Hierarchical Assembly of 2D and 3D Protein-DNA Co-Crystal Materials via AI-Designed Interfaces

PART 1: NON-TECHNICAL SUMMARY This project develops a machine learning protein design protocol in pursuit of advanced biotechnology in the form of a new family of crystalline materials made from both protein and DNA building blocks. At the molecular scale, these designed crystals are three-dimensional pegboards that can capture and hold other biological molecules in a fixed, repeating pattern. Once captured, the structures of these guest molecules can be determined using X-ray crystallography, a powerful technique that reveals the arrangement of atoms within a material. Understanding the atomic-level structure of proteins and other biological molecules is essential for advancing molecular biotechnology, yet obtaining such structures remains slow and unreliable for many important molecules. The proposed scaffold crystals have the potential to dramatically accelerate this process by providing a reusable, modular platform for structure determination. The resulting experimental datasets will also be important to enable further progress of next-generation machine learning models for predicting how proteins read DNA sequences. This research promotes the progress of science and advances national prosperity and welfare, consistent with the mission of the National Science Foundation. Integrated with the research, the project features a multi-level education plan designed to build the next generation of biomolecular engineers. First-year college students gain early research experiences through the Scott Undergraduate Research Experience program. Upper-level undergraduates join the team via an NSF Research Experience for Undergraduates program, and through a Biomolecular Engineering and Synthetic Biology course that is updated annually to teach cutting-edge machine learning methods for protein design. A cascading mentorship model connects graduate students, undergraduates, and high school interns, giving each level guidance and leadership experience. Finally, the project develops a hands-on chemistry education resource -- a book of step-by-step molecular model building instructions -- aimed at sparking curiosity about molecules in younger students before they ever set foot in a chemistry classroom. PART 2: TECHNICAL SUMMARY This biotechnology project combines machine learning (ML) protein design with structural DNA nanotechnology to create a new family of programmable protein-DNA crystalline scaffolds. RFdiffusion and ProteinMPNN are used to convert a monomeric DNA-binding protein (RepE54) into engineered dimers that position two DNA duplexes at a precise 60-degree crossing angle, and AlphaFold2 and AlphaFold3 validate candidate designs prior to experimental testing. Each dimer bridges two DNA duplexes to form a rigid "bowtie"-shaped tile. The research proceeds through three objectives in a hierarchical assembly strategy. In Objective 1, ML-designed protein dimers are expressed, purified, and combined with annealed DNA duplexes to form bowtie tiles. Tile geometry and stability are validated via size exclusion chromatography, mass photometry, fluorescence anisotropy, scanning fluorimetry, and atomic force microscopy (AFM). In Objective 2, asymmetric bowtie tiles are assembled into closed-form "supertiles" -- triangles, rhombi, and rosettes -- using programmed DNA sticky-end hybridization and hierarchical thermal annealing, borrowing protocols proven in DNA nanotechnology. In Objective 3, tiles and supertiles are assembled into two-dimensional and three-dimensional porous crystalline lattices. Chemical or enzymatic ligation stabilizes the lattices, creating robust frameworks with large solvent channels. DNA-binding guest proteins are then loaded into the scaffold crystals, and guest structures are determined via single-crystal X-ray diffraction (XRD). The central hypothesis is that protein-braced bowtie tiles yield rigid lattices capable of better than 3 Angstrom XRD resolution despite pore diameters exceeding 10 nm. A prototype ML-d NSF Award ID: 2607612 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Christopher Snow | Institution: Colorado State University, FORT COLLINS, CO | Award Amount: $360,004 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2607612 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2607612.html