DNAR-06. Discovery and development of novel CNS-penetrating PARP1-selective inhibitors

Abstract Poly(ADP-ribose) polymerase (PARP) is a key enzyme in DNA damage repair (DDR) and inhibition of PARP1 specifically has been shown to be an effective treatment for cancers with DDR deficiencies, such as BRCA mutations. First generation PARP inhibitors have achieved commercial success in the treatment of BRCA-mutated cancers, but are limited in their clinical utility as they are unable to penetrate the blood-brain barrier (BBB) and therefore cannot be used to treat central nervous system (CNS) tumors. These inhibitors also show adverse side effects, likely due to their collateral inhibition of PARP2. Development of a novel PARP1-specific, CNS-penetrating drug could provide a new therapeutic option for patients with brain tumors, both primary and metastatic with the desired advantage of reduced toxicity. Here, we describe the use of artificial intelligence (AI) methods for the discovery and development of a novel, PARP1-selective and CNS-penetrating inhibitor. Deep docking is a process that uses deep learning to accelerate the prediction of binding of target proteins with a large compound library of 1.6 billion compounds. This was combined with generative AI and machine learning techniques that predict CNS penetrance to generate molecules predicted to be CNS-penetrating, potent, and selective for PARP1 inhibition. The most promising of these molecules were synthesized and validating data for these molecules will be presented, including in vitro PARP1 inhibition and selectivity, metabolic stability, in vivo pharmacokinetic profiles, and BBB penetration. This approach has enabled deep and rapid exploration of chemical space, accelerating the drug discovery process compared to traditional drug discovery methods.