Introduction: Recently, kynurenine pathway enzymes have been identified as key regulators of cancer immunity. This has led to the development of several small molecule inhibitors targeting either Indoleamine 2,3-dioxygenase (IDO) or tryptophan 2,3-dioxygenase (TDO), but the results of such drugs were not satisfactory in clinical trials because of pathway redundancies and poor pharmacologic profiles. To overcome these challenges, we employed an artificial intelligence (AI) with deep learning technology to rationally design and discover a novel kynurenine pathway regulator with potent immunotherapeutic efficacy.

Method: We have established IDO- and TDO-specific deep learning models for prediction of compound-target interaction. Each model was applied to screen a chemical library of nearly 2 million small molecules, and the high-ranked compounds were tested for their inhibitory activity in enzyme- and cell-based assays. CT26 colon cancer-bearing mice were treated with a lead compound and/or immune checkpoint inhibitors. Tumors were analyzed by histologic, flow cytometric, and immune profiling assays.

Results: Among dozens of experimentally validated compounds, we identified a novel core structure that exhibits inhibitory activity against both IDO and TDO. A lead compound (designated hereafter as STB001) was then derived from the verified core structure through structure modification and subsequent in vitro assays. STB001 actively inhibits IDO and TDO in enzyme- and cell-based assays. Moreover, it could effectively suppress plasma kynurenine level in an in vivo endotoxin-induced inflammation model. Oral administration of STB001 to tumor-bearing mice led to remarkably delayed CT26 colon cancer growth in a dose-dependent manner. It also showed a strong activation of adaptive anti-cancer immunity with robust T cell proliferation in lymphoid organs and enhanced T cell trafficking into tumor microenvironment. STB001 treatment increased CD8+ cytotoxic T cells and decreased CD4+CD25+Foxp3+ regulatory T cells. The optimal dose and schedule of STB001 (100mg/kg bid, 5 days on and 2 days off) was determined by immunotherapeutic efficacy and toxicity. Finally, the efficacy of STB001-based immunotherapy was further strengthened by concurrent treatment of immune checkpoint inhibitors (αPD-1 and/or αCTLA-4), leading to complete regression of tumors.

Conclusion: Our study demonstrates that AI modeling with deep learning is a valid strategy for a rational and effective development of an immunotherapeutic drug. This AI-based platform can be applied to other molecular targets to speed up the immuno-oncologic drug development.

Citation Format: Won Suk Lee, Jeong Hun Kim, Hye Jin Lee, Hannah Yang, So Jung Kong, Yu Seong Lee, Jong Hui Hong, Jongsun Jung, Beodeul Kang, Hongjae Chon, Chan Kim. Artificial intelligence technology enables a rational development of a potent immunotherapeutic agent [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 2092.