De novo design of pH-responsive self-assembling helical protein filaments

Nature leverages pH to regulate protein assembly, as seen in spider silk formation, pH-sensitive CTP synthase filaments during yeast starvation, and R-body force generation via pH-triggered conformational changes in extended lattices. These natural systems inspire bioengineering efforts to create pH-responsive protein materials with applications in tissue engineering, drug/gene delivery, and self-healing biomaterials. While prior approaches have mimicked silk domains or introduced histidines within coiled-coil motifs to achieve pH-dependent assembly, achieving de novo designed filaments with precise atomic structures and tunable pH transition points has remained unmet. The research question addressed here is whether computational protein design can generate self-assembling helical protein filaments composed of subunits bearing multiple buried, protonation-switchable histidines that confer sharp, tunable, and reversible pH-dependent assembly/disassembly. The authors hypothesized that embedding multiple histidines within buried hydrogen-bond networks in each subunit would produce highly cooperative transitions across extended filaments containing hundreds of subunits, enabling extreme sensitivity to small pH changes and allowing tuning of transition midpoints by modulating the networks.

The study situates itself within pH-responsive protein assemblies found in nature and engineered systems. Examples include spider silk proteins governed by a pH-sensitive relay (Nature 2010), pH-sensitive CTP synthase filaments (eLife 2021), and R bodies that undergo lattice-wide conformational changes upon pH shifts (ACS Synth. Biol. 2016). Engineered materials have incorporated pH-responsive behavior by mimicking silk domains and introducing histidine residues in staggered coiled-coil interactions (Biomacromolecules 2014). Prior de novo designs produced pH-dependent trimeric helical bundles containing nine buried histidines with tunable unfolding transitions between pH 4.0 and 6.5 (Science 2019). The present work builds on and extends these designs from finite oligomers to unbounded helical filaments with atomically precise architectures and controlled pH transition points.

Design strategy: The pH-dependent trimeric scaffold pRO-2.3 (PDB 6MSQ; six buried histidines) was converted to a single-chain monomer by designing short loops to connect the three protomers. Loop candidates were selected from a database of helical-fragment backbones via rigid alignment (≤0.35 Å RMSD), torsion-space alignment with soft constraints, and sequence design using profile constraints derived from source structures; lowest-scoring loops were chosen. Helical filament design: The linked monomers were docked into diverse helical arrangements using Rosetta-based helical docking and interface design. Approximately 45,000 filament backbones were generated. Filtering criteria included ΔE (bound−unbound) < −15 Rosetta energy units, interface area >700 Ų, shape complementarity >0.62, and <5 unsatisfied polar residues. Designs were manually refined by reverting non-contributory mutations, and top-scoring models were selected for experimental testing. Expression and purification: Genes for 18 designs were codon-optimized, cloned into pET29b+ (NdeI/XhoI), transformed into BL21*(DE3) E. coli, and expressed by Studier autoinduction at 37 °C for 24 h. Lysis (Bugbuster or microfluidization) was followed by Ni-NTA purification (wash 40 mM imidazole/500 mM NaCl; elution 400 mM imidazole/75 mM NaCl). Soluble/insoluble fractions were analyzed by SDS-PAGE. Negative-stain EM and fiber quantification: Samples in TBS (25 mM Tris, 75 mM NaCl, pH 8) were applied to glow-discharged, carbon-coated grids and stained (0.75% uranyl formate or Nano-W). Imaging on 100–120 kV TEMs; fiber lengths quantified with cryoSPARC’s filament tracer (template from DpHF19; curvature and cross-correlation filters applied). pH-response assays involved assembling at pH 8, lowering to specific pH values with citric acid (e.g., 6, 5, 4.2, 3.5, 3), and re-raising to 8 with 1 M Tris. Cryo-EM: Grids prepared on holey carbon and plunge-frozen. DpHF19 imaged on Glacios/K2 (counting mode), pixel 1.16 Å; DpHF18 and DpHF7 on Titan Krios/K2 (super-resolution, binned to 1.05 Å). cryoSPARC used for motion correction, CTF estimation, template-free and template-based filament tracing, 2D classification, and 3D helical refinement with non-uniform refinement. DpHF19 modeled with one-start symmetry relating non-contacting subunits; per-particle CTF refinements performed. Density modification via Phenix ResolveCryoEM; atomic model refinement in ISOLDE and Phenix; DpHF7 model built de novo and refined with RosettaCM and Phenix. Fluorescence microscopy (TIRFM): DpHF18 fibers labeled via cysteine conjugation (Oregon488 or sulfo-Cy5). Seeded growth assays mixed pre-assembled Cy5-labeled seeds with freshly disassembled Oregon488 monomers at pH 8, imaged by TIRFM to assess tip growth. Rapid disassembly assays flowed pH 3 buffer into a passivated flow cell while imaging Cy5 fibers at 16 ms/frame; bulk disassembly kinetics quantified via high-content imaging and CellProfiler segmentation. Liquid-phase AFM: Fibers weakly immobilized on poly-lysine–coated mica, imaged in tapping mode under liquid. pH-switching achieved by exchanging buffer from pH 8 to 4.7–4.1 while continuously imaging; individual fiber lengths and endwise disassembly tracked over time. Photoacid experiments used 1 mM 2-nitrobenzaldehyde (Tris pH 5.5) with local 405 nm laser or global 364 nm lamp activation to trigger spatially controlled pH drops and fiber disassembly, with appropriate controls confirming pH-mediated effects. Data, models, and code: Cryo-EM maps deposited (EMD-42075, -42070, -42088) and models in PDB (8UB3, 8UAO, 8UBG). Rosetta protocols and CellProfiler scripts available at https://github.com/shenh2/fiber_design.

- Two of 18 de novo designs (DpHF7, DpHF18) formed filaments by negative-stain EM; redesign of DpHF18 yielded DpHF19 with improved target architecture. - Cryo-EM structures: DpHF18 exhibited an unintended antiparallel D1 dihedral symmetry due to a secondary heterointerface deviating from design. Introducing five substitutions at the second interface (V29D, L82D to disfavor the D interface; E32D, E33K, E87K to strengthen the intended interface) produced DpHF19 with pure helical symmetry closely matching the computational model. - DpHF19 cryo-EM resolution 3.4 Å; Cα RMSD 1.4 Å over interface residues relative to design. One-start helical symmetry with rise 8.4 Å and rotation −148.9°; cross-section shows five subunits per ring; overall architecture is a right-handed helix of two identical, parallel strands. - pH-responsive assembly/disassembly is sharp and reversible, tunable by buried histidine count: • DpHF18 disassembled at pH ~3.5 and reassembled at pH 8. • DpHF19 disassembled at pH ~3 and reassembled at pH 8. • DpHF19_9his (nine buried histidines) disassembled at pH ~4.2 and reassembled at pH 8, shifting the transition higher relative to the six-histidine version. - Filament growth occurs from ends: TIRFM of Cy5-labeled seeds elongated by Oregon488 monomers showed increased Oregon488 intensity at both tips. - Rapid kinetics and sharp transitions: DpHF18 disassembled in <1 s when pH dropped from 8 to 3. Fiber lengths remained stable at pH 3.4 but rapidly decreased at pH 3.1, indicating a transition width of ~0.3 pH units. - AFM kinetics (DpHF19_9his): mean disassembly rates 108 ± 64 nm/min at pH 4.1; 21 ± 9 nm/min at pH 4.4; negligible at ≥4.5, indicating a very sharp transition over 0.1–0.3 pH units. Disassembly proceeds from both ends, sometimes asymmetrically due to surface interactions, and fibers can fragment before endwise disassembly. - Photoacid control: Local (405 nm) or global (364 nm) activation of 2-nitrobenzaldehyde in pH 5.5 buffer induced spatially programmable disassembly patterns by locally lowering pH; controls without active photoacid showed negligible disassembly. - The close packing in fibers lowers the disassembly pH relative to monomers (e.g., nine-histidine trimer PRO-2 disassembles at pH ~5.5 vs. fiber at ~4.2) and produces highly cooperative behavior with little change just above the transition.

The work demonstrates that de novo computational design can create unbounded, micrometre-scale protein filaments with precisely defined atomic structures that undergo sharp, reversible, and tunable pH-dependent assembly. By embedding multiple buried histidines in hydrogen-bond networks within each subunit, the authors achieved highly cooperative transitions across extended filaments, enabling extreme sensitivity to small pH changes. Structural validation by cryo-EM confirmed near-identity between the design and the realized DpHF19 architecture, while iterative redesign eliminated an alternative, unintended dihedral interface present in DpHF18. Functional studies using EM, TIRFM, and liquid-phase AFM established reversible disassembly with transition widths as narrow as 0.1–0.3 pH units and sub-second disassembly kinetics upon pH reduction. Increasing the number of buried histidines shifted the transition midpoint upward, validating the tunability of the pH response by design. These findings address the central challenge of engineering environmentally responsive protein filaments with predictable structures and behaviors, and provide foundational principles for constructing responsive nanomaterials whose cooperative behavior can be modulated by internal hydrogen-bond networks and subunit packing.

The study advances computational protein design by generating micrometre-scale pH-responsive helical filaments with atomically precise architectures and sharply tunable, reversible disassembly. Key achievements include cryo-EM validation of the designed helical symmetry (DpHF19), demonstration of rapid, cooperative disassembly over narrow pH windows, and control of transition midpoints via the number and arrangement of buried histidines. The authors highlight that close packing within fibers shifts disassembly pH below that of corresponding monomers and enhances cooperativity. These designed fibers can be integrated into higher-order materials (e.g., hydrogels) to impart environmentally triggered release and other functionalities. Future research directions include expanding the design space to achieve different pH ranges and response profiles, integrating external stimuli (e.g., light-activated photoacids) for spatiotemporal control, improving interface design accuracy with deep learning-based approaches, and elucidating assembly/disassembly intermediates with advanced in situ imaging methods.

Design success can be limited by small deviations in interface geometry that compound during filament propagation, potentially causing non-specific aggregation or altered geometries (as observed with DpHF18). Increasing interface design accuracy using recent deep learning methods (e.g., RFdiffusion, ProteinMPNN) may improve success rates. Experimentally, capturing early intermediates in assembly/disassembly remains difficult; advances in graphene-based liquid cells and sensitive direct electron detectors for liquid-phase TEM could help visualize these transient states.