Microbiota cervicovaginal como biomarcador preditivo de câncer cervical: meta-análise integrativa com aprendizado de máquina

Abstract

Cervical cancer (CC) is one of the leading causes of female mortality in developing countries. Although persistent infection with human papillomavirus (HPV) is the main etiological factor, growing evidence indicates that alterations in the cervicovaginal microbiota also play an important role in disease progression. This study aimed to investigate variations in microbial composition and diversity across different stages of cervical carcinogenesis, seeking to identify potential bacterial biomarkers for early diagnosis and risk prediction. A total of 1,008 cervical and vaginal samples obtained from public databases (ENA and NCBI) were analyzed, distributed across four clinical groups: healthy controls (n = 441), low-grade squamous intraepithelial lesions (LSIL, n = 112), high-grade lesions (HSIL, n = 170), and cervical cancer (n = 285), covering populations from five continents. 16S rRNA gene sequences were processed using QIIME 2 and analyzed for alpha and beta diversity, taxonomic composition, and discriminative power through machine learning algorithms. Alpha diversity analysis, based on the Shannon index, showed higher diversity in the cancer group. Beta diversity, evaluated using NMDS, indicated marked compositional differences between groups, especially in the cancer group, which exhibited greater microbial dispersion. Taxonomically, healthy controls showed predominance of Firmicutes and the Lactobacillaceae family, while lesion and cancer groups exhibited a higher relative abundance of Bacteroidota, Fusobacteriota, and Proteobacteria, with emphasis on the families Peptostreptococcaceae, Prevotellaceae, Fusobacteriaceae, and Enterobacteriaceae. Additionally, the relationship between microbiota and HPV infection status was evaluated. The abundance of Gardnerella did not differ significantly between HPV-positive and HPV-negative women (p = 0.385). However, Faecalibacterium was more prevalent in HPV-negative women, while the Burkholderia-Caballeronia-Paraburkholderia group was more abundant in HPV-positive women, suggesting their potential as biomarkers associated with the presence or absence of the virus. Discriminant analysis (LDA) identified characteristic microorganisms for each group: Lactobacillus in the control group; Gardnerella and Streptococcus in LSIL; Sneathia and L. iners in HSIL; and Fusobacterium, Staphylococcus, Peptostreptococcus, and Enterococcus in the cancer group. Machine learning model (Random Forest) achieved high accuracy in distinguishing cancer from controls (AUC = 0.96 and 0.86), but lower performance in the intermediate categories (LSIL and HSIL), indicating the need for more robust models. The identification and validation of microbial biomarkers may enable non invasive strategies for screening and early diagnosis of cervical cancer, contributing to better prognosis and increased patient survival.