henhaobio
Sign in

Nature CommunicationsBiochemistryPeer-reviewed journal article

Improved protein binder design using -pairing targeted RFdiffusion

Isaac SappingtoniD, Martin TouliD, David S. LeeiD, Stephanie A. Robinson, Inna GoreshnikiD, Clara McCurdyiD, Tung Ching Chan, Nic Buchholz, Buwei HuangiD, Dionne K. VafeadosiD, Mariana Garcia-SancheziD, Nicole RoullieriD, Matthias GlögliD, Christopher J. KimiD, Joseph L. WatsoniD, Susana Vázquez TorresiD, Koen H. G. VerschuereniD, Kenneth VerstraeteiD, Cynthia S. HinckiD, Melisa Benard-ValleiD, Brian CoventryiD, Jeremiah Nelson SimsiD, Green AhniD, Xinru WangiD, Andrew P. HinckiD, , David A. BakeriD

PDFDOIPreprint
In-depth full-text digest available · Jump to digest ↓

Abstract

Despite recent advances in the computational design of protein binders, designing proteins that bind with high affinity to polar protein targets remains an outstanding problem. Here we show that RFdiffusion can be conditioned to efficiently generate protein scaffolds that form geometrically matched extended beta-sheets with target protein edge beta-strands in which polar groups on the target are nearly perfectly complemented with hydrogen bonding groups on the design. We use this approach to design binders against a set of therapeutically relevant polar targets (KIT, PDGFR[a], ALK-2, ALK-3, FCRL5, and NRP1) and find that beta-strand-targeted design yields higher affinities and success rates than unconditioned RFdiffusion. All by all binding experiments show that the designs have affinities ranging from 137 pM to mid nM for their targets and essentially no off target binding despite the sharing of beta-strand interactions, likely reflecting the precise customization of interacting beta-strand geometry and additional designed binder-target interactions. A co-crystal structure of one such design in complex with the KIT receptor is nearly identical to the computational design model confirming the accuracy of the design approach. The ability to robustly generate binders displaying high affinity and specificity to polar interaction surfaces with exposed beta-strands considerably increases the range and capabilities of computational binder design.

In-depth summary

Claude Sonnet 4.6 · max effort · 8 Nov 2026 · from the full text of version 2 (19 Nov 2024) · CC BY 4.0

Conditioning RFdiffusion on edge beta-strand pairing yields specific binders to polar targets, with in silico success above 9.2% versus 0.98% for hotspot-only conditioning.

Polar target surfaces need precisely placed hydrogen bond partners, and hotspot-only RFdiffusion often gives backbones that fit the shape but not the hydrogen bonding. Conditioning on strand pairing with exposed target edge beta-strands addresses this, including for targets that earlier design campaigns failed to bind.

Key results

  1. ShownBeta-strand interface conditioning raised in silico design success rates and gave more compact binder scaffolds than hotspot-only RFdiffusion, with successes on all seven targets.

    Over 9.2% of strand-conditioned designs met quality metrics vs 0.98% for hotspot-only; hotspot-only gave no in silico successes for ALK-3 and FCRL5; 88.7% vs 25.5% met radius of gyration criteria (500 scaffolds per approach). · beta-strand pairing conditioning improves…

  2. ShownExperimentally tested designs gave binders to all targets with SPR affinities from picomolar to mid-nanomolar, including targets where earlier design attempts had failed.

    Kd: ALK-2bp 164 nM; ALK-3bp 528 pM; α-CTXbp 1.9 nM; NRP1bp 84 nM; FCRL5bp 82 nM; KITbp 64.7 nM; PDGFRɑbp-7LBF 137 pM; PDGFRɑbp-AF2model 4 nM. · TGF-beta superfamily Targets: ALK-2 and ALK-3; Single Exposed Edge beta-strand: ɑ-CTX, FCRL5…; Type III Receptor Tyrosine Kinases: KIT and…

  3. ShownAll-by-all SPR showed each binder responded most strongly to its designed target; the ALK-2, ALK-3, KIT and PDGFRɑ binders showed no evidence of off-target binding to related family members.

    At 1.5 μM and 200nM binder concentrations, every binder gave its strongest response on its cognate target; ALK-2bp did not bind strongly to ALK-3; the all-by-all matrix included ALK-2, ALK-3, KIT, PDGFRɑ, FCRL5, NRP1 and α-CTX. · Beta-strand pairing binder interactions are…; Surface plasmon resonance for measuring…

  4. ShownThe KITbp:KIT co-crystal structure closely matched the design model and showed an extensive hydrogen bond network at the interface.

    2.8 Å resolution; 1.9 Å all-atom RMSD to the design model; binder-only RMSD 2.0 Å all-atom and 1.0 Å backbone; 21 hydrogen bonds at the interface, of which 16 were predicted in the design model. · Structural characterization of a KITbp:KIT…

  5. SuggestedTwo binders showed cellular activity: PDGFRɑbp-7LBF inhibited ligand-driven signaling, and FCRL5bp was internalized by FCRL5-expressing cells.

    PDGFRɑbp-7LBF blocked PDGF-AA signaling via Akt and Erk with nanomolar IC50s; FCRL5bp internalization in HeLa cells tracked FCRL5 expression and reached steady state in 18 hours. · Biological functionality of designed binders; Figure legends

Methods, data and reproducibility

Secondary structure and block-adjacency tensors specifying a binder beta-strand of length L paired with a chosen target edge strand condition RFdiffusion; backbones then pass ProteinMPNN, AlphaFold2 and Rosetta filtering, followed by yeast display, SPR, CD and X-ray crystallography.

  • In silico comparison of strand-conditioned vs hotspot-conditioned RFdiffusion across seven targets (ProteinMPNN sequences, AlphaFold2 and Rosetta scoring) · n = 500 scaffolds per approach; 10 sequences per scaffold
  • Target structures and models: KIT, NRP1, PDGFRɑ (PDB); ALK-2, ALK-3, FCRL5, PDGFRɑ (AlphaFold models) · PDB 2E9W, 2QQI, 7LBF; UniProt Q04771, P36894, Q96RD9, P16234
  • Yeast surface display design libraries for FCRL5, KIT and PDGFRɑ (two PDGFRɑ libraries) · Saccharomyces cerevisiae (EBY100 display strain) · n = 4846 (FCRL5), 1230 (KIT), 3780 and 391 (PDGFRɑ) designs
  • All-by-all SPR of purified binders against all targets, plus CD thermal melts · n = 7 targets
  • Co-crystal structure of KITbp with KIT domains 1-3 · n = 2.8 Å resolution · PDB 9H71
  • PDGFRɑ antagonism assay in CHO cells overexpressing PDGFRα, western blots for pPDGFRα, pAkt, pErk · Chinese hamster (CHO cells) · n = one traditional western blot repeat, plus automated Jess analysis of samples
  • FCRL5bp internalization assay in doxycycline-inducible FCRL5 HeLa cells · human (HeLa cells)
Methods & tools
RFdiffusion, ProteinMPNN, AlphaFold 2, Rosetta ΔΔG, Yeast surface display, SPR (Biacore 8K), Circular dichroism, HBPLUS, Foldseek, Phaser, PHENIX, STARANISO
Reagents & environment
Display strain S. cerevisiae EBY100 with pETcon3 vector; E. coli BL21(DE3) expression; HEK 293 for KIT123; CHO-PDGFRα and Flag-FCRL5 HeLa lines; pHAGE-PDGFRɑ (Addgene 116769); antibodies from Cell Signaling Technology (e.g., #24188, #9271, #9101).
Experimental conditions
Biacore 8K SPR; JASCO-1500 circular dichroism; P13 beamline (PETRA III, EMBL Hamburg) for X-ray data; Incucyte S3 live-cell imager; Biotechne Jess automated western system. No computing hardware or runtimes stated.
Reproducibility gaps
No code or design-pipeline repository URL given in the text; the structure is deposited under PDB code 9H71, but no data URLs are printed.
Limitations (5)
  • SPR Kd fits excluded the 10 μM data because of signal aberrations at that high binder concentration. · stated by the authors
  • KITbp:KIT diffraction data had severe pathologies (strong anisotropy and translational noncrystallographic symmetry), requiring elliptical truncation with STARANISO. · stated by the authors
  • No experimental structure exists for FCRL5, so its binders were designed against an AlphaFold 2 model, and its exact IgG binding site is unknown. · stated by the authors
  • Experimental sets were chosen with per-target cutoffs and, in some cases, extra partial diffusion and sequence sampling, so wet-lab success rates between conditioned and hotspot methods are not strictly matched. · noted in this summary
  • Crystallography covers only one complex (KITbp:KIT) and cell assays only FCRL5bp and PDGFRɑbp-7LBF, with a single traditional western blot repeat, so accuracy and function in other designs are less directly supported. · noted in this summary

Read this if you design protein binders to polar, beta-sheet-rich receptors and want conditioning details, per-target affinities and a crystallographic check of design accuracy.

Citations
33
FWCI
34.34
Field percentile
top 1%

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISBaker '24Jiang '25Britton '26Cao '21Kim '20Glögl '24
74% semantic matchBiochemistry
David A. Baker, Wei Yang, Derrick R. Hicks+25 more·bioRxiv·May 1, 2024· 11 citations