?Background
Advances in cervical cancer screening largely revolve around self-sampling strategies and molecular testing. In this regard, DNA methylation has emerged as biomarker of interest due to its early occurrence in carcinogenesis as well as applicability to vaginal self-samples (VSS) and urine samples. Although recent studies report promising results, methylation detection techniques are limited by the number of methylation biomarkers that can simultaneously be detected. Moreover, bisulfite conversion is known as the golden standard, yet fragments DNA and consequently lowers accuracy especially in low-concentrated samples, such as self-samples. To enhance clinical accuracy in urine, sampling only the first fraction of the urine stream, i.e. first-void urine (FVU), has proven effective due to the capture of biomarker-rich material deposited on external female genital tract organs.
 
Objectives
To overcome above mentioned barriers in methylation detection, a novel technique called IMPRESS (Improved Methylation Profiling using Restriction Enzymes and smMIP Sequencing) has been developed. In this study, we aimed to adapt IMPRESS for cervical screening and triage of HPV positive women using cervical (pre)cancer specific methylation markers.
 
Methods
Previously, we identified hypermethylated biomarkers for cervical cancer detection based on online datasets and developed 1,536 single molecule Molecular Inversion Probes (smMIPs) for methylation detection with IMPRESS. Their performance was tested on 128 samples: 90 FVU, 20 VSS, and 18 clinician-collected cervical samples (CCS) from women referred for colposcopy after aberrant cervical results (NCT02714127, NCT03064087, and NCT03542513). 90 patients had no or low-grade cervical intraepithelial lesions (<HSIL) and 38 high-grade lesions (HSIL).  Undigested patient samples (N = 102) were used to determine the efficiency of the smMIPs. The optimal set of smMIPs was identified to achieve the highest Area Under the Receiver Operator Characteristic (ROC) Curve (AUC) and used to train a random forest classifier to correctly distinguish HSIL vs. <HSIL. For this, histology was used as reference outcome and if a biopsy was not indicated by the physician, colposcopy results were used instead.
 
Results
Based on the sequencing results, 779 inefficient smMIPs were discarded The highest (5-fold cross-validated) AUC of 0.81 was achieved with the combination of the top 18 methylation markers, giving a sensitivity of 75% and specificity of 78%. These markers were used as input for the random forest classifier with 4:1 training and test ratio. In repeated 5-fold cross-validation, the classifier achieved an AUC of 0.79 (95% CI: 0.68-0.89) in the training set and 0.72 (95% CI: 0.53-0.91) in the test set. Significant differences in predicted probabilities were observed for <HSIL vs. HSIL in CCS and FVU (p = 0.04 and 0.005 respectively), but not for VSS.
 
Conclusion
This study demonstrates the feasibility of using IMPRESS for the discrimination of <HSIL vs. HSIL in a referral setting using CCS as well as FVU samples. The developed classifier highlights the potential of methylation marker detection for screening and triage of HSIL. As such, these results could aid in introducing a full molecular cervical cancer screening approach applicable on FVU, warranting further training and validation in a larger cohort, which is ongoing.