are neuroendocrine carcinomas. SCC, squamous cell carcinoma; AC, adenocarcinoma; ASC, adenosquamous carcinoma.
hrHPV DNA and ASCL1/LHX8 results in paired samples. hrHPV DNA and ASCL1/LHX8 DNA methylation in paired urine samples, cervicovaginal self-samples, and clinician-taken cervical scrapes. hrHPV-positive, -negative, and -invalid samples are shown in black, white, and gray, respectively. ASCL1/LHX8-positive, -negative and -invalid samples are shown in green, red, and gray, respectively. In the “Other” category, the first two columns are clear cell carcinomas, and second two columns are neuroendocrine carcinomas. SCC, squamous cell carcinoma; AC, adenocarcinoma; ASC, adenosquamous carcinoma.
Sample agreement and correlation of hrHPV DNA and DNA methylation results
In women with cervical cancer, the overall agreement for hrHPV DNA between urine samples and cervicovaginal self-samples was 89% (moderate agreement; kappa: 0.54; 95% CI, 0.30–0.74), between urine samples and clinician-taken cervical scrapes 89% (moderate agreement; κ: 0.52; 95% CI, 0.23–0.74), and between cervicovaginal self-samples and clinician-taken cervical scrapes 93% (strong agreement; κ: 0.66; 95% CI, 0.41–0.87; Supplementary Table S2). Eleven urine samples and five cervicovaginal self-samples tested hrHPV-negative, while the clinician-taken cervical scrape tested hrHPV-positive. In 8 women with cervical cancer all three samples types tested negative for hrHPV (Fig. 3).
In women with a CIN2 or CIN3 lesion, the overall agreement of hrHPV DNA results between urine samples and cervicovaginal self-samples was 94% (strong agreement; κ: 0.71; 95% CI, 0.51–0.91; Supplementary Table S2). Three urine samples tested hrHPV-negative, while the paired cervicovaginal self-sample tested hrHPV-positive, and three cervicovaginal self-samples tested hrHPV-negative, while the paired urine sample tested hrHPV-positive. Eleven women with a CIN2 or CIN3 lesion tested hrHPV-negative in both urine and cervicovaginal self-sample (Fig. 3).
A very strong agreement for HPV16 and HPV18 was found between paired sample types of both women with cervical cancer (κ: 0.84–1.00) and women with a CIN2 or CIN3 lesion (κ: 0.83–0.88; Supplementary Table S2).
DNA methylation levels of ASCL1, GHSR, LHX8, SST, and ZIC1 among paired sample types were moderately to strongly correlated in women with cervical cancer (Spearman correlation; r: 0.42–0.65), and fairly to moderately correlated between samples of women with a CIN2 or CIN3 lesion (Supplementary Fig. S2).
Discussion
This study presents hrHPV DNA testing and the clinical performance of DNA methylation analysis in urine for the detection of CIN2, CIN3, and cervical cancer. The methylation marker panel ASCL1/LHX8 yielded an AUC of 0.84 to detect CIN3+ in urine. This translates to a sensitivity of 86% at a predefined specificity of 70%. Comparison with paired cervicovaginal self-samples from patients with CIN2, CIN3, and cervical cancer, and paired clinician-taken cervical scrapes from patients with cervical cancer, demonstrated that hrHPV DNA and DNA methylation testing in urine samples offers a promising alternative to detect cervical (pre)cancerous lesions.
Collection of urine is commonly used in clinical practice and is considered easy and noninvasive. Therefore, disease detection in urine has a potential to significantly increase attendance within cervical screening programs (17, 20, 40). In repetitive nonresponders for cervical screening, a subgroup known to be at higher risk to develop cervical cancer (41), a high sensitivity to detect high-grade CIN and cervical cancers is of utmost importance. Over half of all cervical cancer is diagnosed among nonresponders (1, 42). Offering the possibility of a urine test might lead to earlier diagnosis and better prognosis of a substantial amount of women with cervical cancer (43).
Primary hrHPV DNA testing within cervical screening programs currently occurs on clinician-taken cervical scrapes or cervicovaginal self-samples. Cervicovaginal self-sampling, introduced to increase participation, attracts only 5% of women participating in cervical cancer screening and still 44% of women do not participate in screening in the Netherlands (3). Previous studies indicate that nonresponders for cervical screening are willing to collect urine (43, 44). Along other studies (19, 21–24, 26, 45), this study demonstrated the value of hrHPV DNA detection in urine, but a triage test is needed to prevent over referral. DNA methylation analysis has proven to be a valuable triage strategy (14), but is also able to detect primary lesions including the detection of hrHPV-negative lesions (46, 47). This is concordant with 74% of hrHPV-negative urine samples from women with cervical cancer, including all non–HPV-related cancers, were positive for DNA methylation in urine (Supplementary Fig. S2). Since only four urine samples of healthy female controls tested hrHPV-positive, no additional analysis was included to evaluate DNA methylation in urine as a triage test for hrHPV-positive women.
In this study, at a threshold corresponding to 70% specificity and an 86% sensitivity for CIN3+, 68% and 58% of urine samples of women diagnosed with a CIN3 or a CIN2 lesion tested methylation-positive. CIN3 detection in urine is somewhat lower than previous results on cervical scrapes (37). This is in line with present findings on patients with cervical cancer, with a 98% positivity rate on cervical scrapes and a 94% positivity rate on cervicovaginal self-samples and urine at 80% specificity. Nonetheless, women with CIN2 and CIN3 that test methylation negative in urine may not be in need of immediate treatment. The morphologic CIN2/3-diagnosis represents a heterogeneous group known to regress spontaneously in the majority of cases, and overtreatment of these lesions is therefore a potential risk. DNA methylation markers are proposed to differentiate between CIN lesions likely to regress and CIN lesions at risk to progress to cervical cancer, a distinction that cannot be made morphologically (13, 48). It has been shown that a considerable number of CIN2/3 lesions that regress spontaneously have low methylation levels (ref. 49; and unpublished data), while so-called advanced CIN2/3 lesions with a higher cancer risk have high methylation levels (50, 51). Preferably in future screening settings, cervical lesions with low methylation levels remain undetected, and clinically relevant cervical lesions with high methylation levels at risk to progress will be detected. Since all women included in this study were treated, potential overtreatment of methylation negative lesions can only be assumed. Evaluation and revision of the resected tissue during LLETZ, using for example immunohistochemistry as well as methylation analysis of the resected lesions (51), may help to further evaluate this.
From the patients with cervical cancer included in this study, 22% received a diagnostic LLETZ procedure prior to sample collection and before definite primary treatment. Only seven urine samples of women with cervical cancer tested ASCL1/LHX8-negative, of which five also tested negative for hrHPV DNA. These results are explained by minimal residual disease during sampling or by sampling error. Also, 77% of patients with a CIN lesion received a diagnostic biopsy procedure prior to sample collection and the LLETZ procedure. All but one women with a CIN lesion and a hrHPV-negative urine sample had a biopsy before sample collection (Supplementary Fig. S2). These cervical procedures may have influenced our results, since the removal of dysplastic or cancerous cells affects the anatomy and may induce physiologic and immunologic response (52, 53). It is conceivable that hrHPV DNA and DNA methylation analysis will be even more accurate in a primary test situation.
This study is strengthened by the large sample size of 587 prospectively collected paired urine samples, cervicovaginal self-samples, and clinician-taken cervical scrapes that were all tested for hrHPV DNA and five methylation markers. All samples were collected prior to primary treatment and include a histologic endpoint.
A limitation of this study is that the assays used are primarily designed for and validated in cervicovaginal self-samples and clinician-taken cervical scrapes and not urine. Further optimization of the assays, and/or combined testing for HPV-genotyping is expected to increase the detection of CIN2/3 (54). Other limitations are the number and origin of urine samples of female controls, which were not obtained in a screening setting, as well as the absence of paired samples in the controls, and the absence of cervical scrapes from patients with CIN2 and CIN3. Hence to obtain a more accurate comparison of test performance between sample types a future prospective study on paired samples from women without and with cervical (pre)cancer is warranted. Also, the population in this study does not fully reflect the unscreened population for which urine is aimed to offer a solution. Women included in this study, apart from the healthy female control cohort, were screened for cervical cancer or presented with complaints and were all diagnosed with a cervical (pre)cancerous lesion. To further evaluate the potential of urine-based cervical screening, a validation study including a larger study population reflecting a regular screening population with a focus on nonresponders is warranted.
In conclusion, this prospective study indicates that hrHPV DNA and DNA methylation testing in urine can be used to detect cervical cancer and CIN2/3 lesions, and may prove valuable in women currently unreached by conventional screening methods.
Authors' Disclosures
R. van den Helder reports grants from Hanarth Foundation and grants from Stichting Weijerhorst during the conduct of the study. R.D.M. Steenbergen reports other support from Self-screen B.V. outside the submitted work; in addition, R.D.M. Steenbergen has a patent for methylation markers for cervical cancer detection pending to Self-screen B.V. N.E. van Trommel reports grants from Hanarth Foundation during the conduct of the study. No disclosures were reported by the other authors.
Authors' Contributions
R. van den Helder: Conceptualization, resources, data curation, formal analysis, investigation, visualization, methodology, writing–original draft. R.D.M. Steenbergen: Conceptualization, supervision, validation, methodology, project administration, writing–review and editing. A.P. van Splunter: Software, investigation, writing–review and editing. C.H. Mom: Resources, writing–review and editing. M.Y. Tjiong: Resources, writing–review and editing. I. Martin: Data curation, formal analysis, writing–review and editing. F.M.F. Rosier-van Dunné: Resources. I.A.M. van der Avoort: Resources, writing–review and editing. M.C.G. Bleeker: Conceptualization, supervision, funding acquisition, validation, methodology, project administration, writing–review and editing. N.E. van Trommel: Conceptualization, supervision, funding acquisition, validation, writing–original draft, project administration.
Acknowledgments
This research was funded by the Hanarth Foundation and Weijerhorst Foundation. The funders had no role in the design of the study; in the collection, analysis or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.