Survival rates in paediatric oncology continue to improve with current 5-year survival after treatment reaching 80%. However, Cardiac Disease (CD) is the most common non-cancerous long-term adverse effect in this population, causing long-term morbidity and early mortality. Indeed, exposition to anthracyclines/or radiotherapy where the heart was exposed is significantly associated with CD. Nevertheless, these treatments alone cannot explain the individual variability in the prevalence and severity of CD. The current study aims to identify genetic expression biomarkers associated with CD occurrence after childhood cancer treatment, and to build a genomic signature linked to this cardiotoxicity.

We conducted a case-control study nested in the French Childhood Cancer Survivor Study (FCCSS) cohort. A total of 165 validated cardiac diseases cases and their individually matched controls on gender, age at first primary cancer diagnosis (± 3 years), type of first primary cancer diagnosis, and follow-up duration. The expression of 33.000 genes was obtained through a transcriptome microarray for all 330 patients. Clinical and detailed therapeutic information was retrieved in hospital records, and radiation dose to the heart and its structures were estimated. In order to identify the genes expression associated with CD, we performed a Lasso approach extended for matched case-control studies: the conditional logistic Lasso regularization. To obtain a robust selection of genes, we also used: Bolasso, Subslasso and Percentile Lasso. A selection threshold of 50% was set for the Bolasso and Sublasso, and the 95th percentile was applied for the Percentile Lasso. The intersection of the genes selections of the 3 methods formed our final selection.  Model classification performance was evaluated using Area under the Receiver Operating Characteristic curve (AUC).  In order to obtain a more robust AUC, an internal validation of the model was performed using Bootstrap, Split-Sample validation, and Cross-validation. The interactions of cancer treatments doses and selected genes have also been investigated.

For both cases and controls, median age at first cancer diagnosis was 6 years (IQR 2 – 11), and the median follow-up was 20 years (IQR 13 – 29) in cases, and their median age when developing CD was 28.6 years (IQR 36.56 – 20.15). The Bolasso method selected 5 genes, Sublasso 6 genes, and Percentile Lasso 45 genes. Finally 5 promising genes constituted the intersection of these approaches, among which 3 have a protective behaviour against CD, and 2 increase the CD risk. The addition of the selected genes expression to the model including clinical and treatment data, discriminated better between case and controls than the model based solely on clinical and treatments data (Cross-validated AUC = 0.57 vs. 0.90; P < 2.2 e-16). No interaction between treatment doses and expression levels of selected genes was found in our study.

In summary, using high-dimensional data selection methods, we were able to identify five genes potentially related to CD development. Knowledge of these genetic factors could allow us to better understand the biological mechanisms leading to cardiac toxicities, and thus to develop risk-adapted treatment and surveillance strategies.

Childhood Cancer Survivors, Cardiotoxicity, Biomarkers, Gene expression, Lasso

Les auteurs n'ont pas précisé leurs éventuels liens d'intérêts.