Colorectal cancer remains a leading cause of cancer‑related mortality worldwide, with limited therapeutic options targeting dysregulated inflammatory signaling. Glycoprotein 130 (gp130), the common signaling receptor for the IL‑6 cytokine family, drives oncogenic JAK2/STAT3 signaling and represents a high‑potential yet underdeveloped therapeutic target. A major bottleneck in developing gp130 inhibitors is the scarcity of known active compounds for machine learning modeling.
A collaborative team led by Prof. Wenying Yu and Prof. Yixian Liao from China Pharmaceutical University, together with international partners, established an innovative two‑stage AI‑assisted drug discovery pipeline to overcome this challenge. Using transfer learning, the researchers first pre‑trained a predictive model on the well‑characterized STAT3 inhibitor dataset (downstream of gp130) and then fine‑tuned it on a small curated gp130 inhibitor set. This approach efficiently prioritized active molecules from a library of 2,560 natural products under stringent ADMET and structural novelty (MCE‑18) filters, identifying evodiamine as a promising starting scaffold.
Rational structural hybridization and optimization yielded a series of indolopyridine derivatives, from which Compound 8a was selected as the optimal candidate. Biochemical and biophysical assays confirmed that 8a directly binds the gp130 D1 domain with high affinity (K_D = 2.17 μM), markedly superior to evodiamine, rutaecarpine, and bazedoxifene. Mechanistically, 8a selectively inhibits gp130‑mediated JAK2/STAT3 phosphorylation, blocks STAT3 DNA binding, and downregulates the oncogenic targets Bcl‑2 and Cyclin D1. Functional studies demonstrated that 8a robustly inhibits proliferation and induces mitochondrial apoptosis in HT‑29 colorectal cancer cells, with activity dependent on gp130 expression.
In HT‑29 xenograft models, 8a (20 mg/kg) achieved 56.20% tumor growth inhibition without noticeable systemic toxicity, outperforming the clinical gp130 inhibitor bazedoxifene. Preliminary metabolic stability assays in rat liver microsomes showed improved pharmacokinetic properties relative to evodiamine, supporting further translational development.
Notably, this work validates transfer learning as a robust strategy for target‑directed drug discovery under data‑scarce conditions, providing a blueprint for discovering inhibitors against other understudied cytokine receptors and signaling nodes. Compound 8a represents a structurally novel, mechanistically distinct gp130‑targeted antitumor candidate with strong potential for treating colorectal cancer and other malignancies driven by IL‑6/gp130 signaling.
This work, entitled “Transfer learning algorithm assisted in the discovery of novel gp130 inhibitors and their application in colorectal cancer treatment”, was published online March 20, 2026, in Targetome.
DOI: 10.48130/targetome-0026-0010