Therapy resistance remains one of the most formidable challenges in oncology. While pharmacogenetic studies link patient outcomes to genetic variation, they overlook cell-level molecular signatures, such as transcriptional plasticity and epigenetic modification, that capture dynamic regulatory mechanisms beyond genetics alone.
To address this, we recruited a large-scale pan-cancer liquid biopsy (LBIO) cohort of 455 patients with matched whole-genome sequencing (WGS), single-cell RNA (scRNA-seq), and ATAC (scATAC-seq) profiles from peripheral blood mononuclear cells. Patients received chemotherapy, immunotherapy, targeted therapy, or combinations thereof, with clinical outcomes systematically recorded. To maximize the discovery of candidate signatures, we leveraged the largest-to-date single-cell expression (sc-eQTL) and chromatin accessibility (sc-caQTL) QTL mapping results from the TenK10K Phase1 study, comprising 154,932 sc-eQTLs and 243,273 sc-caQTLs from ~2,000 donors. Integrating these population-scale molecular QTLs for discovery with the LBIO cohort for validation enabled the identification of robust multi-omic markers underlying the heterogeneity in therapeutic outcome.
We mapped the QTLs to a curated list of pharmacogenes (key regulators of drug metabolism and transport, and potential drug targets) from ClinPGx. We identified epigenetic markers by colocalising sc-eQTLs and sc-caQTLs. This yielded a systematic catalogue of risk loci and chromatin peaks across 112 genes critical to cancer treatment efficacy and safety. Our findings recapitulate known pharmacogenetic associations while revealing 86 novel genetic and 193 epigenetic signatures linked to pharmacogenes. For instance, we replicated the well-established MTHFR 677C>T association with MTHFR expression in CD4 naive cells (p = 4.5 × 10⁻¹⁶, β = –0.046) and identified previously unreported markers, such as rs55763963(T>C) and chr16:28609104-28611222 associated with SULT1A1, offering new candidate signatures into the molecular determinants of therapeutic efficacy.
Overall, our study provides a valuable resource of genetic and epigenetic markers for therapy resistance, establishing a foundation for developing large-scale, multi-omics AI models that enable personalized prediction of ineffective cancer interventions.