Background

Minimal residual disease (MRD) detection after curative-intent surgery determines whether patients receive adjuvant therapy. Current ctDNA-based approaches (e.g., Signatera, FoundationOne Tracker) use tumor-informed mutation panels with sensitivity of 0.01-0.1% VAF. This misses patients who relapse despite negative ctDNA — the false negative rate is 30-50% in stage II-III colorectal and NSCLC.

Hypothesis

Integrating ctDNA mutation tracking with cell-free methylation signatures (cfMeDIP-seq or targeted bisulfite panels), circulating tumor cell (CTC) enumeration with single-cell transcriptomics, and exosomal protein/miRNA panels into a unified Bayesian classifier will achieve MRD sensitivity of 0.001% (10-fold improvement over ctDNA alone) with specificity >95%, reducing the false negative rate from ~40% to <10% in stage II-III solid tumors.

Rationale

1. ctDNA and methylation are complementary. ctDNA tracks somatic mutations (clonal evolution) while methylation captures tissue-of-origin and differentiation state. Tumors that shed minimal ctDNA (low tumor burden, poor vascularization) often still release methylation-altered cfDNA because methylation changes are clonal and stable.

2. Exosomal cargo fills the gap for non-shedding tumors. Some solid tumors (particularly low-grade, well-encapsulated, or brain tumors behind the BBB) release minimal cfDNA but actively secrete exosomes. Exosomal miRNA signatures (miR-21, miR-155, miR-210) and surface proteins (EpCAM, CD63) provide an orthogonal detection channel.

3. CTC single-cell transcriptomics resolves heterogeneity. Bulk ctDNA averages across subclones. CTCs captured and profiled at single-cell resolution reveal therapy-resistant subpopulations (EMT-high, stemness-high) that predict relapse pattern — information invisible to bulk liquid biopsy.

4. The Bayesian integration is critical. Each analyte alone has different sensitivity/specificity tradeoffs across tumor types. A hierarchical Bayesian model that weights each channel by tumor type, stage, and prior treatment can dynamically optimize the detection threshold.

Testable predictions

• In a prospective cohort of 500 stage II-III CRC patients post-surgery, the multi-omic panel will detect MRD in >90% of patients who relapse within 24 months (vs ~60% for ctDNA alone)
• The integrated panel will identify the dominant resistant subclone at MRD detection in >70% of cases, enabling rational adjuvant therapy selection
• Time-to-MRD-detection will precede radiographic recurrence by median 6+ months (vs 3-4 months for ctDNA alone)

Limitations

• Cost: multi-omic panels will initially cost $3,000-5,000 per timepoint vs $500-1,000 for ctDNA alone. Health-economic modeling needs to show that avoiding unnecessary adjuvant chemo (in true negatives) and earlier intervention (in true positives) justifies the cost.
• Standardization: each analyte requires different pre-analytical handling (cfDNA extraction, exosome isolation, CTC enrichment). Sample splitting and processing complexity increase failure rates.
• The Bayesian classifier requires large training cohorts per tumor type. Rare cancers will be underpowered.

Why this matters

The current standard of care — treat-all or ctDNA-guided — either overtreats 60% of patients (stage II CRC) or misses 30-40% of relapses. A 10x sensitivity improvement with maintained specificity would fundamentally change adjuvant therapy decisions for millions of cancer patients annually.