AI-Driven Discovery of Novel STAT3 Inhibitors for Fibrotic Diseases

Introduction

Signal Transducer and Activator of Transcription 3 (STAT3) is a key transcription factor involved in numerous physiological processes, including cell growth, differentiation, and immune response. Dysregulation of STAT3 signaling has been implicated in various pathological conditions, particularly fibrotic diseases where excessive extracellular matrix deposition leads to organ dysfunction.

Despite its well-established role in disease progression, STAT3 has long been considered an "undruggable" target due to the challenges in developing selective inhibitors that do not interfere with related STAT family members or other signaling pathways. Previous attempts at STAT3 inhibition have been hampered by poor specificity, limited bioavailability, and significant toxicity.

Methods

AI-Driven Compound Discovery

Our approach leveraged a multi-stage computational pipeline integrating several machine learning models:

• Structure-based virtual screening: Using high-resolution crystal structures of the STAT3 SH2 domain, we developed a deep learning model trained on known protein-ligand interactions to identify potential binding pockets and predict binding affinities.
• Generative chemistry: A variational autoencoder (VAE) was trained on a library of drug-like molecules to generate novel chemical scaffolds with optimized properties for STAT3 binding.
• Multi-parameter optimization: Reinforcement learning algorithms guided compound optimization across multiple parameters including potency, selectivity, metabolic stability, and predicted toxicity.

Experimental Validation

Lead candidates identified through computational screening were synthesized and subjected to rigorous experimental validation:

• Biochemical assays measuring inhibition of STAT3 phosphorylation and dimerization
• Cellular assays in multiple fibroblast lines and primary cells
• Selectivity profiling against other STAT family members and kinases
• Pharmacokinetic and toxicity studies in multiple species
• Efficacy evaluation in mouse models of pulmonary and hepatic fibrosis

Results

Our AI-driven approach identified several novel chemical scaffolds with high affinity for the STAT3 SH2 domain. Through iterative optimization, we developed TTI-101, a first-in-class oral STAT3 inhibitor with exceptional selectivity and drug-like properties.

In preclinical models of pulmonary fibrosis, TTI-101 demonstrated dose-dependent reduction in collagen deposition and improvement in lung function. Similarly, in models of liver fibrosis, the compound significantly reduced fibrotic area and improved hepatic function markers.

Remarkably, TTI-101 not only prevented the progression of fibrosis but also reversed established fibrotic changes when administered in therapeutic regimens. These effects were associated with reduced expression of pro-fibrotic genes and decreased myofibroblast activation.

Discussion

The development of TTI-101 represents a significant breakthrough in targeting STAT3, a protein long considered challenging to drug effectively. Our AI-driven approach enabled the discovery of a chemical scaffold that achieves the delicate balance of potent target engagement with minimal activity on related proteins.

The observed efficacy in models of both pulmonary and hepatic fibrosis suggests broad potential in treating fibrotic disorders across multiple organ systems. The ability to reverse established fibrosis is particularly promising, as most current therapies can only slow disease progression.

Furthermore, the favorable safety profile of TTI-101, with minimal effects on physiological STAT3 signaling required for immune function and tissue homeostasis, addresses a key challenge in previous attempts to target this pathway.

Conclusion

This study demonstrates the power of AI-driven drug discovery in addressing previously "undruggable" targets. TTI-101 represents a first-in-class selective STAT3 inhibitor with promising efficacy in fibrotic disease models. These findings have led to the initiation of clinical trials, including the ongoing RENEW-IPF Phase 2 study in patients with idiopathic pulmonary fibrosis.

Our success in developing TTI-101 validates our computational approach and establishes a new paradigm for precision targeting of transcription factors and other challenging targets in drug discovery.