S. Yadav, A. Stegmann, B. Helfrecht, H. Pyles, C. Mundy, S. Zhang, D. Baker, J. de Yoreo
Pacific Northwest National Laboratory,
United States
Keywords: biomolecular self-assembly, in-situ characterization, self-assembly dynamics, machine learning, Monte Carlo Simulation
Summary:
Understanding and controlling of dynamic assembly of proteins is essential to construct supramolecular structures and develop functional biomaterial. Recent advances in protein engineering have promoted accurate design of protein building blocks and protein-protein interface. Although several hierarchical structures that exploit the designed interfaces and the anisotropic nature of protein building blocks have been developed, the dynamics of protein monomer assembly into supramolecular structures are still not properly understood. To investigate the emergence of order, we used high speed AFM (HS-AFM) to capture the assembly of protein nanorods into two-dimensional crystal at liquid-mineral interface. The protein nanorods have arrays of up to 54 carboxylate residues geometrically matched to the potassium ion (K+) sublattice on muscovite mica (001) and have been previously shown (Pyles et al. Nat. Let. 2019) to form two-dimensional liquid-crystal phases on the mica surface. Depending on the K+ ion concentration and the underlying mineral atomic structure, these nanorods can assemble into high density smectic phase or orientationally disordered phase. The emergence of order with active rotational degrees of freedom is then explored using a computational workflow which combines artificial intelligence - based semantic segmentation and Fourier transformation convolution - based instance segmentation to analyze orientation and protein rod domain dynamics. Additionally, to gain a more fundamental understanding of the protein rod self-assembly, we performed an ensemble of Monte Carlo simulations for hard rods on a surface. By systematically varying the chemical potential, rod aspect ratio, and rod mobility across the simulation suite, we can determine which conditions lead to different levels of nematic ordering, smectic ordering, and high-density disorder. By comparing the simulation results against experiments, we can ascertain whether the hard rod model is sufficient to describe the experimentally observed protein rod self-assembly. This study is essential to demonstrate that protein–inorganic lattice interactions can be systematically programmed and are a steppingstone for designing protein–inorganic hybrid materials. Moreover, this study highlights the importance of high spatiotemporal resolution to visualize the rotational and translational motion and dynamics of nanorods during the emergence of order, the potential of artificial intelligence to analyze the emergence of this order and significance of Monte Carlo simulations to determine the parameters that lead to this order. With the ensemble of experimental, artificial intelligence and theoretical techniques we can extract the underlying physical mechanisms of assembly.