98P - Translating cancer tissue methylation to cell-free DNA methylation for minimally invasive cancer detection

Background

Cell-free DNA (cfDNA) consists of circulating DNA fragments originating from cells throughout the body. It can be found in plasma and urine and is under heavy investigation for its potential to detect cancer both early, and as a minimally invasive biopsy. DNA methylation is strongly associated with tissue and cell type, and in the context of cancer, it is often significantly altered. Tumour derived cfDNA has been shown to contain tumour specific methylation patterns. In this study, we attempt to correlate tissue specific methylation with cfDNA methylation.

Methods

We extracted tumour tissue specific methylation signals from The Cancer Genome Association (TCGA) methylation data and used our RenovaroCube (AI platform) to mine tumour specific CpG sites from Illumina methylation array data. As Illumina methylation array data for liquid biopsies is scarce due to the high input requirements, we investigated an alternative methylation detection method: cell-free Methylation DNA ImmunoPrecipiation (cfMeDIP-seq). cfMeDIP-seq captures solely 5mC methylated cfDNA fragments, which are then sorted into 300bp windows. From 97 publicly available cfDNA plasma cfMeDIP samples (24 control, 23 CRC, 25 lung cancer, 25 BRCA) we extracted the 300bp windows containing mined TCGA CpG sites. Using random forest classifiers, we performed leave-one-out-cross-validations (LOOCV) for the classifcation of cancer types. We further validated the lung cancer model using a cfMeDIP-seq validation set (62 control, 55 lung cancer).

Results

From the TCGA data, we mined 86 NSCLC, 95 BRCA and 139 CRC specific CpG sites, resulting in 72, 74 and 121 300bp panels, respectively. The performance of the panels was evaluated through LOOCV, resulting in (Specificity/Sensitivity/Accuracy): NSCLC model: 0.96/0.59/0.75, BRCA model: 1.00/0.72/0.86, CRC model: 1.00/0.841/0.92. In the validation lung cancer dataset, we obtained a performance of 0.97/0.49/0.74.

Conclusions

It is possible to detect tissue specific methylation levels extracted from methylation array data using RenovaroCube platform in cfDNA using cfMeDIP-seq data.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The author.

Funding

Has not received any funding.

Disclosure

The author has declared no conflicts of interest.