Background: Immunotherapy has emerged as an essential treatment modality for enhancing survival in triple negative breast cancer (TNBC). Despite demonstrated improvements in pathologic complete response (pCR), both toxicity and adverse events from immuno-oncology (IO) drugs remain a significant limitation. Currently, a lack of tests to differentiate patients likely to respond to IO vs. poor responders precludes a tailored approach to immunotherapy. Here we describe an imaging biomarker that allows physicians to target breast cancer patients with the highest likelihood of response to immunotherapies.
Methods: We identified a rapidly assessable, non-invasive biomarker of tumor response to immunotherapy. This biomarker uses radiological imaging (DCE-MRIs) coupled with the biophysical simulation platform TumorScope® to predict a patient’s likelihood of pCR following treatment with immunotherapy and backbone chemotherapy. While this biomarker does not depend on transcriptomic data, it was designed by matching biological function (transcriptomics) to tumor microenvironmental features (as observed in biophysical simulations) derived from the SimBioSys TumorBank®. With this simulation-derived biomarker in hand, we validated our methods in a small, independent cohort. We additionally applied the SimbIOScope (TM) IO-prediction analysis to assess a large immunotherapy-naïve cohort, in order to validate if an increase in pCR rates in the presence of immunotherapy correlated with the anticipated rate of response to immunotherapy.
Results: In TNBC tumors prior to neoadjuvant therapy, we found that a high immune evasive capability was associated with low nutrient utilization. Tumor immune evasion (including the PD-1/PDL-1 axis) is strongly correlated with tumor hypoxia (r = 0.45, p < 1×10-6). Similarly, in HR+/HER2- tumors prior to neoadjuvant therapy, we found that immune evasion was negatively correlated with angiogenesis (r = -0.40, p = 0.006), suggesting that low tumor vascularization is associated with immune evasion capability. As these associations were identified from available transcriptomic data obtained from a single biopsy site within each patient’s tumor, they were unable to account for tumor heterogeneity. We therefore sought to identify a spatially-resolved biomarker for immune evasive potential in TNBC and HR+/HER2- tumors. We used publicly available DCE-MRIs of patients treated with the immunotherapy drug pembrolizumab and paclitaxel from the ISPY2 trial (n=63) to train a model to predict pCR in IO-treated tumors. Critically, the resulting model’s predictive power matched that obtained from transcriptomics data. SimbIOScope was then tested on IO-treated patients in a small, independent cohort and correctly predicted pCR in >91% (n=12). We further validated SimbIOScope by predicting the expected pCR rate in 292 patients from our virtual TumorBank in response to immunotherapy. Consistent with empirical increase (13.6% in TNBC, Keynote522) anticipated as seen from clinical trials, we found that SimbIOScope predicted a 14% increase in pCR rate in TNBC patients (and 9% increase in HR+ patients) with addition of immunotherapy vs chemotherapy backbone alone.
Conclusions: The SimbIOScope platform offers physicians a rapid, non-invasive biomarker to differentiate patients likely to respond to immunotherapy from non-responders. This innovative technology thereby personalizes oncologic care and mitigates the potential for adverse effects by helping to optimize selection of patients best suited for immunotherapy.
