TL;DR
This study identifies a novel DNA methylation profile in peripheral blood that can distinguish breast cancer patients from healthy individuals with high accuracy. This profile is linked to immune system regulation, suggesting a new avenue for developing non-invasive early detection biomarkers for breast cancer.
What Problem Does This Paper Address?
Early detection of breast cancer is crucial for improving treatment outcomes and reducing mortality. However, current screening methods like mammograms and clinical exams have limitations, including costly false positives and overdiagnosis. While blood-based biomarkers offer a promising alternative, an accurate and reliable peripheral blood DNA methylation profile for early breast cancer detection is still lacking. Previous studies have primarily focused on European populations and used older array technologies. This paper aims to address this by identifying a robust DNA methylation profile in an Asian population using a more comprehensive array and advanced machine learning techniques, and to explore the biological mechanisms underlying this association.
Methods
The researchers employed a multi-step approach:
- Epigenomic Profiling: They used the Infinium MethylationEPIC array to profile DNA methylation at over 850,000 CpG sites in peripheral blood samples.
- Data Preprocessing and Quality Control: Standard procedures were used to remove low-quality probes, including those near known SNPs or that were cross-reactive. Samples with outlier values in principal component analysis or cell-type composition were also removed.
- Feature Selection: A rigorous feature selection process was implemented on the training set. This involved removing CpG sites correlated with DNA storage duration or patient treatment history. Subsequently, a Boruta feature selection method and simulated annealing with random forest were used to identify a core set of cancer-associated CpGs.
- Machine Learning Model Training and Evaluation: Three machine learning algorithms (Linear SVM, Random Forest, and XGBoost) were trained using the selected CpG features. The models were then evaluated on an independent testing set to assess their predictive performance.
- Benchmarking: The newly identified methylation profile was compared against four previously reported breast cancer-associated methylation profiles from the literature.
- Enrichment Analysis: To understand the biological relevance of the identified CpG sites, enrichment analyses were performed to identify transcription factors that bind to these genomic regions and to explore the functional pathways associated with nearby genes.
Datasets
- Training and Testing Cohorts: The study utilized a cohort of 524 ethnic Chinese individuals, comprising 256 breast cancer patients and 268 age- and ethnicity-matched healthy controls. This cohort was split into a training set (179 cases, 187 controls) and a testing set (75 cases, 81 controls).
- External Datasets for Benchmarking: Data from four previously published studies on breast cancer-associated blood methylation were used for benchmarking. These studies primarily used the HumanMethylation450 array.
Key Insights and Ideas
- Immune System Link: The most striking insight is the strong association of the identified methylation profile with immune-related pathways and transcription factors (specifically the AP-1 family). This suggests that the host's immune response to cancer, rather than solely cancer-specific genes, might be a key driver of the observed methylation differences in peripheral blood.
- Superior Predictive Performance: The newly identified 51-CpG methylation profile, when used with machine learning, significantly outperformed previously reported breast cancer methylation profiles, even those derived from larger CpG sets. This highlights the potential of focusing on specific, biologically relevant CpG sites.
- Population Generalizability (with caveats): While the study focused on an Asian population, one of the previously identified profiles (from a predominantly European cohort) performed reasonably well, suggesting some degree of generalizability. However, the authors caution that further studies are needed to fully characterize this.
- Beyond Cancer Genes: The findings challenge the conventional approach of solely targeting cancer-related genes for methylation biomarkers, emphasizing the importance of considering immune system interactions.
Key Results
- Predictive Accuracy: The identified 51-CpG methylation profile, when used with an XGBoost machine learning algorithm, achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.827 on an independent testing set. This demonstrates a 75% sensitivity and 78% specificity in distinguishing breast cancer patients from healthy controls.
- Benchmarking Performance: This profile outperformed all four previously reported breast cancer-associated methylation profiles, with the best previous profile achieving an AUC of 0.792.
- Enrichment Findings: Three transcription factors from the AP-1 family (JUND, BATF3, and FOS) were significantly enriched. Pathway analysis revealed enrichment in immune-related pathways, including IL-12, IL-21, Th17 cell lineage commitment, and NK cell activation.
Limitations and Future Work
- Confounding Factors: The study acknowledges potential confounding from DNA storage duration and treatment effects. While efforts were made to mitigate these during feature selection, the authors note that the predictive model showed no association with these factors.
- Orthogonal Validation: The lack of orthogonal methods (like pyrosequencing or qPCR) to validate the immune-related DNA methylation profile is a limitation, attributed to insufficient blood samples.
- Cohort Specificity: The study cohort consisted of patients with early-onset or a family history of breast cancer, recruited from genetic testing clinics. Therefore, the applicability of these results to the early detection of sporadic breast cancer requires further investigation.
- Future Directions: The authors suggest that further studies are needed to:
- Assess the identified methylation profile in larger breast cancer cohorts and in patients with other cancer types to establish specificity.
- Validate the findings using orthogonal methods.
- Explore the precise mechanisms by which immune-related methylation changes contribute to breast cancer development and detection.
- Investigate the generalizability of these biomarkers across different populations and environmental contexts.
Summary updated