设计ATP门控蛋白用于肿瘤选择性药物递送

Chemotherapy remains limited by lack of specificity in many existing treatments, motivating the need for therapeutic strategies that exploit molecular features unique to the tumor microenvironment (TME). A defining hallmark of the TME is the elevated concentration of extracellular ATP (eATP), which can rise ~1000-fold from nanomolar levels in healthy cells to ~100 µM in tumors. eATP promotes tumor growth and immune evasion, making it an attractive biochemical feature for tumor targeting. In this work, we used state of the art deep learning-based protein design methods to generate scaffolds capable of selectively binding ATP and experimentally validated their binding as an initial step toward developing ATP-gated systems for tumor-specific drug delivery. To accomplish this, we adapted our lab's Neural Iterative Selection and Expansion (NISE) workflow, which integrates deep learning-based sequence design and structure prediction. In this framework, ATP was first docked into candidate protein scaffolds to define the binding-site geometry. A sequence design model then generated amino acid sequences predicted to fold into stable structures accommodating the positioned ligand, after which a structure prediction network modeled the complex to assess self-consistency between the designed sequence and structure. The designed sequences were then filtered based on high structural similarity (indicating designability) and high confidence (reflecting complex plausibility), and top-performing designs were used as input for the next round of sequence generation. Across twenty NISE refinement cycles, the algorithm progressively enriched for protein-ATP complexes by optimizing the joint space of sequence, structure, and ligand conformation. The workflow ultimately produced ~2,600 candidate complexes, from which a small set of high-quality designs were selected for experimental validation. The designs were tested using NMR spectroscopy. One top construct displayed ¹H-chemical-shift perturbations upon ATP titration (Kd ≈ 250 µM), consistent with the predicted isoleucine-adenine stacking interaction. Subsequent design rounds aimed to increase affinity by introducing features observed in natural ATP-binding proteins such as π-π stacking and expanded hydrogen-bonding motifs. In addition, the candidate protein structures were modified from 4-helix to 7-helix bundles to expand and solvate the binding pocket, enhancing hydrogen bonding, electrostatic complementarity, and water-mediated interactions. Together, these efforts represent the first step toward developing ATP-gated protein systems that can serve as tumor-selective drug delivery platforms, reducing the side effects commonly associated with chemotherapy in future applications.