Toward prediction of abscopal effect in radioimmunotherapy: Pre-clinical investigation

Abstract

Purpose

Immunotherapy (IT) and radiotherapy (RT) can act synergistically, enhancing antitumor response beyond what either treatment can achieve separately. Anecdotal reports suggest that these results are in part due to the induction of an abscopal effect on non-irradiated lesions. Systematic data on incidence of the abscopal effect are scarce, while the existence and the identification of predictive signatures or this phenomenon are lacking. The purpose of this pre-clinical investigational work is to shed more light on the subject by identifying several imaging features and blood counts, which can be utilized to build a predictive binary logistic model.

Materials and methods

This proof-of-principle study was performed on Lewis Lung Carcinoma in a syngeneic, subcutaneous murine model. Nineteen mice were used: four as control and the rest were subjected to combined RT plus IT regimen. Tumors were implanted on both flanks and after reaching volume of ~200 mm ³ the animals were CT and MRI imaged and blood was collected. Quantitative imaging features (radiomics) were extracted for both flanks. Subsequently, the treated animals received radiation (only to the right flank) in three 8 Gy fractions followed by PD-1 inhibitor administrations. Tumor volumes were followed and animals exhibiting identical of better tumor growth delay on the non-irradiated (left) flank as compared to the irradiated flank were identified as experiencing an abscopal effect. Binary logistic regression analysis was performed to create models for CT and MRI radiomics and blood counts, which are predictive of the abscopal effect.

Results

Four of the treated animals experienced an abscopal effect. Three CT and two MRI radiomics features together with the pre-treatment neutrophil-to-lymphocyte (NLR) ratio correlated with the abscopal effect. Predictive models were created by combining the radiomics with NLR. ROC analyses indicated that the CT model had AUC of 0.846, while the MRI model had AUC of 0.946.

Conclusions

The combination of CT and MRI radiomics with blood counts resulted in models with AUCs of 1 on the modeling dataset. Application of the models to the validation dataset exhibited AUCs above 0.84, indicating very good predictive power of the combination between quantitative imaging and blood counts.