69MO - 5-Hydroxymethycytosine analysis reveals stable epigenetic changes in tumor tissue that enable cfDNA cancer predictions

Background

5-Hydroxymethylcytosines (5hmCs) are stable epigenetic marks in the regions of active gene expression and regulation that underlie development and tumorigenesis. To investigate 5hmC patterns in tumor tissues, we characterized a collection of 267 fresh frozen tumor and non-tumor tissue samples using Bluestar Genomics’ genome-wide 5hmC profiling technology. The tumor samples represent five tissue types at various stages of cancer: breast, colon, lung, ovary, and pancreas.

Methods

Fresh-frozen samples of five tissue types were obtained by autopsy or surgery from individual subjects, 217 are tumor and 50 are non-tumor; cfDNA samples were obtained from 1009 cancer subjects and 1678 non-cancer subjects. Genomic or cfDNA were extracted, then enriched for 5hmC containing fragments. Whole genome and 5hmC-enriched libraries were generated and sequenced.

Results

Differential analysis found that up to 50% of all genes are hyper- or hypo- hydroxymethylated between tumor and normal samples and a global decrease of 5hmC in tumors across all tissue types. The 5hmC landscape reveals both a common theme across tissues as well as tissue-specific changes during tumorigenesis--changes that appear to be stable during cancer progression. We constructed a multinomial model that incorporates both non-tumor and tumor tissue 5hmC data. The model predicts Tissue of Tumor Origin (TOTO) and cancer status with 88% overall accuracy. There is a significant overlap of tumor-differentiating signatures between tissue and cfDNA, with highest concordance in colon, ovarian and pancreatic cancers, and weaker concordance in lung and breast; the strength of concordance trends with plasma cfDNA concentration. Further, we found the top tumor-differentiating 5hmC genes (DhMR) increase cancer prediction sensitivity over non-DhMR genes for cfDNA samples.

Conclusions

Our results suggest that the majority of 5hmC changes occur during the early onset of tumorigenesis and remain stable over cancer stage and therefore can be used to improve cancer and TOTO predictions for both early- and late-stage tumors as well as cfDNA from patients.

Clinical trial identification

Editorial acknowledgement

Dr. Elizabeth Stewart for her assistance in the writing of the abstract.

Legal entity responsible for the study

Bluestar Genomics.

Funding

Bluestar Genomics.

Disclosure

F. Feng: Stocks/Shares, Financial Interests, Personal: Bluestar Genomics. Y. Ning, Y. Xue, V. Friedl, D. Hann, B. Gibb, A. Bergamaschi, G. Guler, K. Hazen, A. Scott, T. Phillips, E. McCarthy, C. Ellison, R. Malta, A. Nguyen, V. Lopez, R. Cavet, S. Chowdhury, W. Volkmuth, S. Levy: Other, Institutional, Full or part-time Employment, employee only authors: Bluestar Genomics.