RADIomics to Predict HER2 Status And T-DXd Efficacy in Metastatic Breast Cancer: the RADIOSPHER2 Study

Overview

RADIOSPHER2 study is a monocentric, retrospective, observational study aiming at identifying a radiomics signature able to predict HER2 expression (0 vs low vs overexpression) and trastuzumab deruxtecan efficacy in metastatic breast cancer patients. The study also encompasses translational analyses and inter-modal correlations in order to provide novel insights about HER2 spatial and temporal heterogeneity, at the macroscopic and microscopic levels.

The primary objective of the study is the identification of a radiomic signature ("radiobiopsy") from several features extracted from radiological images to predict the HER2 status on different specific metastatic sites of metastatic breast cancer patients. Accordingly to the study rationale, it will chosen to analyse images from those metastatic sites usually difficulty to approach with biopsy, such as lung, liver, pleural and bone lesions. The "radiobiopsy" model will be applied to retrospectively calculate HER2 status from the baseline scan of HER2+ or HER2-low mBC patients underwent any-line trastuzumab deruxtecan (T-DXd) treatment for metastatic disease. The T-DXd PFS of patients assigned to the three cohorts (HER2-0, HER2-low or HER2-overexpressed) based on "radiobiopsy" model will be then calculated and differences among groups will be tested with Cox regression analysis. Clinical characteristics (e.g., number of therapy line for metastatic disease, HR status, HER2 IHC expression) of the same cohort will be extracted from clinical health records, in order to build a multivariate Cox regression model. To assess the HER2 spatial heterogeneity at intra-tumor microscopic level, pathology slides derived from biopsies of patients enrolled to build the "radiobiopsy" model will be also reviewed from pathologists for HER2 heterogeneity evaluation and digitalized for pathomics analyses. The HER2 heterogeneity will be evaluated by pathologist based on IHC expression and distribution. In order to evaluate if radiomic features can predict HER2 heterogeneity at intra-tumor macroscopic level, a different radiomic model will be developed for this purpose, using the HER2 heterogeneity score assigned by pathologist in the previous task as outcome. In addition, in order to confirm and quantify inter-lesion HER2 spatial heterogeneity, the features included in the discovered "radiobiopsy" signature will be extracted and the HER2 status calculated for the other metastatic lesions. A correlation analysis between HER2 status assignment from different lesions of the same patients will be performed to quantify the prevalence of HER2 variability among lesions in the cohort. In order to clarify if HER2 heterogeneity at microscopic level is associated with inter-lesion heterogeneity, a single-patient correlations between the heterogeneity pathomic score and the inter-lesion radiomic heterogeneity will be performed. To have insight into the features included into the "radiobiopsy" model, transcriptomics analysis will be conducted through bulk-RNA sequencing from available tumor FFPE specimens of lung, pleural and liver biopsies. The differential expression of several genes and transcriptomic signatures with known prognostic/predictive relevance, such as immune and metabolic genes and signatures, will be assessed among samples with differently expressed features included in the model. It will allow to dissect tumor biology by finding the biological counterpart of agnostic radiomics features that would remain otherwise unexplained.