Oncology · Medical Literature
Genome-wide variation in cell-free DNA end-motif entropy predicts immunotherapy response in head and neck cancer.
Abstract
BACKGROUNDMinimally invasive biomarkers predicting the immunotherapy response in head and neck squamous cell carcinoma (HNSCC) remain an unmet clinical need.METHODSIn a prospective, multi-institutional phase II trial, we performed whole-genome sequencing of 185 longitudinal plasma cell-free DNA (cfDNA) samples from 68 patients with locally advanced, surgically resectable HNSCC who received neoadjuvant and adjuvant pembrolizumab. We developed the regional motif diversity score (rMDS), a fragmentomic metric that quantifies the entropy of cfDNA 5'-end motifs across genomic regions.RESULTSUnsupervised analysis showed that rMDS robustly distinguished responders from nonresponders, outperforming established fragmentomic metrics and copy number alterations while remaining independent of technical confounders. Longitudinal rMDS changes localized to regions enriched for immune-, lectin-, and keratinization-related genes - hallmarks of squamous cell carcinoma - reflecting tumor-peripheral immunity interplay during treatment. The most dynamic regions clustered at telomere-proximal loci, suggesting a link between telomere biology and cfDNA fragmentation. An rMDS-based machine learning classifier achieved AUC 0.89-0.99 across validation settings, with the highest accuracy after treatment, outperforming PD-L1 expression and tumor fraction in matched samples. Predicted responders showed improved disease-free survival (log-rank P = 0.035; HR 2.67, 95% CI 1.03-6.92).CONCLUSIONrMDS represents a biologically meaningful and clinically actionable biomarker for the immunotherapy response in HNSCC, and merits integration into future risk assessment frameworks.TRIAL REGISTRATIONClinicalTrials.gov NCT02641093.FUNDINGNational Human Genome Research Institute (NHGRI), NIH grant R56HG012360; startup funds from Cincinnati Children's Hospital Medical Center, Northwestern University, and Robert H. Lurie Comprehensive Cancer Center; Science Olympiad Alumni Research Grant, Science Olympiad USA Foundation; Merck Sharp & Dohme Corp.
Clinical relevance
In patients with locally advanced, surgically resectable head and neck squamous cell carcinoma (HNSCC) receiving neoadjuvant and adjuvant pembrolizumab, can a novel cfDNA fragmentomic metric—the regional motif diversity score (rMDS)—predict immunotherapy response more accurately than established biomarkers such as PD-L1 expression and tumor fraction? __SNAPSHOT__ {"studyDesign":"Prospective, multi-institutional phase II clinical trial with embedded translational biomarker study (NCT02641093); whole-genome sequencing of longitudinal plasma cfDNA samples with machine learning classifier development and validation","setting":"Multi-institutional (specific institutions not individually enumerated in abstract); United States (inferred from authorship affiliations including Northwestern University, Cincinnati Children's Hospital Medical Center, and Robert H. Lurie Comprehensive Cancer Center)","sampleSize":"n=68 patients; 185 longitudinal plasma cfDNA samples analyzed","population":"Adults with locally advanced, surgically resectable HNSCC receiving neoadjuvant and adjuvant pembrolizumab; additional eligibility criteria (e.g., HPV status, PD-L1 CPS thresholds, prior therapy) not reported in abstract","intervention":"Development and validation of the regional motif diversity score (rMDS), a fragmentomic metric quantifying Shannon entropy of cfDNA 5′-end motifs across genomic regions, derived from whole-genome sequencing of serial plasma samples","comparator":"Head-to-head comparison against established fragmentomic metrics, copy number alterations, PD-L1 expression, and tumor fraction in matched samples","primaryOutcome":"Discrimination of immunotherapy responders from non-responders by rMDS; classifier performance reported as AUC across multiple validation settings; association with disease-free survival (DFS)","followUp":"Longitudinal (serial sampling across neoadjuvant and adjuvant treatment phases); exact duration of follow-up for DFS not reported in abstract","fundingSource":"Mixed: Government (NIH/NHGRI grant R56HG012360); Non-profit/institutional startup funds (Cincinnati Children's Hospital Medical Center, Northwestern University, Robert H. Lurie Comprehensive Cancer Center; Science Olympiad Alumni Research Grant, Science Olympiad USA Foundation); Industry (Merck Sharp & Dohme Corp. — manufacturer of pembrolizumab)"}
Key findings
The rMDS machine learning classifier achieved AUC 0.89–0.99 across multiple validation settings, with the highest discriminative accuracy observed at post-treatment timepoints, outperforming PD-L1 expression and tumor fraction in matched samples. Patients classified as predicted responders by rMDS demonstrated significantly improved disease-free survival compared to predicted non-responders (log-rank P = 0.035; HR 2.67, 95% CI 1.03–6.92). The lower bound of the CI barely crosses 1.0, warranting cautious interpretation of clinical effect size despite statistical significance. No absolute DFS rates (e.g., 2-year DFS percentages) or median survival times are reported in the abstract, precluding NNT calculation. In unsupervised analyses, rMDS robustly separated responders from non-responders and remained independent of technical confounders, outperforming established fragmentomic metrics and copy number alterations. Longitudinal rMDS changes localized to genomic regions enriched for immune-, lectin-, and keratinization-related genes, and the most dynamic regions clustered at telomere-proximal loci. Adverse event data are not reported in the abstract.
Practice implications
This study establishes proof-of-concept that a novel cfDNA fragmentomic metric (rMDS), derived from whole-genome sequencing of serial plasma samples, can distinguish immunotherapy responders from non-responders in locally advanced HNSCC with high discriminative performance (AUC up to 0.99 post-treatment) and is associated with improved DFS. Importantly, rMDS outperformed currently used clinical biomarkers—PD-L1 expression, tumor fraction, and copy number alterations—in matched samples, addressing a well-recognized limitation of existing predictors. However, several critical uncertainties limit immediate clinical translation: (1) the sample size is small (n=68), raising concern about overfitting of the machine learning classifier despite multi-setting validation; (2) the 95% CI for the DFS hazard ratio (1.03–6.92) is very wide, reflecting imprecision; (3) HPV status, PD-L1 CPS, disease stage distribution, and response definition criteria are not described in the abstract, limiting assessment of population generalizability; (4) rMDS is not yet analytically validated or standardized for clinical laboratory implementation; (5) the highest classifier accuracy was post-treatment, which limits its pre-treatment prognostic utility; (6) Merck (pembrolizumab manufacturer) provided partial funding, introducing potential for industry bias. Only the abstract was available for this review, which precludes evaluation of subgroup analyses, model architecture, cross-validation strategy, and definitions of 'responder.' Findings are not applicable to recurrent/metastatic HNSCC, non-squamous head and neck cancers, or settings outside the neoadjuvant/adjuvant pembrolizumab context. __LIMITATIONS__ 1. Small sample size (n=68) with a large number of longitudinal samples (185), creating risk of overfitting in the machine learning classifier despite reported multi-setting validation. 2. Wide confidence interval for the primary survival outcome (HR 2.67, 95% CI 1.03–6.92) indicates imprecision and the result is marginally statistically significant (P=0.035). 3. Abstract does not report response definitions, HPV status breakdown, staging details, or PD-L1 CPS distribution, limiting generalizability assessment. 4. Partial industry funding (Merck Sharp & Dohme Corp.) introduces potential sponsorship bias. 5. Highest classifier accuracy is observed post-treatment, limiting utility as a pre-treatment selection tool. 6. rMDS requires whole-genome sequencing of plasma cfDNA, which is not yet standardized or broadly available in clinical laboratories, limiting near-term implementation. 7. The study is restricted to locally advanced, surgically resectable HNSCC on a specific immunotherapy regimen; findings cannot be extrapolated to other HNSCC populations or other checkpoint inhibitors without further validation. 8. Duration of DFS follow-up is not reported in the abstract. 9. Only the abstract was available for full methodological appraisal, including cross-validation strategy, feature selection, and model transparency. __BOTTOM_LINE_CATEGORY__ promising_but_preliminary __BOTTOM_LINE__ rMDS, a novel cfDNA 5′-end motif entropy metric derived from whole-genome sequencing, demonstrated strong discriminative performance (AUC 0.89–0.99) for immunotherapy response in locally advanced HNSCC and was associated with improved disease-free survival in this small prospective multi-institutional cohort. These findings represent a meaningful advance over current liquid biopsy and tissue-based biomarkers, but the small sample size, wide survival confidence intervals, partial industry funding, and requirement for whole-genome sequencing infrastructure necessitate independent validation in larger cohorts before clinical adoption. Clinicians should regard rMDS as a highly promising but not yet practice-ready biomarker pending prospective validation and analytical standardization.