A NEW INTEGRATED MODEL TO PREDICT NEOADJUVANT TREATMENT RESPONSE IN BREAST CANCER PATIENTS

A NEW INTEGRATED MODEL TO PREDICT NEOADJUVANT TREATMENT RESPONSE IN BREAST CANCER PATIENTS Background and rationale. Neoadjuvant systemic therapy (NST) has improved outcomes for locally advanced and early breast cancer (BC) patients, expanding surgical indications, facilitating breast-conserving surgery, and providing insights into individual long-term prognosis, since pathological complete response (pCR) is considered a surrogate endpoint for invasive disease-free and overall survival. Neoadjuvant treatment response is affected by a complex tumor-host interaction in which the role of environmental factors has not yet been completely elucidated. Microbiota, identified in extra gut sites like breast tissue, may influence the immune-metabolic profile of BC cells. [18]F-FDG PET/CT scan, allows advanced image analysis using computer-automated software to extract high-throughput quantitative features that quantify disease heterogeneity. In the present study, we aimed to combine advanced imaging features from [18]F-FDG PET/CT scan with microbiota characterization and clinical-pathological data in order to develop a robust NST prediction model. Methods. BC patients eligible for standard NST were enrolled in a prospective observational study. Before starting NST, a staging [18] F-FDG PET/CT scan was performed, and BC tissues, blood, oral, and fecal samples were collected. BC tissues, fecal and oral samples were employed for microbiota analysis. After NST, patients underwent a re-staging [18]F-FDG PET/CT scan. Patients eligible for genetic counseling were evaluated during NST, and a blood sample was collected for genetic testing. The association of pCR with clinical-pathological characteristics, tissue microbiota, genetic information, and imaging features from [18]F-FDG PET/CT scan was analyzed. Features were selected using least absolute shrinkage selection operator regression (LASSO) and a support machine learning (ML) model was employed. Model performance was evaluated using the receiver operating characteristic (ROC) and calculating the area under curve (AUC). Results. A total of 110 patients were enrolled: 58 (52.4%) HER2 positive BC, 11 (9.5%) luminal-like BC, and 41 (38.1%) triple negative BC. At surgery, the pCR rate was 46.6%. In microbiota analysis, when comparing the prevalence at the genus level with treatment response, we observed that Corynebacterium was present in 19/51 patients (36.7%) who achieved pCR and in 35/56 (62.5%) patients with residual disease (p=0.01). No differences in the prevalence of other baseline bacteria were observed between pCR and non-pCR patients. Different integrated models were then generated to predict response to NST, with the best performances achieved by ‘Combined Model 1’ and ‘Combined Model 2’. ‘Combined Model 1’ (AUC: 0.75 ± 0.11) integrated data on germline Pathogenic variants (GPVs), tissue presence of Corynebacterium, and three radiomic features (i.e., ‘GLSZM Zone Size Variance’, ‘Intensity-based kurtosis’, and ‘GLCM Cluster Shade’). ‘Combined Model 2’ (AUC: 0.79 ± 0.09) included data of GPVs, presence of Corynebacterium in BC biopsy, and three radiomic features (NGTDM Busyness’, ‘GLCM Cluster Shade’, and ‘Intensity Histogram Variance SUV’). Conclusion. This study represents the first integrated translational model for predicting pCR to NST, combining GPVs, microbiota in the baseline biopsy, and radiomic features. This represents a significant step towards understanding the mechanisms linking tumor microenvironment and treatment response in the BC neoadjuvant setting. Further validation studies with larger cohorts are required to confirm these results. If confirmed, our model could pave the way for pre-treatment patient stratification, thereby providing personalized and more effective treatments to increase patient survival.