Anastasia A Artyukh ^1* Mikhail K Ivanov ^1,2* Sergei E Titov ^1,2 Victoria V Dzyubenko ¹ Sergey E Krasilnikov ^3,4 Anastasia O Shumeikina ^3,4,5 Nikita A Afanasev ⁶ Anastasia V Malek ⁷ Sergei A Glushkov ¹ Eduard F Agletdinov ¹

• ¹ AO Vector-Best, Novosibirsk, Russia
• ² Institute of Molecular and Cellular Biology (RAS), Novosibirsk, Novosibirsk Oblast, Russia
• ³ Federal Research Center of Fundamental and Translational Medicine, Siberian Branch of the Russian Academy of Sciences, Novosibirsk, Novosibirsk Oblast, Russia
• ⁴ Novosibirsk State University, Novosibirsk, Novosibirsk Oblast, Russia
• ⁵ Meshalkin National Medical Research Center, Novosibirsk, Novosibirsk Oblast, Russia
• ⁶ Saint-Petersburg City Clinic №17, Saint-Petersburg, Russia
• ⁷ N.N. Petrov National Medical Research Center of Oncology, Saint Petersburg, Russia

Cervical screening, aimed at detecting precancerous lesions and preventing cancer, is based on cytology and high-risk HPV testing. Both methods have their limitations, the main ones being the variable diagnostic sensitivity of cytology and the moderate specificity of HPV testing. Various molecular biomarkers are proposed in recent years to improve cervical cancer management, including a number of mRNAs encoded by human genes involved in cervical carcinogenesis. Many scientific papers have shown that the expression patterns of cellular mRNAs reflect the severity of the lesion, and their analysis in cervical smears may outperform HPV testing in terms of diagnostic specificity. However, such analysis has not yet been implemented in broad clinical practice. Our aim was to devise an assay detecting severe cervical lesions (≥HSIL) via analysis of cellular mRNA expression in cytological smears. Through logistic regression analysis of a reverse-transcription quantitative PCR (RT-qPCR) dataset generated from analysis of six mRNAs in 167 cervical smears with various cytological diagnoses, we generated a family of linear classifiers based on paired mRNA concentration ratios. Each classifier outputs a dimensionless decision function (DF) value that increases with lesion severity. Additionally, in the same specimens, the HPV genotyping, viral load assessment, diagnosis of cervicovaginal microbiome imbalance and profiling of some relevant mRNAs and miRNAs were performed by qPCR-based methods. The best classifiers were obtained with pairs of mRNAs whose expression changes in opposite directions during lesion progression. With this approach based on a five-mRNA combination (CDKN2A, MAL, TMPRSS4, CRNN, and ECM1), we generated a classifier having ROC AUC 0.935, diagnostic sensitivity 89.7%, and specificity 87.6% for ≥HSIL detection. Based on this classifier, a two-tube RT-qPCR based assay was developed and it confirmed the preliminary characteristics on 120 cervical smears from the test sample. DF values weakly correlated with HPV loads and cervicovaginal microbiome imbalance, thus being independent markers of ≥HSIL risk. Thus, we propose a high-throughput method for detecting ≥HSIL cervical lesions by RT-qPCR analysis of several cellular mRNAs. The method is suitable for the analysis of cervical cytological smears prepared by a routine method. Further clinical validation is necessary to clarify its clinical potential.