Tumor habitats are phenotypically and spatially distinct intratumoral regions. Hypoxic habitats are associated with reduced accessibility of chemotherapeutics, resulting in resistance. Hypoxia targeted prodrugs (HAPs) may overcome this resistance. However, in a phase III clinical trial, the HAP evofosfamide (TH-302) did not improve survival of sarcomas when used in combination with the standard of care doxorubicin (Dox). This may have been due to a lack of patient stratification based on hypoxic status. Thus, identification of hypoxic tumor habitats (Jardim-Perassi et al. 2019) would allow for pre-therapy patient stratification, and monitoring of habitats post-therapy. Our goal was to use multiparametric (mp) MRI and co-registered histology to classify hypoxic habitats in preclinical models of sarcoma, and to use Habitat Imaging to monitor therapy response. A patient-derived xenograft (PDX) of rhabdomyosarcoma (obtained from St. Jude's) and a syngeneic model of fibrosarcoma (RIF-1) were randomized into four groups, receiving saline (control), Dox, TH-302, or the combination (Dox + TH-302). mpMRI was acquired pre and post therapy, and parameter maps were calculated for T2, T2*, diffusion weighted imaging (DWI), and Dynamic contrast-enhanced (DCE). MRI based 3D printed tumor molds were used to guide MRI and histology co-registration. Machine learning was used to classify MRI into viable, hypoxic and non-viable tumor habitats using their co-registered histological counterparts as ground-truth. TH-302 monotherapy or the Dox + TH-302 combination reduced tumor growth and increased overall survival in the PDX model (p<0.05). The combination therapy was superior, consistent with TH-302 controlling hypoxic habitats and Dox controlling viable normoxic regions. Pimonidazole-positive (hypoxic) areas were reduced with TH-302 monotherapy (p<0.05) but not with the Dox + TH-302 combination. This suggests that TH-302 decreases the hypoxic area, concurrently as Dox may have increased this habitat, possibly as a resistance mechanism in tumors undergoing prolonged therapy. Notably, the RIF-1 model was resistant to both Dox and TH-302. This is not due to inherent resistance, as RIF-1 cells in vitro were sensitive to TH-302 under hypoxia and to alkylating agents, indicating that resistance is dependent on an intact tumor microenvironment. Co-registration of MRI and histology allowed spatial comparison of imaging modalities to develop a machine learning method for classification of these tumor microenvironments. In conclusion, the use of HAPs to target hypoxic tumor habitats shows superiority over conventional chemotherapy in rhabdomyosarcoma, but was without effect in RIF-1. Imaging habitats have the potential to classify different microenvironments within a tumor, and this concept may allow the choice of therapy to be based on habitat distribution.

Citation Format: Bruna Victorasso Jardim Perassi, Wei Mu, Dominique F. Abrahams, Michal R. Tomaszewski, Jan Poleszczuk, Jack Higgins, Gary V. Martinez, Robert J. Gillies. Targeting hypoxic habitats with hypoxia pro-drug evofosfamide in preclinical models of sarcoma [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 2772.