# Tissue of origin prediction for cancer of unknown primary using a targeted methylation sequencing panel

## TL;DR

This paper introduces a methylation-based classifier and sequencing panel to accurately predict the tissue of origin in cancer of unknown primary, improving diagnosis and treatment options.

## Contribution

A novel targeted methylation sequencing panel and classifier (MFCUP) for high-accuracy tissue-of-origin prediction in cancer of unknown primary.

## Key findings

- MFCUP achieved 97.2% accuracy in predicting tissue of origin across 25 cancer types.
- The targeted methylation sequencing panel correctly identified tissue of origin in 88.5% of FFPE samples.
- Specific CpG methylation levels can distinguish cancer types, suggesting potential as diagnostic biomarkers.

## Abstract

Cancer of unknown primary (CUP) is a group of rare malignancies with poor prognosis and unidentifiable tissue-of-origin. Distinct DNA methylation patterns in different tissues and cancer types enable the identification of the tissue of origin in CUP patients, which could help risk assessment and guide site-directed therapy.

Using genome-wide DNA methylation profile datasets from The Cancer Genome Atlas (TCGA) and machine learning methods, we developed a 200-CpG methylation feature classifier for CUP tissue of origin prediction (MFCUP). MFCUP was further validated with public-available methylation array data of 2977 specimens and targeted methylation sequencing of 78 Formalin‐fixed paraffin‐embedded (FFPE) samples from a single center.

MFCUP achieved an accuracy of 97.2% in a validation cohort (n = 5923) representing 25 cancer types. When applied to an Infinium 450 K array dataset (n = 1052) and an Infinium EPIC (850 K) array dataset (n = 1925), MFCUP achieved an overall accuracy of 93.4% and 84.8%, respectively. Based on MFCUP, we established a targeted bisulfite sequencing panel and validated it with FFPE sections from 78 patients of 20 cancer types. This methylation sequencing panel correctly identified tissue of origin in 88.5% (69/78) of samples. We also found that the methylation levels of specific CpGs can distinguish one cancer type from others, indicating their potential as biomarkers for cancer diagnosis and screening.

Our methylation-based cancer classifier and targeted methylation sequencing panel can predict tissue of origin in diverse cancer types with high accuracy.

The online version contains supplementary material available at 10.1186/s13148-024-01638-6.