Analysis of precancerous lesion-related microRNAs for early diagnosis of cervical cancer in the Thai population

Abstract

The incidence rate of cervical cancer (CC) is three times greater in Southeast Asia (SEA), where screening tests are less common than in Northern America, underlining a need for convenient self-diagnostic methods. The expression pattern of microRNAs (miRNAs) has been considered a molecular tool for non-invasive cancer diagnosis and prognosis. This study aimed at the development of the first miRNA biomarker panel for early detection of CC in Thai women. Genome-wide miRNA expression profiling was performed on cervical tissue and discharge samples from high-grade squamous intraepithelial lesion (HSIL) and adenocarcinoma in situ (AIS) subjects. Machine learning was used for handling imbalanced data and feature selection before differential expression analysis to identify significantly dysregulated miRNA panels. Pathway analysis was conducted to provide the cellular functions involved in CC progression. The study identified a shared 18-miRNA panel for both tissue and discharge, with which the prediction model distinguished HSIL and AIS from normal samples with an accuracy of 90.9%. Three dysregulated miRNAs comprised of miR-125b-1-3p, miR-487b-3p, and miR-1180-3p in CC were first described. Most of the miRNAs in the panel were down-regulated, whereas merely miR-142-3p was significantly up-regulated in HSIL and AIS, suggesting a convenient biomarker for detecting precancerous conditions. Moreover, our miRNA panel highlighted important roles played by the cell-cell interaction pathways in CC. Together, our miRNA panel hold promise as a biomarker for the early detection of cervical cancer with cervical discharge, offering the possibility for developing non-invasive diagnostic tools.

Keywords: Adenocarcinoma in situ; Cervical cancer; High-grade squamous intraepithelial lesion; Machine learning; microRNA.

Fig. 1

The overall workflow of the miRNA biomarker analysis in this study.

Fig. 2

NOISeq analysis of important miRNAs. (A) The heatmap clustering of 223 important miRNAs after filtering out low counts and normalized with TMM. The total numbers of samples (tissue + discharge) for each group are indicated in parentheses in the graph legend. (B) The Principle component analysis (PCA) plot of important miRNAs expression from discharge samples (DC_HSIL and DC_ AIS), tissue samples (HSIL and AIS), and normal tissue samples (Normal). The total numbers of samples for each group are indicated in parentheses in the graph legend. The differential expression plot of miRNAs in HSIL (C) and AIS (D). Up-regulated miRNAs within the probability of differential expression threshold (q) = 0.9, 0.8, and 0.7 are labeled and colored in bright red, dark red, and pink, respectively. The downregulated miRNAs are in bright blue, dark blue, and light blue, respectively.

Fig. 3

Analysis of the 18-miRNA panel. (A) The Venn diagram represents dysregulated miRNAs selected from machine learning. Up-regulated miRNAs are in red characters and the down-regulated miRNAs are in blue. (B) The PCA plot represents sample classification of HSIL and AIS using the 18-miRNA panel. The total numbers of samples for each group are indicated in parentheses in the graph legend.

Fig. 4

Heatmap of KEGG pathway enrichment analysis of the miRNAs panel for PC. The labels of up-regulated miRNAs are underlined. The names of the pathways related to cell-cell interactions are highlighted in red.