Real time multimodal ctDNA and imaging assessment enables adaptive treatment in HPV related oropharyngeal cancer

Purpose: Circulating tumor DNA (ctDNA) has emerged as a sensitive biomarker for minimal residual disease detection, but its potential for real-time adaptation during curative therapy remains unexplored. Integrating ctDNA dynamics with imaging could enable more precise, risk-adapted treatment modifications beyond current radiological monitoring alone. HPV-associated oropharyngeal cancer (OPC) presents an ideal context given high cure rates, active treatment de-escalation efforts, and protracted chemoradiation courses allowing mid-treatment adaptation. We sought to determine whether longitudinal ctDNA profiling integrated with multimodal MRI could inform response-adapted treatment strategies. Methods: A total of 158 patients with HPV OPC were enrolled on a personalized de-escalation trial (NCT03323463). Blood samples for ctDNA analysis were collected pre-treatment and weekly during therapy, with 980 longitudinal samples obtained from 119 patients over up to 126 weeks (mean 8.2 samples/patient). We developed a dual Personalized Cancer Monitoring (PCM)-HPV ctDNA assay combining patient-specific tumor-informed variant detection (average 47 variants/patient) with probes targeting high-risk HPV subtypes, especially E6 and E7 genes of HPV-16 and HPV-18, both using anchored multiplex PCR (AMP) followed by high-throughput sequencing with Invitae (now Labcorp, San Francisco, CA). Weekly MRIs (T2-weighted) were performed concurrently to determine tumor volume. Results: Integrating somatic mutations with viral ctDNA detection increased baseline ctDNA detection from 80.3% (HPV ctDNA only) and 89.4% (PCM ctDNA only) to 93.9%. Baseline circulating HPV levels correlated with tumor volume, apoptotic signaling, necrosis, and primary tumor viral load (combined R²=0.54). During treatment, ctDNA showed broader dynamic range and faster kinetics than volumetric imaging. Critically, absolute ctDNA fraction at week 2 of treatment predicted patients requiring treatment intensification (p=0.0003). A multimodal model integrating on-treatment ctDNA assessment and imaging improved identification of patients with high-risk disease, surpassing either modality alone. Conclusions: Early on-treatment ctDNA dynamics capture aggressive disease biology during a critical window for adaptation. Multimodal integration of molecular and imaging assessments provides superior risk stratification, demonstrating that they capture complementary biology early in therapy and establishing a framework for real-time treatment personalization.