Genetic predisposition to radiotherapy-induced toxicity

Abstract

[eng] During radiotherapy (RT) some healthy tissues are irradiated inevitably causing side effects, without being able to predict nowadays which patients will develop this toxicity. To date, the biomarker with the best evidence as a predictive test of late reactions is the radiation-induced lymphocyte apoptosis (RILA) assay. This thesis delves into exploring genetic predispositions to radiation-induced toxicities by analysing gene expression, single nucleotide polymorphisms (SNPs), and microRNA data in patients experiencing toxicity We first aimed to investigate the molecular basis underlying the distinctive RILA levels in patients with toxicity by using gene expression analysis in patients with and without late effects and in whom we had previously identified differences in RILA levels. Peripheral blood mononuclear cells of 10 breast cancer (BC) patients with late severe side effects and 10 patients without symptoms were mock-irradiated or irradiated with 8 Gy. The 48-h response was analysed in parallel by RILA assay and gene expression profiling with Affymetrix microarrays. While the differentially expressed mRNAs did not reach a significant adjusted p-value when comparing patients experiencing clinical toxicity versus those without, the enriched pathways revealed substantial differences between these groups. We found that patients at risk of developing late toxicity have a distinctive pathway signature driven by deregulation of immune and cell cycle pathways related to senescence, which in turn may underlie their low RILA phenotype. Secondly, we investigated the ability of a genetic profile, comprising candidate SNPs located in non-coding RNAs and variants previously linked to toxicity, to predict for RT- induced side effects in BC patients, and to create predictive models integrating the SNPs with relevant clinical variables. The study included 1,979 BC patients followed for two up to eight years after RT as part of a multi-centre study. We split the patient cohort into training and validation datasets according to recruiting centre. In the validation set, SNP- based models showed limited predictive power for both acute and late toxicity (0.49 and 0.54 AUC, respectively). The model using clinical variables alone or the combination of clinical and SNPs had the same performance with an AUC of 0.60 for acute and a c-index of 0.59 for late toxicity. In contrast, accuracy improved in the combination model, reaching 62% for acute and 57% for late toxicity. Our findings did not support a predictive value for the set of assessed non-coding RNA variants and underscore the significance of adding clinical data alongside genetic markers to model for RT toxicity in BC. Lastly, we aimed to identify differentially expressed microRNAs as novel biomarkers for radiosensitivity in whole blood of BC and prostate cancer (PC) patients. Small RNA sequencing was conducted in a discovery cohort, comprising 136 BC and 122 PC patients. Subsequently, a separate cohort was utilized for the validation of the top microRNAs, which included 205 BC patients along with 248 PC patients. In the BC cohort, 90 significantly differentially expressed microRNAs were identified between patients experiencing overall toxicity and those without it. We found pathways related to cellular senescence, transcription and signalling by TGFβ family members upregulated in the toxicity group. In the validation phase, we identified five microRNAs that maintained significant differential expression and fold change direction for BC patients. For the PC cohort, we did not find differentially expressed microRNAs or pathways between cases and controls. However, it is worth noting that we detected nine significant enriched pathways with a non-adjusted p ≤ 0.01 mainly related to cellular senescence and interleukin signalling. Although without adjusted significance, we confirmed in the validation cohort, the under-expression of one microRNA in patients with urinary retention compared to PC patients without toxicity. In summary, our research underscores molecular pathways linked to senescence potentially influencing both the observed late toxicity and RILA phenotype. Our developed predictive models emphasize the significance of integrating relevant clinical variables for toxicity prediction and indicated the limited predictive capability of the studied candidate SNPs. Additionally, we identified a five-microRNA signature that holds promise as a potential biomarker for predicting toxicity outcomes in BC patients.