Molecular Diagnosis of Hereditary Syndromes and Cancer Using Genomic DNA Methylation
1. Department of Pathology and Laboratory Medicine, Western University, London, ON, Canada; 2. Molecular Genetics Laboratory, Molecular Diagnostics Division, London Health Sciences, London, ON, Canada; 3. Greenwood Genetics Center, Greenwood, SC, USA; 4. Children’s Hospital of Eastern Ontario Research Institute, University of Ottawa, Ottawa, ON, Canada; 5. Department of Pathology and Laboratory Medicine, McMaster University, Hamilton, ON, Canada; 6. Departments of Pediatrics, Biochemistry and Oncology, Western University and Children's Health Research Institute, London, ON, Canada
Introduction: DNA methylation of the CpG dinucleotides plays an integral role in the regulation of the processes that control normal development and organ stability. Aberrant DNA methylation in early development leads to neurodevelopmental syndromes, while its disruption in somatic tissues is associated with carcinogenesis. As a relatively stable functional modification, genomic DNA methylation has been the focus of biomarker discovery. Currently, the diagnosis of neurodevelopmental syndromes is challenging due to complex and overlapping clinical presentations and un-interpretability of the genetic variants of unknown significance. Similarly in prostate cancer, a non-germline disease, the current diagnostic methods based on pathology examination of prostate needle biopsies produce high false negative rates due to the temporospatial, molecular, and morphological heterogeneity of prostate adenocarcinoma.
Hypothesis: We hypothesize that both neurodevelopmental conditions and cancer generate specific DNA methylation patterns which can be utilized in molecular diagnosis and disease screening.
Methods: Peripheral blood samples from disease-specific cohorts of patients with Mendelian neurodevelopmental syndromes caused by defects in epigenomic machinery, as well as archival prostate cancer and benign tissues were assessed for genome-wide methylation changes using Illumina Infinium 450k and EPIC arrays. Supervised and unsupervised machine learning techniques, including support vector machine, LASSO, and unsupervised hierarchical clustering, were utilized to train predictive models for each disorder. Independent cohorts were used to validate the performance of the classification models.
Results: We identified highly sensitive and specific peripheral blood DNA methylation epi-signatures in a number of genetic conditions including Floating-Harbor Syndrome, autosomal dominant cerebellar ataxia, deafness, and narcolepsy, alpha thalassemia/mental retardation X-linked syndrome, Kabuki syndrome, Sotos syndrome, CHARGE syndrome, Claes-Jensen syndrome, Genitopattelar syndrome, Coffin-Siris syndrome, and Nicolaides-Baraitser syndromes. Using ~1,000 selected CpG probes from every epi-signature, we trained a multi-class machine-learning-based prediction model, enabling concurrent classification of the mentioned disorders, with 100% accuracy as determined by applying our model to multiple external and internal validation cohorts. We demonstrated the ability of the algorithm to identify undiagnosed subjects through screening a targetted population, to resolve the ambiguous clinical cases carrying variants of unknown significance, and to assign a new diagnosis to patients for whom the initial clinical diagnostic assumption was not correct. Similarly, in prostate cancer, using four CpG loci, we achieved 96% sensitivity and 98% specificity in differentiating tumors from benign tissues as confirmed using an external cohort of 234 tumors and 92 benign samples. We also showed that this method could sensitively detect metastatic lesions in bone, lymph node, and soft tissue.
Conclusion: This study demonstrates the power of DNA methylation analysis combined with machine learning in accurate detection of both hereditary genetic syndromes and acquired conditions.