A Combinatorial Functional Precision Medicine Platform to Evaluate Immunotherapy-Based Combinations in Non-Hodkin Lymphoma

PURPOSE:
Non-Hodgkin lymphoma (NHL) is a heterogenous malignancy with many different subtypes. Although patients initially respond to treatments, approximately 40% of this cohort will subsequently relapse. Beyond second-line salvage regimen, determining the next option is largely empirical. We have developed a combinatorial functional precision medicine platform, Optim.AI, which utilizes efficiently designed experiments and small data AI to predictively rank all actionable treatments within the list of interrogated drug panel. Utilizing this approach, subsequent lines of treatment can be rationally determined for patients who have exhausted the standard repertoire of therapies. The clinical feasibility of using Optim.AI to guide treatment has been previously demonstrated in NHL. However, the platform previously did not interrogate immunotherapy due to the limitations of the ex vivo model. This study will thus explore the feasibility of using Optim.AI in evaluating immunotherapy-based drug combinations.

METHODS:
Initial optimization of the in vitro co-culture model were carried out using T-cell lymphoma (TCL) cell lines, SU-DHL-1 or SR-786, with transformed, natural killer cell line, NK-92, at different effector to target cell ratios. An investigational monoclonal antibody (mAB) was added to evaluate the cytotoxicity of the co-culture models, as reflected from the cell death of the target cells.

These models were interrogated with 50 mathematically derived test combinations consisting of six drugs at varying concentrations for 48 hours. The drug panel consists of the investigational mAb and 5 drugs FDA-approved for TCL. Post-drug treatment cell death was measured using Caspase 3/7 assay and quantified using a high content imaging system, Operetta. The percentage of tumor cell death was subsequently used for Optim.AI analysis, to rank and compare 729 possible combinations to determine the top-ranked therapeutic options across the different models.

RESULTS:
In the optimization phase, there was a distinct dose-dependent increase in mAB-treated target cell death as compared to the target only controls at an effector to target cell ratio of 1:5. This ratio was eventually used for downstream experiments. Subsequent Optim.AI analyses highlighted differences in drug sensitivity between the two cell lines, where brentuximab-based combinations or investigational mAb-based combinations were top ranked for SU-DHL-1, while chemotherapy drugs were more dominant for SR-786. Additionally, the investigational mAb was identified as a higher ranked single drug and showed greater cell killing efficacy primarily in SU-DHL-1 as compared to SR-786, confirming prior dose optimization tests.

CONCLUSIONS:
As a proof-of-concept, we have demonstrated Optim.AI’s ability to evaluate immunotherapy drugs in combination by integrating Optim.AI with a high-content screening co-culture model. Moving forward, this workflow will be applied towards NHL primary patient samples to evaluate clinical feasibility in predicting immunotherapy prediction. In addition, these preliminary results suggest that Optim.AI can be useful in rapidly identifying appropriate partners to increase efficacy of immunotherapy-based combinations, for both current and emerging antibody candidates.