A cfDNA methylation-based tissue-of-origin classifier for cancers of unknown primary

Area of Science:

• Oncology
• Genomics
• Bioinformatics

Background:

• Cancers of Unknown Primary (CUP) present significant diagnostic and therapeutic challenges due to tumor heterogeneity and limited treatment efficacy.
• Molecular prediction of tissue-of-origin (TOO) can refine CUP diagnosis, and liquid biopsies offer a less invasive approach compared to tissue acquisition.
• Previous TOO prediction methods using liquid biopsies have not been explored in CUP cohorts.

Purpose of the Study:

• To develop and validate a machine learning classifier (CUPiD) for predicting the tissue-of-origin (TOO) in Cancers of Unknown Primary (CUP) using cell-free DNA (cfDNA) methylation.
• To assess the accuracy and clinical utility of CUPiD in diverse cancer types and a CUP cohort.

Main Methods:

• Development of CUPiD, a machine learning classifier utilizing cfDNA methylation patterns for TOO prediction across 29 tumor classes.
• Validation of CUPiD on 143 cfDNA samples from patients with 13 cancer types and 27 non-cancer controls.
• Testing CUPiD on an independent cohort of 41 patients diagnosed with CUP.

Main Results:

• CUPiD achieved an overall sensitivity of 84.6% and a TOO accuracy of 96.8% in the validation cohort.
• In the CUP cohort, CUPiD made predictions in 78.0% of cases (32/41).
• 88.5% of CUPiD predictions were clinically consistent with subsequent or suspected primary tumor diagnoses (23/26 patients).

Conclusions:

• CUPiD demonstrates high accuracy in predicting tissue-of-origin from cfDNA methylation, offering a promising tool for CUP diagnosis.
• The integration of CUPiD with cfDNA mutation data shows potential for re-classifying diagnoses and guiding treatment strategies in difficult-to-treat CUP cases.
• This approach mitigates tissue acquisition barriers and advances the molecular profiling of CUP through non-invasive liquid biopsies.