Skip to main content

Towards a diagnostic panel for cancer detection and classification based on DNA methylation

Wednesday, September 13, 2017 — Poster Session II

3:30 p.m. – 5:00 p.m.
FAES Terrace


  • G Margolin
  • N Jameel
  • Y-C Chen
  • HM Petrykowska
  • L Elnitski


Early detection of cancerous neoplasia is crucial for successful intervention and treatment of many cancers. It is especially important for cancer types that develop asymptomatically until late stages, such as ovarian and pancreatic cancers, when treatments can rarely save patient lives. Non-invasive biomarkers could provide the necessary tool for this purpose, however, biomarkers with high specificity and sensitivity are required for use in clinically applicable, blood-based diagnostic testing. In the event of early and late stage human malignancies, cell-free DNA (cfDNA) in blood plasma can have traces of circulating tumor DNA (ctDNA) that contain genetic and epigenetic signatures unique to the cell of origin. Thus, DNA methylation at specific loci can be used to identify DNA from specific tissues or tumor types making it a useful biomarker that survives long-term storage in fresh-frozen or formalin-fixed, paraffin-embedded samples and blood samples. We investigated DNA methylation array (Infinium HumanMethylation450 BeadChip) data to identify biomarkers as candidates for blood-based screening approaches and developed computational methods for combinatorial biomarker selection. Using DNA methylation data from 13 different tumor types, we report that a panel of 8 loci can robustly detect tumors of all considered types (4052 tumor samples) with sensitivity of 91.4%. Specificity was > 99.96% based on healthy peripheral blood (2711 samples), with a slightly lower value of 98.8% based on normal genomic DNA from 12 tissue types (646 normal samples). Furthermore, a panel of 39 loci can correctly identify the tumor type with a rate ranging from 98% to 69% (median 86%). As a proof of concept, we compared these results to enrichment bisulfite-sequencing data collected in the same genomic regions, using the highly parallel Fluidigm matrix system. DNA was collected from 25 solid tumor samples including five of breast, lung, stomach, liver, and colon, two normal samples from each tissue, and 13 normal blood samples. Despite noticeable data quality issues associated with low DNA amounts, these pilot results were in line with our array-based estimates. The ability to classify samples as tumor or normal, and identify the tumor type using a panel of 46 markers provides advancement towards a precision diagnostic tool for non-invasive cancer screening, monitoring tumor burden, and inferring drug sensitivities. Further experiments with blood plasma are in progress, and refinements of the biomarker panels will be made as needed.

Category: Genetics and Genomics