https://www.nature.com/articles/s41587-026-03081-9

Artificial allosteric protein switches with machine-learning-designed receptors

Key figures

• Figure 1: Shows that circularly permuted natural and machine-learning-designed receptor domains can drive ligand-dependent activation of a β-lactamase reporter without requiring global conformational change.
  
• Figure 3: Reveals the proposed mechanism by linking minimal secondary-structure change, altered local dynamics by ^19F-NMR, HDX-protected regions, and reduced ligand affinity after circular permutation.
  
• Figure 5: Demonstrates practical deployment of the switches in living bacteria and on an electrochemical device, moving the work beyond in vitro enzyme assays.
  

1) Thesis (one sentence)

To address the gap of constructing artificial allosteric switches without relying on natural receptors or large ligand-induced conformational changes, in chimeric enzyme, luminescent, bacterial, and bioelectronic biosensor systems, machine-learning-designed circularly permuted receptor domains cause ligand-dependent activation and logic-gated output by reducing conformational entropy and propagating local dynamic stabilization to distal reporter domains, supported by enzymatic, luminescent, electrochemical, HDX-MS, and ^19F-NMR evidence.

2) Evidence card (three bullets only)

• Strongest result: (Fig. 1e, Fig. 2b-h, Fig. 5) Artificial ML-designed steroid and peptide binders yielded functional single-component switches with outputs spanning nearly 400-fold dynamic range for a cortisol-responsive β-lactamase chimera, >20-fold gain from a duplicated YES-gate architecture, a fivefold AND gate, reporter portability to GDH and NanoLuc, fully synthetic LuxSit Pro-based switches, steroid-dependent bacterial growth, and a hormone-responsive bioelectrode.
  
• Method enabler: (Fig. 3; research type + tools) Circular permutation of compact designed receptor domains plus insertion into reporter loops enabled switch construction, and mechanistic dissection combined far-UV CD, ^19F-NMR with site-specific p-(O)CF3-tyrosine labeling, HDX-MS, competitive titration, ITC, and biolayer interferometry to show that ligand binding stabilizes dynamics rather than triggering a large fold change.
  
• Critical limitation: (Fig. 3g,h; Fig. 2g,h) Circular permutation that enabled switching also reduced receptor affinity by about four- to fivefold, and fully synthetic LuxSit Pro chimeras with one receptor showed only 2-3-fold dynamic range before receptor duplication, indicating that topology and linker architecture still require empirical optimization.
  

Optional

Quote bank (2–4 short excerpts)

• Quote 1: “machine-learning-engineered minimal ligand-binding domains act as efficient receptors in single-component allosteric switches” (Abstract, page 1)
  
• Quote 2: “despite lacking global conformational change” (Abstract, page 1)
  
• Quote 3: “ligand binding reduces the conformation entropy of the system” (Abstract, page 1)
  

Key comparisons (1–3 lines)

• Compared to: Traditional protein switches built from natural ligand-binding domains chosen partly for obvious ligand-induced macromolecular conformational change.
  
• Win: Shows that compact ML-designed receptors without global structural rearrangement can still generate allosteric enzyme, luminescent, logic-gate, cellular, and electrochemical outputs.
  
• Tradeoff: Performance remains architecture-sensitive, with meaningful penalties from circular permutation and some reporter combinations needing receptor duplication to achieve strong dynamic range.
  

Methods I might copy (protocol hooks)

• Construct design / Models: Open reading frames were synthesized and cloned into kanamycin-resistant pET-28a(+); receptor domains were circularly permuted at loop residues and inserted into reporter loops; examples include cpHCY129.1-35-BLA-253, cpOHPFA1952-20-BLA-253, 2cpOHPFA1952-20-BLA-41-197, cpOHPFA1952-20-cpCPH02-52-BLA-41-197, cpOHPFA1952-20-GDH-404, cpOHPFA1952-20-NanoLuc-157, and LuxSit Pro fusions.
  
• Conditions / Instruments: Proteins were expressed in E. coli BL21(DE3) in LB with 50 μg mL^-1 kanamycin, induced with 0.3 mM IPTG, and grown overnight at 18 °C; lysis used 50 mM Na2HPO4 pH 8.0, 300 mM NaCl, 20 mM imidazole, 1 mM AEBSF, and DNase I, followed by 27 kPsi cell disruption, Ni-NTA HisTrap FF purification on an ÄKTA Express, dialysis into 20 mM Tris-HCl pH 7.0 and 100 mM NaCl, and storage at -80 °C; representative assays used 100 nM chimera with 50 μM UW154, 10 nM GDH chimera, 20 nM NanoLuc chimera, 10 nM LuxSit Pro chimera, 2 μM 17-OHP, and 10 μM C-peptide; the bioelectrode used 25 mM HEPES pH 7.2, 100 mM Na2SO4, 3 mM calcium acetate, 10 mM glucose, 5 μM 17-OHP, and 5 mV s^-1 scanning versus Ag/AgCl/3 M KCl at room temperature.
  
• Readout / Analysis: Enzyme activity was fit to quadratic binding equations to estimate apparent Kd and dynamic range; ^19F-NMR used site-specific probes at positions 105, 229, 230, 256, 389, and 437 with 0.6 mM labeled protein and 1.5 mM 17-OHP; HDX-MS mapped apo-to-holo stability changes across the chimera; competitive titration and ITC quantified affinity loss after circular permutation; biolayer interferometry measured kon and koff using 17-OHP-PEG-biotin immobilized on streptavidin chips.
  

Open questions / Theoretical implications (2–5 bullets)

• How much baseline weak target affinity is needed before ligand-driven entropy reduction can be converted into a useful conditional binder rather than just a biosensor?
  
• Could receptor duplication or logic-gate architectures be repurposed to sharpen ON/OFF discrimination in protein-targeting systems, not just reporter assays?
  
• Does circular permutation mainly work by exposing a mechanically coupled helix-linker junction that can be redesigned, or by globally redistributing local conformational entropy?