Integrated longitudinal analysis of ctDNA, radiologic response, and tumor volume reveals spatial and temporal heterogeneity in advanced melanoma

Abstract Background In advanced melanoma, longitudinal disease monitoring remains limited by infrequent biomarker assessment and predominantly categorical imaging readouts. Although circulating tumor DNA (ctDNA) correlates with tumor burden, its behavior under high-frequency sampling and its relationship to quantitative metastatic tumor volume are poorly defined. Clarifying these dynamics is essential to establishing ctDNA as a clinically actionable tool for real-time treatment monitoring. Methods We retrospectively analyzed 42 patients with unresectable stage III/IV melanoma treated with immune checkpoint inhibitors. Plasma ctDNA was quantified longitudinally across 241 time points using a UMI-based amplicon next-generation sequencing (NGS) assay detecting BRAF, EGFR, KRAS, NRAS, and PIK3CA. We paired ctDNA measurements with radiologic staging and correlated them with response categories ( n = 99 time points) and volumetric tumor burden ( n = 73 time points) using nonparametric statistics and receiver operating characteristic (ROC) analyses, including stratification by the ctDNA-imaging time interval and descriptive assessment of discordant patterns. Results ctDNA levels were significantly associated with radiologic response according to RECIST 1.1, increasing across worsening response categories (Spearman ρ = 0.31, 95% CI 0.20–0.52, p = 0.002). Discriminative performance for identifying progression was moderate (AUC = 0.69). Threshold analyses demonstrated a sensitivity–specificity trade-off, with higher ctDNA levels providing greater specificity (e.g. ≥25 mutant molecules (MM)/mL: specificity ≥ 83%, sensitivity ≤ 30%), while lower thresholds showed limited sensitivity and specificity (e.g. 1–2 MM/mL: ~56% each). ctDNA concentrations were positively associated with total tumor volume (ρ = 0.46, p < 0.0001), with stronger correlations observed for temporally aligned measurements (0–30 days: ρ = 0.56). In contrast, dynamic changes in ctDNA were not significantly correlated with changes in tumor volume (ρ = 0.20, p = 0.25). Organ-associated analyses demonstrated marked heterogeneity in ctDNA shedding, with stronger associations in lymph node and peritoneal metastases and limited detectability in lung, liver, and brain metastases. Concordance between ctDNA dynamics and radiologic response was high in responding disease (CR/PR: 85.0%) but lower in stable or progressive disease (72.9%), with discordant cases associated with tumor heterogeneity, timing differences, and treatment-related factors. Conclusions ctDNA reflects radiologic disease status and tumor burden, with increasing levels observed as response worsens. Despite heterogeneity in ctDNA shedding across metastatic sites, its significant correlation with tumor volume, particularly when closely timed to imaging, supports the potential value of ctDNA as a complementary biomarker for longitudinal disease monitoring while highlighting important biological and methodological factors that influence its clinical interpretation.