Machine learning-driven rational design of ternary carrier-free nanomedicine: Huanghao Yang’s team at Fuzhou University achieves non-invasive intranasal administration for treating orthotopic and brain-metastatic non-small cell lung cancer

Background
The clinical treatment of non-small cell lung cancer (NSCLC) with brain metastasis faces severe challenges due to acquired drug resistance and inherent pharmacokinetic limitations such as poor bioavailability and high systemic toxicity. Carrier-free nanodrugs (CFNDs), which rely solely on intermolecular interactions for self-assembly without requiring the introduction of therapeutically inactive exogenous carriers, have emerged as a highly promising new drug delivery paradigm. This type of system can significantly improve the solubility and stability of poorly soluble drugs and achieve synergistic multidrug delivery through co-assembly, thereby effectively expanding therapeutic efficacy. However, the development of CFNDs is heavily dependent on the self-assembly properties of drug molecules, and especially in multicomponent systems, it remains constrained by empirical, low-efficiency screening approaches. Machine learning (ML) offers a powerful tool to overcome these cost-intensive and inefficient development bottlenecks, but existing models are mostly limited to binary-component systems and small-scale training datasets, making it difficult to support the rational design of complex multicomponent CFNDs.

Lipoic acid (LA), as an endogenous natural amphiphile, can significantly enhance drug solubility, stability, and self-assembly efficiency. Its unique dithiolane ring structure endows it with efficient transmembrane delivery capacity, enabling potent penetration of mucosal barriers and solid tumor barriers. Through dynamic thiol–disulfide exchange, LA achieves specific drug release in the tumor microenvironment with high glutathione (GSH) levels, providing key support for the synergistic treatment of primary tumors and secondary malignant metastases. Moreover, LA possesses dual pharmacological activity: it exerts antioxidant effects in normal tissues, alleviating chemotherapy-induced offtarget inflammation and toxic side effects, while acting as a pro-oxidant in the tumor microenvironment to help reverse immunosuppression. In addition, LA inhibits tumor progression and metastasis by regulating multiple signaling pathways and the epithelial–mesenchymal transition (EMT) process. In summary, LA offers unique and comprehensive advantages for constructing universal multicomponent nanomedicine systems.

Research Progress

Professor Yang Huanghao's team at Fuzhou University, in collaboration with Professors Li Jingying and Zhou Jie, employed endogenous lipoic acid (LA) as a universal self-assembling building block. They constructed a ternary coassembly database comprising 63 clinically used smallmolecule drugs, yeilding1352 distinct assembly combinations. After digitizing molecular features with extended-connectivity fingerprints (ECFPs), they systematically optimized four machine learning algorithms. The final ensemble model, Best@LA, achieved a 100% positive predictive rate in experimental validation, with all top-ten predicted combinations successfully forming stable nano-assemblies.

Based on the predictions from the above ensemble model, the research team further screened and constructed a ternary coassembled DTL nanomedicine composed of lipoic acid (LA), dabrafenib (Da), and trametinib (Tra) for targeted therapy of EGFRmutant nonsmall cell lung cancer. In this system, LA mediates a thioldependent nonendocytic delivery pathway, promoting efficient cellular uptake and endosomal escape of the nanomedicine. At the molecular level, LA simultaneously regulates multiple key pathways involved in tumor progression, including EGFR, MET, and MEK, as well as the epithelial-mesenchymal transition (EMT) process. Moreover, through its prooxidant properties, LA induces mitochondrial and endoplasmic reticulum stress, triggering downstream inflammatory cascades. Concurrently, the combination of dabrafenib and trametinib (Da&Tra) vertically blocks the MAPK (RAF-MEK-ERK) signaling axis, effectively reversing resistance to EGFR inhibitors. This assembly strategy maximizes drug bioavailability at the delivery level and, at the molecular level, establishes a multitarget, multimechanism synergistic therapeutic network that triggers multiple tumor cell death pathways—including apoptosis, pyroptosis, ferroptosis, and immunogenic cell death—thereby inhibiting tumor metastasis and remodeling the tumor immune microenvironment. This “drug repurposing” strategy not only expands the application scenarios of already approved clinical drugs but also achieves potent and synergistic tumor inhibition.

To address the critical clinical challenges of low drug brain penetration and poor blood–brain barrier crossing in the treatment of NSCLC brain metastasis, this study adopted a noninvasive intranasal administration route and performed systematic therapy in orthotopic and brainmetastatic xenograft mouse models of NSCLC. This noninvasive delivery strategy enabled efficient accumulation of DTL in primary lung tumors. Through a thiolmediated transbarrier transport mechanism, the nanomedicine rapidly penetrated the tight junctions of the nasal mucosal epithelium, bypassed the blood–brain barrier via the olfactory nerve and trigeminal nerve pathways, and achieved precise accumulation in intracranial metastases. As a result, it significantly suppressed tumor growth while minimizing chemotherapyrelated toxic side effects, demonstrating outstanding “efficacyenhancing and toxicityreducing” therapeutic advantages.

Future Outlook

This study has achieved several key advances. First, it established a highthroughput integrated “experiment–machine learning–prediction” research platform, overcoming the limitations of traditional empirical screening for nanomedicine formulation and providing a generalizable paradigm for the synergistic selfassembly of multicomponent drugs. Second, the rationally designed DTL nanomedicine derived from this platform has demonstrated, through omics and functional validation, significant therapeutic advantages and clinical translational potential for the treatment of EGFRmutant nonsmall cell lung cancer. Finally, the noninvasive intranasal administration strategy substantially improves patient compliance, minimizes systemic toxic side effects, and maximizes therapeutic benefits for both primary lung tumors and brain metastases.

Certain limitations exist in this study: current tumor accumulation relies primarily on passive targeting mediated by local drug delivery and the EPR effect, and the active targeting capability needs further enhancement. Future work will focus on expanding the structural diversity and generalizability of the machine learning database, as well as optimizing lipoic acid (LA) molecular derivatization to endow the nanosystem with stronger active tumortargeting ability. The ultimate goal is to provide safe and efficient clinical solutions for patients with advanced nonsmall cell lung cancer accompanied by brain metastasis, as well as for other related diseases.

Sources: https://spj.science.org/doi/10.34133/research.1180