Targeting adhesins that sit outside the bacterial cell
The paper asks whether protein design can create new antimicrobials by neutralizing bacterial adhesins, the surface proteins that let bacteria recognize and hold onto host tissue. The authors chose adhesins because they sit at the tip of bacterial pili, well beyond the protective lipopolysaccharide shell, so a designed protein can physically reach them. Two targets were used: FimH from uropathogenic Escherichia coli, and the Abp1D and Abp2D adhesins from Acinetobacter baumannii strain ACICU.
The two targets map onto two clinical settings. FimH mediates uncomplicated urinary tract infection, while the Abp adhesins let A. baumannii bind fibrinogen deposited on implanted catheters, a step the paper describes as critical for establishing catheter-associated UTI. That split matters for reading every result that follows: the FimH and Abp programs are separate design campaigns with separate assays, not one molecule acting on both pathogens.
Hotspot-guided design produced binders aimed at the binding pocket
The design pipeline started from crystal structures of the adhesins and specified "hotspot" residues corresponding to the substrate-binding pocket. Using RFdiffusion with hotspot conditioning, the team generated roughly 10,000 backbone designs per target, assigned sequences with ProteinMPNN, filtered with AlphaFold2 Initial Guess predictions, and iteratively resampled with partial diffusion. Final designs had to pass interface and monomer-quality thresholds, including predicted binding energy and shape-complementarity scores.
For the Abp campaign, the authors deliberately designed against crystal structures showing the flexible anterior binding loop in its open, low-affinity conformation, so the putative pocket was exposed. This is a design choice with a mechanistic consequence: the resulting binder A7 locks that loop open, holding the adhesin in a low-affinity state rather than mimicking a natural high-affinity ligand. The paper reports that the exact structural interactions between fibrinogen and these adhesins remain unknown, so the pocket was inferred from prior mutagenesis work rather than from a solved ligand complex.
F7 binds FimH tightly and tolerates sequence variation
The FimH binder F7 bound the low-affinity FimH state with a dissociation constant of 119 nM and the high-affinity state with 713 nM, measured by surface plasmon resonance. It was thermostable up to 75 degrees Celsius by circular dichroism, and a co-crystal structure matched the design model with a C-alpha RMSD of 0.69 angstroms. The authors describe F7 as converting the high-affinity state toward the low-affinity state, which is the proposed neutralization mechanism.
Specificity was tested against the other program's targets: F7 showed insignificant binding to Abp1D and Abp2D, and A7 showed undetectable binding to FimH. F7 also inhibited hemagglutination by clinical E. coli strains from clonal groups ST73, ST95, ST69, ST1193 and ST131-H30, and by a strain expressing Klebsiella pneumoniae FimH, which shares roughly 85% sequence identity with E. coli FimH. That cross-strain activity is a hemagglutination readout, not a binding-affinity measurement, so it supports breadth of inhibition rather than quantified affinity against each variant.
Blocking adhesion in bladder cells and biofilms
In a T24 bladder cell adhesion assay, F7-treated wells had a mean of 2.85 bacteria per T24 cell, compared with 17.26 for buffer-treated cells (p = 0.0012) and 29.79 for the control minibinder A4 (p < 0.0001). F7 was not significantly different from the antibody mAb926, which averaged 2.0 bacteria per cell. In inhibition ELISA, F7 had an IC50 of 1.7 micromolar for detachment and mAb926 reached 9 nM, so the antibody was substantially more potent in that particular assay even though the cell-adhesion outcome was comparable.
For the Abp program, A7 inhibited A. baumannii ACICU binding to fibrinogen-coated plates with an IC50 of 2.6 nM and displaced already-adherent bacteria with an IC50 of 40.0 nM. At 1 micromolar, A7 prevented ACICU biofilm formation (p < 0.0001) and almost completely dispersed preformed biofilms after 2 hours (p = 0.0021), resembling the phenotype of an ACICU strain with both abp1 and abp2 deleted. On fibrinogen-coated silicone catheters, 100 nM A7 significantly reduced ACICU attachment and dispersed bound bacteria relative to a BSA control.
Lower bacterial titers in two mouse infection models
In a murine UPEC UTI model, bladder titers 48 hours post-infection were significantly higher in buffer-treated mice (n = 14) than in F7-treated mice (n = 10), with p = 0.0089 by Mann-Whitney U test. In the CAUTI model, mice infected with 10^8 ACICU cells were assessed 3 hours post-infection: A7 treatment (n = 12) lowered bladder titers relative to buffer (n = 9, p = 0.0035) and lowered catheter titers (p = 0.0006). The abp1/abp2 double-deletion strain also reduced bladder titers relative to wild-type ACICU treated with buffer (p = 0.0453), linking the effect to the adhesin target.
These are short-horizon readouts in mice, and the two models differ in timing and endpoint: 48 hours for bladder titers in the UPEC model versus 3 hours for bladder and catheter titers in the CAUTI model. The paper does not report a head-to-head in vivo comparison of the miniproteins against mAb926 or mannoside drugs; the authors list that comparison as future work.
Two routes to higher potency, with different mechanisms
The paper tested two optimization routes. Resampling sequences on the F7 backbone produced C8, with about 10-fold higher FimH affinity (low-affinity state Kd = 15.0 nM; high-affinity state Kd = 243 nM) and more than 30-fold greater inhibitory activity in ELISA (IC50 = 37 nM). Separately, fusing A7 to de novo designed homo-oligomeric domains created multivalent binders; oligomer C11 reached picomolar Kd values for both Abp1D and Abp2D and an inhibition ELISA IC50 of 0.36 nM, versus A7 in the same assay.
The distinction is worth keeping: C8 improves the intrinsic affinity of a single binding site, while C11 gains potency through avidity from multiple displayed copies. The paper frames both as generally applicable strategies for improving in vivo efficacy, but the reported gains are in binding and ELISA assays, not in the mouse models.
Resistance, off-target binding and dosing remain open
The authors argue that targeting the substrate-binding pocket should make rapid resistance unlikely, because mutations that disrupt binder binding may also disrupt adhesion. That is a hypothesis, not a measured outcome; the paper explicitly calls for future studies on whether resistance can be evolved and how the minibinders affect the broader microbiome. Similarly, specificity was checked against the other adhesin in the study, but off-target binding to host proteins was not assessed and is named as future work.
Pharmacokinetics are also unresolved for these specific molecules. The paper notes that some intraperitoneally delivered miniproteins clear from serum with half-lives under 15 minutes, but states that the half-lives of these miniproteins in the kidney are unknown and will be determined in future work. Route of administration, bioavailability and cross-strain reactivity are listed as requirements for clinical development, with intravesical instillation through the catheter proposed as one delivery option.
The adhesin target, not the scaffold, drives the case
The most decision-relevant comparison in this paper is between the two adhesin programs and their assay contexts. A7 is the stronger single molecule on the metrics reported: 2.6 nM inhibition IC50 against ACICU, 40.0 nM detachment IC50, biofilm prevention and dispersal at 1 micromolar, and significant reductions in both bladder and catheter titers in the CAUTI model. F7 is the more thoroughly cross-validated binder across clinical E. coli strains, but its ELISA IC50 of 1.7 micromolar trails mAb926's 9 nM, and its in vivo evidence is a single bladder titer comparison at 48 hours.
For readers weighing whether designed miniproteins could substitute for antibodies or mannosides, the paper's own framing sets the next threshold: in vivo head-to-head comparisons against mAb926 or mannoside drugs, plus pharmacokinetic characterization in the kidney. Until those exist, the supported claim is narrower and still notable — de novo design can hit a chosen conformational state of a bacterial adhesin, and that binding translates into reduced adhesion, biofilm disruption and lower bacterial burden in two mouse UTI settings.
Sources used in this answer
De Novo Design of Miniprotein Inhibitors of Bacterial Adhesins
De novo designed miniproteins targeting FimH and Abp adhesins blocked host-receptor recognition, disrupted biofilms, and reduced bacterial titers in murine uncomplicated and catheter-associated UTI models.