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

Introduction

Cancer of unknown primary (CUP) is a rare disease in which malignant cells are present while the originating organ remains unknown. This makes it difficult to treat, as selecting an appropriate therapy is problematic without knowing the tumour type. Predicting tissue-of-origin (TOO) is complicated by the absence of biopsies on which to perform molecular characterisation. One solution is liquid biopsies; however, these have yet to be undertaken in CUP. This study describes CUPiD, a machine learning tool which predicts TOO by assessing the methylation patterns of circulating cell-free DNA (cfDNA).

Main Points

• Using cfDNA to classify TOO is challenging due to the high proportion of non-cancerous DNA present. To mitigate this, the authors created in silico mixtures of non-cancer and cancer DNA methylation patterns to mimic differing tumour fractions (TF). Identifying differentially methylated regions (DMRs) within these genomes, they were then able to produce a prediction matrix appliable to a wide range of cancer types, which they named CuPID.
• CUPiD was applied to a retrospective cohort of 143 patients with 13 cancer types, where it correctly classified 84.6% of tumours. This was possible despite the variable estimated tumour fraction in the cfDNA, suggesting strong utilities in CUP cohorts.
• The authors used CUPiD to assess cfDNA from a prospective cohort of 41 CUP patients. cfDNA methylation profiling was undertaken on material isolated from plasma and processed using the T7-MBD-seq method. Samples were in vitro transcribed to RNA, sequencing adapters ligated, reverse transcribed and Illumina index adapters appended. Libraries were then paired-end sequenced using Illumina platforms.
• This method included a complex bioinformatics pipeline, with Nextflow DSKL2 being used for T7-MBD-seq alignment, QSA to perform Methylation Enrichment Analysis, IchorCNA to estimate tumour fraction and NGSCheckMate to perform sequencing QC.
• cfDNA methylation profiling was successful for all 41 patients in the cohort, and when CUPiD was applied, it yielded a tumour prediction in 78% (32/41) of cases. Of these 32 tumours, the most common types were hepato-pancreatobiliary (7/32), female genital tract (6/32) and lung (5/32).
• Of the 33 patients from the cohort with a clinically resolved tumour type, 26 had a CuPID prediction, and 23 of these (88.5%) of the predictions aligned with the confirmed tumour type or suspected diagnosis.
• The authors suggest that the next steps are further validation of CUPiD in larger cohorts or a prospective clinical trial.

Conclusion

This study suggests that CUPiD accurately predicts tumour type in cfDNA. It enables cfDNA mutation and methylation to be assessed simultaneously, tailoring treatment plans. The tumour types predicted by CUPiD have considerably different treatment strategies compared to CUP, almost all cases being considered for immuno or targeted therapies, potentially improving the standard of care for patients.

EPISTEM SERVICES

If you need to investigate molecular targets such as DNA methylation in challenging clinical samples, Epistem can help. Our science team has in-depth experience across a range of tissue types and disease areas. We offer a range of targeted panels, including:

• Cancer Hotspot

Somatic mutations across 50 genes with known associations to cancer.

• BRCA

Somatic and germline variants across BRCA1 and BRCA2.

• Immune Repertoire+

T cell diversity and clonal expansion through sequencing T cell receptor beta chain.

• Myeloid

Relevant targets for a range of conditions including acute myeloid leukaemia and myeloid dysplastic syndrome.

• HLA

Human leukocyte antigen typing across seventeen loci including all exons for all genes.

We also offer comprehensive data analysis support, with our bioinformatics team able to undertake gene expression, differential methylation, and genotyping analyses in-house. If you have project specific requirements, our BD team would be delighted to discuss them with you.