A new cancer drug candidate developed by Lawrence Livermore National Laboratory (LLNL), BBOT (BridgeBio Oncology Therapeutics) and the Frederick National Laboratory for Cancer Research (FNLCR) has demonstrated the ability to block tumor growth without triggering a common and debilitating side effect.
In early clinical trials, the compound, known as BBO-10203, has shown promise in disrupting a key interaction between two cancer-driving proteins - RAS and PI3Kα - without causing hyperglycemia (high blood-sugar levels), which has historically hindered similar treatments. Published in Science, the findings mark a major milestone for the collaboration, offering a potential breakthrough for patients with aggressive, treatment-resistant cancers.
The discovery of BBO-10203 brings together DOE high-performance computing with AI and biomedical expertise to accelerate drug discovery. LLNL is leveraging its Livermore Computer-Aided Drug Design (LCADD) platform - combining AI and machine learning with physics-based modeling - and world-class DOE supercomputing resources like Ruby and Lassen, to simulate and predict drug behavior long before any compound is synthesized.
"This is a precise, targeted strike on a long-standing cancer vulnerability," said LLNL Biochemical and Biophysical Systems Group Leader Felice Lightstone, co-author of the study. "What's especially exciting is that this was achieved using a computational pipeline - reducing what traditionally takes many years."
A "breaker" disrupting the RAS-PI3Kα pathway
BBO-10203 works by blocking the interaction between two proteins that often help cancer grow. These proteins - part of the RAS and PI3K pathways - are frequently mutated in cancer but have been notoriously difficult to target safely and effectively with drugs. What makes BBO-10203 different is how precisely it cuts off the cancer signal without interfering with normal blood sugar control - a common problem in existing treatments, according to researchers.
In lab tests and animal models, the drug candidate slowed tumor growth across several cancer types, including HER2-positive, PIK3CA-mutated and KRAS-driven cancers. It also enhanced the effectiveness of existing therapies used to treat breast, lung and colorectal cancers, suggesting it could be combined with standard treatments to improve outcomes.
The development of the BBO-10203 molecule - which the team called the "breaker" for its unique ability to disrupt RAS-PI3Kα binding - traces back to a 2018 collaboration initiated by FNLCR scientists and builds on years of foundational work in structural biology, particularly efforts to understand and model the interaction between two key proteins frequently mutated in cancer.
"Our six-year journey from concept to clinic addresses the urgent need to target the interaction between the two most common cancer drivers: RAS and PI3Kα," said Dhirendra Simanshu, lead author and principal scientist at FNLCR. "We discovered a first-in-class way to block this interaction in tumors without affecting insulin signaling. This achievement highlights how strategic partnerships among BBOT, LLNL and the National Cancer Institute's RAS Initiative at FNLCR can translate structural biology insights into novel therapies, advancing cancer treatment from bench to bedside."
FNLCR researchers began with a "molecular glue" compound that stabilized the RAS-PI3Kα interaction and enabled detailed structural studies. Recognizing that this interaction could also be disrupted, they conceived the idea of converting the glue compound to breaker, and through close collaboration with BBOT and LLNL, the team designed key features of the molecule to block the binding interface rather than stabilize it.
With early compounds and insights on more than 50 crystal structures the FNLCR team solved during lead optimization, BBOT and LLNL's LCADD platform iteratively refined the molecule for potency, selectivity and pharmacokinetics. This work transformed the compound into a therapeutic candidate, targeting a previously "undruggable" protein interface and laying the foundation for BBO-10203's development.
HPC-driven drug discovery: from molecule to medicine
The rapid design and development of BBO-10203 is part of a larger effort to apply DOE computing capabilities and AI/ML for drug discovery. In six years, the LLNL/BBOT/FNLCR team has advanced three small-molecule cancer drug candidates into clinical trials, BBO-10203 being the second to reach patients. The first - BBO-8520 - entered human trials in 2024 and targets KRASG12C mutations in non-small cell lung cancer.
"This collaboration represents the future of cancer drug discovery - faster, smarter and more direct," said Pedro Beltran, chief scientific officer of BBOT and co-lead author of the paper. "We're excited by these results and the potential to expand treatment options for patients with numerous types of previously undruggable cancers."
BBO-10203's Phase 1 trial involves individuals with advanced tumors, including breast, colorectal and lung cancers - some of the most common cancers driven by RAS protein mutations. The goal is to evaluate the drug's safety, dosage and preliminary efficacy.
Traditional cancer-drug development is time and energy-intensive, costly and fraught with setbacks. But with a computational-first approach combining AI, simulation and structural modeling, researchers were able to dramatically reduce the cost and timeline of drug development to design molecules before synthesizing them in the lab and increase the odds of success.
After FNLCR's structural biology team helped define the protein-drug molecule binding site, researchers used the LCADD platform to evaluate millions of molecules, narrowing the field to a few top candidates for lab validation. These compounds were evaluated in biochemical and cellular assays, and their binding poses were determined through crystallography. Through this design loop, the team produced a highly selective molecule with a novel mechanism and improved pharmacological properties, advancing the candidate to clinical testing.
"This is about moving faster without cutting corners," Lightstone said. "We're combining cutting-edge DOE supercomputing with state-of-the-art chemistry and biology, and we're delivering results."
The computational work was supported by LLNL's Institutional Computing Grand Challenge Program, with experimental validation carried out in collaboration with BBOT and FNL. Researchers at FNLCR also leveraged DOE user facilities, including the Advanced Photon Source at Argonne National Laboratory, to guide structure-based design.
As clinical data from BBO-10203 continues to emerge, researchers are optimistic about its potential to set a new standard for PI3Kα pathway inhibitors and hope the compound could represent a new class of cancer therapeutics that avoids the toxicities of previous generations.
"We've built a powerful engine for drug design - and we're just getting started," Lightstone said.
LLNL's effort began with a Cooperative Research and Development Agreement (CRADA) with Theras/BBOT aimed at advancing discovery of novel RAS inhibitors for the treatment of cancer. The CRADA and license agreement with BBOT for the drug candidate were negotiated through LLNL's Innovation and Partnerships Office by Business Development Executive Yash Vaishnav.
