All chemicals were analytical grade. Recombinant human LCN2 was produced and purified by GenScript. The MMP-9 protein used in SPR studies was acquired from antibodies.com (code A331071-50). Twin-Strep–tag magnetic beads, Buffer W and Buffer BXT were from IBA LifeSciences. Dynamic-light-scattering (DLS) and nano-DSF assays were carried out in phosphate-buffered saline (PBS; 20 mM sodium-phosphate, 150 mM NaCl, pH 7.4). All other solutions are described below.

The generation of miniproteins to bind LCN2 was carried out using the procedure described by Baker and co-workers (20) (https://github.com/RosettaCommons/RFdiffusion). The crystal structure of LCN2 (PDB ID: 1DFV) was selected as template structure (21). Following the protocol author’s suggestion, we generated 10,000 backbone scaffolds using RF diffusion (20) and subsequently designed 10,000 sequences with ProteinMPNN (22) to create LCN2 binders. Finally, the ability of the generated sequences to correctly fold and to bind the target was evaluated using AlphaFold2 (AF2) (23).

For each model we recorded the mean pLDDT and the interface predicted aligned-error (pAE_interface). Designs with pAE_interface<10 were retained, yielding 1,184 candidates.

Next, to select the five most promising binders, we estimated the LCN2-binder affinity with PRODIGY, a contact-based predictor of protein-protein affinity (24). Finally, the five proteins reporting the most negative ΔG were expressed and purified.

To identify the molecular determinants of the MinP-2/LCN2 interaction, we analyzed the AF2-predicted complex structure using the DrugScorePPI server (25).

The DNA sequences encoding the five select miniproteins were synthesized and cloned by GenScript into the pET30b plasmid, incorporating a Twin-Strep tag for purification. E. coli BL21 (DE3) pLysS cells were transformed with the plasmids and plated onto five separate LB-agar plates. Following overnight incubation, colonies that successfully incorporated the plasmids were selected.

Before starting the protein production and purification, we assessed miniprotein expression in E.coli BL21 (DE3) pLysS cells analyzing induced and not induced cultures by Western-Blot (WB).

Samples (20 µl) from both IPTG-induced and non-induced cultures were mixed with 20 µl of sample buffer containing Tris-Glycine SDS and a sample reducing agent. From these prepared stocks, 20 µl were loaded onto the gel. Following electrophoresis, the SDS-PAGE gel was transferred into a membrane. After the transfer, the membrane was activated with methanol and subsequently washed with Tris-buffered saline containing 0. 1% Tween^® 20 (TBST). Following the washing step, blocking was performed with TBST and milk to prevent nonspecific binding. Subsequently, another washing step was conducted, and the primary antibody (IBA Lifesciences 2-1507-001; Anti-Strep) was added.

The next day, the primary antibody was removed, and the secondary antibody (anti-mouse IgG1 conjugated with horseradish peroxidase (HRP)) was applied. After the incubation step, another washings step was performed, and reagents were added to develop the Western Blot.

Following positive WB results, a bacterial preculture was prepared by inoculating a single colony into 7 mL of Luria-Bertani (LB) medium supplemented with chloramphenicol and kanamycin, and incubated at 37 °C to promote bacterial growth.

Subsequently, 1 mL of the preculture was inoculated into 200 mL of LB medium and incubated at 37 °C until the culture reached an OD₆₀₀ of 0.6. Protein expression was then induced by adding IPTG to a final concentration of 1 mM. After overnight incubation, bacterial cells were resuspended in a PBS added with two tablets of protease inhibitor (cOmplete™, Mini, EDTA-free Protease Inhibitor Cocktail) and lysed by sonication. The lysate was then filtered through a 0.22 μm membrane to remove debris before purification.

Strep-tag purification was performed using magnetic beads (IBA LifeSciences). Beads were washed with 1x Buffer W, followed by an incubation step to allow protein binding. After additional washing, elution was carried out using 1X Buffer BXT, yielding purified miniproteins in the eluate. Residual biotin in the elution buffer was removed by dialysis. Purity was verified by SDS-PAGE.

The size distribution and thermal stability of all miniproteins were analyzed using a Prometheus Panta (NanoTemper). Measurements were performed in triplicate, with miniproteins directly loaded into capillaries at concentrations of 0.32, 0.22, 0.18, 0.10, and 0.2 mg/mL for MinP-1 to MinP-5, respectively.

Interactions between the designed miniproteins and LCN2 were analyzed using biolayer interferometry (BLI). LCN2 was immobilized by His-tag on BLI sensors Octet^® Ni-NTA (NTA) Biosensors using a concentration of 5μg/ml. All the experiments were run using a Octet^® R8 instrument. All miniproteins were tested at a concentration of 4 μM. For MinP-2, which showed clear binding, titration experiments were performed at 1000, 500, 250, 125, 62.5, 31.2, and 15.6 nM. The signal from a reference sensor without immobilized protein was subtracted to remove background noise and aspecific binding. Association and dissociation phases were 300 s each, and data were fitted globally with a 1:1 and 2:1 binding models in the Octet Analysis software.

SPR experiments were performed using a Biacore 8K system (Cytiva, Marlborough, MA, USA). MinP-2 dissolved in 10 mM sodium acetate (pH 4.5) at a concentration of 2 μM was immobilized via amine coupling onto a CM5 sensor chip (GE Healthcare, Chicago, IL, USA). MMP-9 was immobilized using the same procedure using protein concentration of 300 nM.

Five concentrations of LCN2 (7.8, 15.6, 31.2, 62.5, and 125 nM) were injected over the immobilized MinP-2 at a flow rate of 30 μL/min. The contact time for each injection was 120 s, while the dissociation phase was run for 600 s. All the experiments were performed at 25 °C. Following blank subtraction, the binding data were fitted and analyzed using Biacore Insights Evaluation Software v. 5.0.18. In the case of the MMP-9/LCN2 interaction a single concentration of 200 nM LCN2 alone or mixed with an equimolar concentration of MinP-2 was injected for 45s followed by 60s of dissociation.

In the initial phase of the project, we applied the computational pipeline originally proposed by Baker and colleagues (20). Starting from 10,000 backbone scaffolds generated with RFdiffusion, a single ProteinMPNN sequence was designed for each scaffold. AlphaFold2 rescoring retained 1,184 binders with a pAE_interaction score below 10 - a threshold shown to enrich for functional hits (26) ( Figure 2A ) 1S.

However, expressing and purifying such a large number of proteins is impractical. To overcome this limitation, we applied a consensus scoring approach to refine the selection. Specifically, all 1,184 candidates were re-ranked with the contact-based affinity predictor PRODIGY (24), and the five miniproteins with the most negative predicted ΔG were selected for expression ( Table 1 , Supplementary Figure S3 ).

Given that the proteins were designed by AI, before large-scale production we assessed the expression of the five constructs in E. coli by WB ( Supplementary Figure S1 ). This preliminary analysis confirmed that all proteins were expressed, although MinP-2 and MinP-3 showed lower expression levels. Consequently, we proceeded with large-scale expression and purification of all five constructs. Purification was performed using magnetic beads functionalized with a modified streptavidin (Strep-Tactin^®XT, IBA Lifesciences), yielding highly pure proteins in sufficient quantities for subsequent biophysical characterization ( Supplementary Figure S2 ).

Before evaluating LCN2 binding, all purified constructs were characterized for size distribution via dynamic light scattering (DLS) and thermal stability via nanoDSF. DLS analysis revealed that all constructs - except MinP-4 - were monodisperse with an apparent hydrodynamic radius of approximately 2 nm, consistent with their predicted molecular weight and structural models ( Figures 3A–E , 4A–E ). In contrast, MinP-4 showed a polydispersity index (PDI) of 0.38, indicative of aggregation. Thermal stability analysis by nanoDSF showed that only MinP-5 displayed a clear melting temperature (Tm at 46.8 °C). This is notably lower than the high thermal stability (Tm > 90 °C) reported for some AI-designed miniprotein binders by Baker and colleagues (16).

Preliminary binding assays at 4 µM were performed to evaluate the interaction of the five miniproteins with LCN2. Among them, MinP-2 emerged as the most promising binder ( Figure 5A ), whereas the other constructs displayed only weak or negligible binding.

Therefore, to fully characterized affinity and kinetics of the LCN2-MinP-2 interaction we performed a titration across seven concentrations ( Figure 5B ). This experiment showed that MinP2 strongly binds LCN2, with two dissociation constants (Kd1 = 0.7 and Kd2 = 35 nM) obtained using a 2:1 heterogeneous model. The requirement for a 2:1 model may stem from surface crowding on the sensor, partially blocking MinP-2 binding sites and resulting in a secondary, lower-affinity interaction.

To verify if the immobilization strategy used in the BLI experiment could be the source of the 2:1 binding, we performed surface plasmon resonance (SPR) experiments ( Figure 5C ). In this case, confirming our hypothesis about the influence of the immobilization strategy on the results, the data were best fitted with a 1:1 model, yielding a Kd for the LCN2/MinP-2 complex of 4.2 nM.

To better understand the molecular determinants of the LCN2/MinP-2 interaction, we performed a computational alanine scanning analysis on the structure AlphaFold2-predicted complex using the DrugScorePPI server. The results ( Supplementary Tables S1-2 ) and visual inspection of the interface ( Figure 2B ) reveled that binding is determined by multiple intermolecular interactions, notably: (i) backbone hydrogen bonds between LCN2 residues Cys96, Asp97, and Tyr98 and the MinP-2 residues Ile38 and Val40 forming a short antiparallel β-augmentation interface; (ii) additional hydrogen bonds between LCN2 Lys95, Asn116, and Tyr120 with MinP-2 residues Thr4, Glu31, and Gly33; and (iii) a salt bridge between Arg37 of MinP-2 and Asp97 of LCN2. Taken together, these interactions rationalize the observed low-nanomolar affinity.

To evaluate whether MinP-2 can prevent the interaction between MMP-9 and LCN2, we performed an SPR experiment in which LCN2 - alone or in presence of an equimolar (200 nM) concentration of MinP-2 - was injected over immobilized MMP-9. This experiment ( Figure 5D ) demonstrated that MinP-2 effectively LCN2/MMp-9 interaction.