Download this poster
Credits
- Robert J Osborne¹
- Fabio Puddu¹, Annelie Johansson¹
- Aurélie Modat¹
- Jamie Scotcher¹
- Riccha Sethi¹
- Shirong Yu¹ ²
- Nick Harding¹
- Mark Hill¹
- Ermira Lleshi¹
- Casper Lumby¹
- Jean Teyssandier¹
- Michael Wilson¹ ³
- Robert Crawford¹
- Tom Charlesworth¹
- Shankar Balasubramanian¹ ⁴ ⁵
- Páidí Creed¹
1 biomodal Ltd, The Trinity Building, Chesterford Research Park, Cambridge, UK.
2 Current address: Tagomics Ltd, The Cori Building, Little Abington, Cambridge, UK.
3 Current address: Department of Astrophysical Sciences, Princeton University, New Jersey, US.
4 Cancer Research UK Cambridge Institute, University of Cambridge, Cambridge, UK.
5 Yusuf Hamied Department of Chemistry, University of Cambridge, Cambridge, UK.
1. Introduction
Detecting cancer at an early stage can improve treatment success and patient survival. This study explores the potential of 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) as biomarkers for identifying early-stage colorectal cancer (CRC) in cell-free DNA (cfDNA). Whole-genome sequencing using duet evoC was used to analyse cfDNA from 37 treatment-naive CRC patients and 32 healthy individuals. Our results show that assessing 5mC and 5hmC separately significantly improves diagnostic accuracy (AUC = 0.95) compared to conventional methods that merge them into a single modified cytosine signal (modC, AUC = 0.66). Notably, 71.7% of differentially methylated regions (DMRs) with increased 5hmC in stage I cfDNA exhibited reduced 5mC in stage IV, indicating that 5hmC may serve as a marker of progressive demethylation during tumour evolution. These findings support the idea that distinguishing between 5mC and 5hmC enhances the sensitivity of liquid biopsy tests for early cancer detection.
Figure 1. duet multiomics solution evoC is a 6-base calling technology that reads all four canonical bases plus 5mC and 5hmC.
2. 5mC and 5hmC increase sensitivity of early cancer detection
Figure 2.
- Diagram illustrating the enzymatic steps in cytosine methylation and demethylation. DNA methyltransferases catalyse the transfer of a methyl group to cytosine within CpG sites. This modification is subsequently oxidized by TET enzymes to 5hmC and further oxidation products, 5‑formylcytosine (5fC) and 5‑carboxylcytosine (5caC). Thymine DNA glycosylase (TDG) then excises 5fC and 5caC, followed by gap-filling repair to restore unmodified cytosine.
- Illustration of genomic regions undergoing hypermethylation due to increased 5mC in advanced cancer. Changes in 5mC and total modified cytosine (modC) remain largely comparable, making both suitable as biomarkers for distinguishing disease progression.
- Depiction of hypomethylation in late-stage disease. In early cancer, 5hmC alterations are more pronounced than those of 5mC or modC. As demethylation progresses, 5mC and modC changes become more evident, though modC signals remain indistinct at early stages due to the conflation of 5mC and 5hmC, only emerging as distinct markers at mid-to-late stages.
3. Complete and accurate 5mC and 5hmC data for feature extraction
Figure 3.
- Volcano plot displaying differentially methylated regions (DMRs) between stage IV colorectal cancer (CRC) tissue and matched adjacent normal tissue, based on TCGA data. Point density is color-coded, with yellow indicating higher density.
- Venn diagram illustrating genomic regions with statistically significant differences (q-value < 0.05) in 5mC and 5hmC in stage I plasma.
- Scatterplots showing mean methylation differences in stage IV tissue (versus adjacent normal tissue) and mean methylation differences in stage I (upper panel) and IV (lower panel) cfDNA versus healthy control cfDNA. Each plot is annotated with the coefficient of determination R².
4. 5mC and 5hmC are synergistic biomarkers for early detection of CRC
Figure 4.
- ROC curves for modC only, 5mC only, 5hmC only, and 5mC and 5hmC models using generalised linear models and leave- one- out cross validation. The combined model outperforms individual markers in detecting stage I CRC from cfDNA.
- Box plot of AUC values across 500 sub-cohorts confirms that models incorporating both 5mC and 5hmC consistently achieve the highest performance, demonstrating that the signal is robust to changes in the cohort.
- Venn diagram of correctly predicted CRC samples (threshold 0.5) for 5mC only, 5hmC only, and 5mC and 5hmC models highlighting the distinct and complementary contributions of 5mC and 5hmC to model performance.
5. 5hmC contributes more at early stage and 5mC at late stage
Figure 5.
- In the comparison between stage IV patients and healthy controls, modC (AUC = 0.88) and 5mC alone (AUC = 0.83) provided the highest AUC values, indicating that 5mC contributes most of the discriminatory signal at later stages of cancer.
- 5hmC contributed significantly to distinguishing between stage I and stage IV CRC with similar AUCs for 5hmC alone (AUC = 0.95) and 5hmC and 5mC (AUC = 0.91), supporting the hypothesis that 5hmC offers additional discriminatory power in the earlier stages of cancer progression.
6. Conclusion
This study demonstrates that the combination of 5mC and 5hmC is highly effective for early-stage CRC detection in cfDNA. Using both biomarkers together offers a distinct advantage over using them individually and represents a significant improvement compared to conflating them into a single modC marker. The results highlight the crucial role of 5hmC in detecting early-stage changes in disease progression. We believe there is significant potential in applying duet evoC 6-base sequencing to larger clinical cohorts across various diseases to further assess how 6-base sequencing data could enhance early disease detection.
7. References
- Füllgrabe J. et al. Simultaneous sequencing of genetic and epigenetic bases in DNA. Nat Biotechnol. 2023 Oct;41(10):1457-1464.
- https://portal.gdc.cancer.gov/, January 2024
- Friedman J. et al.. Regularization Paths for Generalized Linear Models via Coordinate Descent. J Stat Softw. 2010;33(1):1-22.
- Tay J.K. et al. Elastic Net Regularization Paths for All Generalized Linear Models. J Stat Softw. 2023;106:1.
