Single-cell transcriptomics links HPV transcript abundance to malignant epithelial heterogeneity and redistribution of cell-cell communication across cervical lesions
Abstract
Persistent high-risk Human Papillomavirus (HPV) infection is a prerequisite for most cervical cancers, but only a subset of infections progress to invasive disease. This study investigated associations between HPV transcript abundance, malignant epithelial heterogeneity, inferred cell-state trajectories, and microenvironmental organization at single-cell resolution. Single-cell RNA sequencing data from GSE208653 and a supplementary cervical squamous cell carcinoma dataset were mapped to a combined human-HPV reference. The analytical framework integrated inferCNV-based copy-number analysis, GeneNMF metaprogram analysis, pseudotime inference, CellChat, network-based in silico knockout, survival analysis, and assessment in an independent single-cell dataset. After quality control, 105,799 cells were retained, and transcripts from HPV16, HPV18, and HPV59 were detected. Putative malignant epithelial cells separated into HPV high /MP6-high and HPV dim /MP6-low states. Among the 10 metaprograms, MP6 showed the strongest association with HPV transcript abundance and shared 56 genes with the intermediate-peaking C2 dynamic module (adjusted P = 9.58 × 10⁻²²). The two states had similar aggregate interaction numbers and strengths but differed in the distribution of inferred pathways. Compared with HPV-positive normal cervical samples, high-grade squamous intraepithelial lesion samples showed fewer and weaker inferred interactions, together with a shift from immune-centered networks toward epithelial and epithelial-stromal communication. In silico perturbation prioritized COL6A1, ABCC3, STARD10, and AOC1, with stronger external support for the first three than for AOC1. Together, these analyses support associations between HPV transcript abundance, malignant epithelial states, and microenvironmental organization but do not establish causality. Experimental validation and larger multi-patient cohorts are required.