TOPLINE:

A novel multifeature human papillomavirus (HPV) whole genome sequencing liquid biopsy assay detected circulating tumor HPV DNA (ctHPV DNA) with 79% sensitivity and 100% specificity up to 7.8 years before the clinical diagnosis of oropharyngeal cancer. Application of a machine learning model improved detection sensitivity to 96% and extended the maximum lead time to 10.3 years.

METHODOLOGY:

• Despite decades-long preclinical development, HPV-associated oropharyngeal squamous cell carcinoma lacks screening tests. Although ctHPV DNA is a promising biomarker, existing droplet digital polymerase chain reaction (ddPCR) methods lack sensitivity for detecting ctHPV DNA in prediagnostic samples. To address this gap, the present study developed and analyzed a novel HPV whole genome sequencing liquid biopsy assay: HPV-DeepSeek.
• Researchers analyzed plasma samples from 28 patients with HPV-associated oropharyngeal squamous cell carcinoma collected 1.3-10.8 years before diagnosis and from 28 age-, sex-, and plasma volume-matched control individuals from the Mass General Brigham biobank.
• Samples were processed using the HPV-DeepSeek hybrid capture assay that targeted 43 HPV genotypes and analyzed for viral reads, genome coverage, integration events, and fragmentomics, among other features. A predefined cutoff of ≥ 10 unique HPV reads and ≥ 10% genome coverage indicated a positive result; ddPCR was performed in parallel with HPV antibody assays.
• A machine learning model was trained and tested on an independent cohort of 153 patients and 153 control individuals to improve detection sensitivity.

TAKEAWAY:

• ctDNA was detected in 22 of 28 prediagnostic samples (overall sensitivity, 79%) — with a maximum lead time of 7.8 years — and all control individuals tested negative for ctHPV DNA, maintaining specificity at 100%.
• In prediagnostic samples, HPV-DeepSeek outperformed ddPCR and HPV serology, doubling ddPCR sensitivity (36% vs 79%) and surpassing antibody detection sensitivity (77% vs 79%), particularly within 4 years of diagnosis compared with HPV serology (P =.004), underscoring the superior early detection accuracy of the sequencing assay.
• Application of a machine learning model (Naive Bayes model) increased the detection sensitivity to 96% and extended the maximum lead time for ctHPV DNA detection to 10.3 years.
• HPV genome coverage emerged as the most influential predictor across all models, while fragmentomics features — mean fragment length, fragment skewness, and fragment kurtosis — further enhanced the classification accuracy. Viral genome molecular fingerprinting confirmed that 18 of 22 cases contained unique viral genomes, with matching patterns observed between prediagnostic plasma and diagnostic tumor tissue.

IN PRACTICE:

"Using plasma samples from asymptomatic individuals who later developed HPV+OPSCC [oropharyngeal squamous cell carcinoma] and population-level controls, paired with a novel ultrasensitive multifeature HPV [whole genome sequencing] liquid biopsy," the authors demonstrated a "sensitive and specific blood-based early detection of HPV+OPSCC extending to a decade prior to clinical cancer diagnosis."

SOURCE:

The study, led by Dipon Das, PhD, Harvard Medical School in Boston, was published online in the Journal of the National Cancer Institute.

LIMITATIONS:

The primary limitation of this study was the small sample size, thereby limiting its statistical power and generalizability.

DISCLOSURES:

The study was supported by NIH/NIDCR. One author disclosed receiving research funding from Bristol-Myers Squibb, Calico, Predicine, BostonGene, NeoGenomics, and Haystack (Quest) and consulting fees from Merck, Noetic, Chrysalis Biomedical Advisors, NeoGenomics, Arcadia, GT Molecular, and Focus. Another author declared serving on advisory boards for Merck Sharp & Dohme. Additional disclosures are noted in the original article.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.