Single-cell multi-omics profiling enables high-resolution, high-dimensional understanding of heterogeneous biological processes and identifies rare cell types. However, identification of transcriptome signatures and new cell types are frequently not biologically informative. Advanced bioinformatic analysis and machine learning models are required to reveal biological insights that can stand wet-bench validations. Using HIV infection as a model, we build bioinformatic analysis and machine learning models to answer biological questions, reveal mechanisms of viral immune escape, and provide insights for treatments.
Despite effective antiretroviral therapy, HIV persists as integrated DNA provirus in CD4+ T cells. HIV-infected cells undergo clonal expansion and proliferate over time. The clonally expanding HIV-infected CD4+ T cells survive viral cytopathic effect, resist immune clearance, and are the major barrier to cure. We used single-cell ECCITE-seq, which captures surface protein expression, cellular transcriptome, HIV RNA, and TCR sequences within the same single cell to track clonal expansion dynamics in longitudinally archived samples from six HIV-infected individuals (during viremia and after suppressive antiretroviral therapy) and two uninfected individuals, in unstimulated conditions and after CMV and HIV antigen stimulation. Despite antiretroviral therapy, persistent antigen and TNF responses shaped T cell clonal expansion. HIV resided in Th1 polarized, antigen-responding T cells expressing BCL2 and SERPINB9 that may resist cell death. HIV RNA+ T cell clones were larger in clone size, established during viremia, persistent after viral suppression, and enriched in GZMB+ cytotoxic effector memory Th1 cells. Targeting HIV-infected cytotoxic CD4+ T cells and drivers of clonal expansion provides a new direction for HIV cure.