De novo design of dual-target mini-binders simultaneously neutralizing Streptococcus equi subspecies zooepidemicus M-like protein and host TNFR1 confers complete protection against lethal infection Authors: Ke Ming, Yang Hu, Banbin Xing, Meiting Chen, Yuhua Chen, Xiujuan Chen, Miaomiao Geng, Zhiyong Wu, Yanxia Liu, Lin Wei, Jun Weng, Zigong Wei Published: 2026 ABSTRACT Computationally designed mini-binders have shown great promise, yet current strategies primarily yield molecules targeting a single target. For complex infections, simultaneously engaging multiple pathogenic pathways may offer superior efficacy. Streptococcus equi subspecies zooepidemicus (SEZ) infection, characterized by a highly reproducible and rapid lethal phenotype in mice, serves as an ideal model to test this hypothesis. Based on our previous success with SzM- and TNFR1-targeting mini-binders, which conferred partial protection alone and enhanced survival in combination, we developed a RFdiffusion-based platform to de novo design dual-target binders. Two distinct methods, Superimposition-Optimization (SIO) and Fusion-Diffusion (FD), were developed to explore various configurations of the designed binders. All binders exhibited micromolar affinities to both targets. Functional validation confirmed their abilities to inhibit SEZ adhesion to host cells and block TNFα-TNFR1 signaling. SIO_b5 and SIO_b40 conferred complete protection against lethally infected mice, with a significant reduction in bacterial loads in organs and a suppression of serum TNFα. This work established an effective RFdiffusion-based strategy for designing dual-target mini-binders, providing a novel therapeutic strategy for combating SEZ infection and a generalizable platform for developing multi-target therapeutics against diverse complex diseases. Introduction Computationally de novo designed specific mini-binders with enhanced stability and smaller size could have advantages over conventional monoclonal antibodies (mAbs), achieving diverse biological functions [[1], [2], [3], [4], [5], [6]]. They also have been demonstrated to possess high thermal stability and low immunogenicity [1]. These features circumvent the high production costs and cold-chain dependence associated with mAbs, facilitating broader application. The computational approach has demonstrated considerable potential in treating bacterial infections [[7], [8], [9]], positioning it as a promising alternative to conventional antibiotics. Current de novo design strategies allow for the generation of high-affinity binders against virtually any target of interest [10,11]. However, these efforts have predominantly yielded binders capable of engaging only a single target. For complex infections where pathogenesis involves multiple pathways, a therapeutic strategy capable of simultaneous multi-target engagement may lead to significantly superior efficacy. The validity of this multi-targeting concept is well established by the success of bispecific antibodies and the development of other forms of dual-function antibodies [[12], [13], [14]]. Inspired by these precedents, we sought to develop dual-target mini-binders that overcome the limitations of conventional antibody-based formats. Streptococcus equi subspecies zooepidemicus (SEZ) infection represents an ideal model for validating the dual-target binders, both due to its significant clinical and economic impact and the availability of a well-established murine model that exhibits highly reproducible and rapid lethality, with consistent 100% mortality within 72∼96 h post-infection [[15], [16], [17], [18]]. SEZ is a well-known zoonotic pathogen with a wide host range, including equine, swine, canines, and human [16,19,20]. The infection manifests as diverse severe inflammatory conditions such as sepsis, meningitis, endocarditis, and arthritis [21,22], posing substantial challenges to both public health and the economy. Although human infections are exceedingly rare, they are often fatal [23]. Close contact with infected animals or their products is a major risk factor for human infection with SEZ, as illustrated by an outbreak in Spain linked to inadequately pasteurized dairy [24]. Notably, humans can also serve as carriers, potentially transmitting SEZ back to animal populations [25]. In the agricultural sector, SEZ outbreaks in swine industry have resulted in mortality rates as high as 30-50% within weeks post-infection in North America [26,27]. Similarly, in China, SEZ remains one of the principal pathogens for swine streptococcal infection [28], with sporadic outbreaks still occurring to this day [29]. The pathogenicity of SEZ is largely attributed to the massive bacterial proliferation, as well as uncontrolled host inflammatory response and the resulting extensive tissue injury. The M-like protein (SzM) serves as a key determinant for the virulence of SEZ [17,30]. In its free form, M protein can trigger vascular leakage, tissue damage [31], and activation of the NLRP3 inflammasome [32,33]. SzM is essential for SEZ adhesion, immune evasion, and breaching the blood-brain barrier (BBB) [15,34]. Studies have shown that SzM promotes SEZ survival in blood by binding surface fibrinogen, thereby inhibiting phagocytosis. The bacterial virulence is markedly diminished by szm gene deletion [34]. MAbs directed against SzM have shown remarkable protection against SEZ infection [16]. In our previous study, we designed mini-binders targeting SzM and achieved considerable protection against lethally infected mice [17]. In addition to SzM, other toxin factors of SEZ such as superantigens are also responsible for the excessive immune system activation and host damage. The robust inflammatory response induced by Streptococcus is characterized by elevated production of pro-inflammatory cytokines, particularly TNFα [35,36]. TNFα plays a central role in initiating and sustaining inflammatory cascades [37]. It is considered as the first appeared cytokine and other cytokines such as IL-1β and IL-6 rely on the prior release of TNF [38,39]. Upon pathogenic infections, tissue damage, or stress, immune cells such as macrophages and T cells are activated to secrete abundant TNFα [40]. TNFα signals through two receptors, TNFR1 and TNFR2, with TNFR1 primarily responsible for pro-inflammatory effects via NF-κB activation, leading to further cytokine production [41]. Persistent activation of the TNFα/TNFR1 signal can lead to systemic inflammatory response syndrome (SIRS), multiple organ dysfunction, and septic shock [42,43], contributing to bacterial infection-associated mortality independent of bacterial burden [44]. This understanding of SEZ pathogenesis, combined with the availability of a highly reproducible mice model, provided an ideal platform for our therapeutic strategy. We built upon our previous success in developing specific binders against SzM (SzM_binder3) and TNFR1 (TNFR1_s1b2) [17,45]. The demonstrated efficacy against SzM or TNF signaling provides a strong rationale for engineering dual-target binders capable of simultaneously targeting bacterial adhesion and inflammatory response. In this study, we proposed a strategy for designing such dual-target binders based on existing single-target binders. We introduced two novel methods, Superimposition-optimization (SIO) and FusionDiffusion (FD), and reported the development of binders that concurrently target SzM and TNFR1. Our results indicated that these dual-target binders can synergistically reduce bacterial load and attenuate TNFα-mediated inflammatory responses, significantly improving survival in lethal SEZ infection mouse model. This work highlighted the potential of dual-target binders as novel therapeutic against SEZ and provided a framework for the design of other multi-functional binders. Materials and methods Ethic statement All animal experiments reported herein were approved by the Animal Ethic and Welfare Committee of Hubei University (approval number: 20260021). Cells and bacterial strains HEK-293T cells (CRL-3216), L929 cells (CCL-1) and HEp-2 cells (CCL-23) were originally obtained from ATCC. The cells were cultured at 37 °C in a humidified environment with 5% CO2 in DMEM medium supplemented with 1% penicillin-streptomycin and 10% fetal bovine serum (FBS). The SEZ strain ATCC 35246 was purchased from American National Standard Preservation Center (ATCC) and maintained at 37 °C on tryptone soya agar (TSA) or tryptone soya broth (TSB) (Difco Laboratories, Detroit, USA) supplemented with 10% bovine sera under aerobic conditions. Mice Male BALB/c mice were purchased from Laboratory Animal Center of Huazhong Agricultural University. Male mice were used exclusively to avoid sex-related hormonal influences and ensure experimental consistency. Mice were kept under even temperature (22 ± 2 °C) and 12 h light/dark cycle (7 a.m. to 7 p.m.) in specific-pathogen-free cages with free access to water and food. Antibodies For immunoblotting: NF-kB p65/RelA Rabbit mAb (A19653, ABclonal Technology, China), Phospho-NF-kB p65/RelA-S536 Rabbit mAb (AP1294, ABclonal Technology, China), β-Actin Rabbit mAb (AC026, ABclonal Technology, China), HRP-conjugated Goat anti-Rabbit IgG (H + L) (AS014, ABclonal Technology, China), Cy3 conjugated Goat Anti-Rabbit IgG (H + L) (GB21303, Servicebio Technology, China). For flow cytometry: CD45 Monoclonal Antibody (30-F11), FITC (11-0451, Thermo Fisher Scientific, USA), GR-1 Monoclonal Antibody (RB6-8C5), APC (17-5931, Thermo Fisher Scientific, USA). RFdiffusion-based design of the dual-target mini-binders Dual-target mini-binders were designed based on two developed functional binders and were designed using two different strategies in this study, Superimposition-optimization (SIO) method for triple-helix binders and FusionDiffusion (FD) method for penta-helix binders, respectively. For SIO method, a total of eight steps were included. Step 1, calculate the binding free energy of the developed binder (TNFR1_s1b2 and SzM_binder3) to the target by molecular dynamics (MD) simulation (GROMACS 2024.5) through using AMBER14SB_parmbsc1 force field and TIP3P water model [46,47]. The binding free energy (ΔGbind) between the binder and its target was calculated using the gmx_MMPBSA tool on stable trajectory segments [48,49]. Step 2, helix bundle extraction and reassembly. Individual helices were extracted from the original helix-bundles and were ranked based on the calculated ΔG from Step 1 and key residue contributions. The top two helices from one binder (SzM_binder3) and the top helix from the other binder (TNFR1_s1b2) were selected. Their relative orientations were manually adjusted using PyMol to position the binding interfaces towards the surface of the nascent target proteins scaffolds. Step 3, linker design between helices using RFdiffusion [11]. Linker lengths were constrained to 4-7 amino acids, and 100 candidate structures were generated. Visual inspection of the 100 RFdiffusion-generated linker structures was performed in PyMol with the following criteria: (i) linker backbone continuity and junction compatibility (the linker backbone must exhibit no breaks or discontinuities, and its N- and C-termini must seamlessly connect to the adjacent helix ends without gaps, steric overlaps, or backbone strain, ensuring the 4-7 amino acid linker appropriately bridges the inter-helical distance); (ii) Ramachandran-allowed backbone dihedrals (all linker residues must fall within favored or allowed regions, verified using Discovery Studio); and (iii) absence of steric clashes (no van der Waals overlap greater than 0.5 Å was allowed between the linker and adjacent helix backbones). Step 4, sequence design for linkers and triple-helix bundles using ProteinMPNN. The amino acid sequences for the linkers and the residues within the core of the triple-helix bundle were re-designed using ProteinMPNN to optimize packing and minimize steric clashes [50]. Step 5, structure prediction and filtering. The designed sequences from Step 4 were folded in silico using ColabFold [51]. Generated models were filtered based on pLDDT scores and the Cα root-mean-square deviation (RMSD) between the designed model and the original model. Step 6, interface sequence optimization using ProteinMPNN-FastRelax. A ternary complex model (SzM-binder-TNFR1) was constructed in PyMol, and the sequence optimization on the binder interfaces was performed using the ProteinMPNN and FastRelax [52]. Step 7, screening for dual target binding using ColabFold. ColabFold was used to predict the structures of binder-SzM and binder-TNFR1 complexes. Complexes were ranked based on predicted pLDDT and ipTM scores. High-scoring complex models were visually inspected to confirm (i) binding epitope preservation (the interface residues of each helix must maintain the same contact pattern as in the original parent binders, SzM_binder3 and TNFR1_s1b2); (ii) helix orientation fidelity (helix 1 must exclusively contact SzM, helix 3 must exclusively contact TNFR1, and helix 2 must engage both targets without steric occlusion); and (iii) interface quality (absence of inter-molecular clashes at both binding interfaces). Step 8, MD simulation and binding affinity calculations. Promising dual-target binder candidates were subjected to MD simulation and binding free energy calculation. The calculated ΔG values for binder-SzM and binder-TNFR1 complexes were used to select final candidates for experimental validation. For FD method, the non-binding helix at C-terminal of TNFR1-s1b2 served as a central structural scaffold. The binding helices from both parent binders (both helices from SzM_binder3 and the two binding helices from TNFR1-S1B2) were connected to this central scaffold, resulting in a five-helix bundle structure. The design workflow included five steps. Step 1, structural assembly of helices using PyMol. The relative positions of the two parent binders were manually adjusted using PyMol. The manual adjustment was guided by structural rationale rather than candidate screening. The non-binding C-terminal helix (scaffold helix) from TNFR1-S1B2 was positioned centrally. The binding helices from both binders were positioned on either side of this central helix, forming the preliminary structure of the five-helix bundle. Care was taken to orient the binding interfaces of these four functional helices towards the solvent-exposed surface. Step 2, linker design using RFdiffusion. Linkers connecting the functional helices to the central scaffold helix were designed using the RFdiffusion [11]. A total of 1000 linker backbone structures were sampled to explore diverse conformational possibilities. Step 3, linker sequence design using ProteinMPNN. Three sequences were sampled per linker backbone. Amino acid sequences for the designed linkers (from Step 2) were generated using ProteinMPNN [50]. The HyperMPNN model, specifically trained on thermostable proteins, was utilized for sequence design to enhance the potential stability of the linkers [53]. Step 4, structure prediction and screening. The designed penta-helix bundle sequences were folded in silico using ColabFold [51]. Generated models were rigorously screened based on pLDDT score and ipTM score. Models with high pLDDT were prioritized, indicating confident backbone atom prediction. High ipTM scores indicate high confidence in the predicted interface formation for each respective target. Visual inspection criteria were applied to ensure the final penta-helix bundles maintained the intended topology, specifically an intact five-helix bundle without steric clashes and with correct helix-helix packing geometry. Step 5, MD simulation and binding affinity calculations as described in Step 8 of SIO method. Protein expression For the expression of the designed binders, the gene sequences of the binders were synthesized based on the amino acids of the protein. Then the genes were cloned into pET-22b vectors (69744-3, Addgene) by In-Fusion cloning. The vectors were then transformed into E. coli BL21 competent cells and the protein expression was induced by supplementing IPTG and culturing at 18 °C for 20 h. The harvested bacterial cells were sonicated, and the expressed protein was collected. Proteins were purified by Ni-NTA His column, followed by size-exclusion chromatography (SEC) using fast protein liquid chromatography (FPLC) system equipped with a Superdex 75 Increase 10/300 column (Cytiva, Sweden). The Avi-tag labeled SzM and TNFR1 were prepared via E. coli expression system. Briefly, the genome of SEZ was extracted by using Rapid Bacterial Genomic DNA Isolation Kit (B518225, Sangon Biotech, China), and the gene of SzM protein was amplified using specific primers. Then the gene was cloned into pET-22b vectors (69744-3, Addgene) by In-Fusion cloning, and the protein was expressed and purified as described above. Differential scanning fluorimetry The thermal stability of the protein was assessed using differential scanning fluorimetry (DSF) with SYPRO Orange dye (S5692, Sigma-Aldrich, USA). The protein was diluted in 50 mM Tris buffer (pH 8.0) to desired concentrations. A total reaction volume of 20 μL was assembled in a 100 μL PCR tube strip. The final protein concentrations in the assay were 40 μM. Melting curve analysis was performed using CFX Connect real-time PCR system (Bio-Rad, USA) with the following program: temperature ramp from 35.0 °C to 95.0 °C with increments of 0.2 °C per step, holding for 10 s at each temperature followed by a plate read. Data were processed using the CFX Maestro™ software (Bio-Rad, USA). Mouse infection models Six-week-old specific-pathogen-free (SPF) mice were randomly divided into different groups. To evaluate the impact of the drug on animal survival, a mouse model of SEZ infection was established via intraperitoneal injection of the bacteria (2 × 107 CFU per mouse). Different binders were administered concurrently, with a control group receiving PBS.Then the mice were treated once daily by intraperitoneal injection. Mortality was recorded daily across all groups. Once the infected mice died, major organs including heart, liver, spleen, lung, kidney and brain, were collected and homogenized in sterile PBS for bacterial colony counting. The experiment was terminated on day 10, and the final survival rates were calculated for each group. Surviving mice were euthanized, and bacterial loads in their organs were quantified. To assess the effects of the drug on bacterial burden and histopathological changes in various organs, a separate group of infected mice received the binder treatments immediately after model establishment and subsequently every 24 h. At 48 h post-infection, the animals were humanely euthanized. Different organs were harvested. One portion of each tissue was fixed for subsequent hematoxylin and eosin (H&E) staining, while another portion was homogenized for bacterial enumeration via plate counting. Serum concentrations of the binders The serum concentrations of the binders were measured using a fluorescence-based assay. Briefly, cysteine residue was added to the C-terminal of the binders for FITC conjugation. After expression in E. coli BL21 system, the binders were incubated with FITC at a 1:1 M ratio in a shaking incubator at 37 °C overnight. Unconjugated FITC was removed, and the conjugated binders were concentrated to 15 mg/mL through using ultra centrifugal filters (Sigma-Aldrich, USA). The fluorescence of the conjugate was confirmed by measuring absorption at excitation/emission wavelengths of 494 nm and 520 nm, respectively. The FITC-labeled binders were 2-fold diluted from 1 mg/mL using serum isolated from health mice, and a standard curve was constructed. Mice were administered the FITC-labeled binders via intraperitoneal injection at a dose of 10 mg/kg. Blood samples were collected at 0.25 h, 0.5 h, 1 h, 2 h and 4 h post-administration. Serum was separated, and 50 μL of supernatant was diluted with an equal volume of PBS. Fluorescence was measured in a plate reader, and the background fluorescence from untreated mouse serum was subtracted from all samples. The serum concentration of the binder was quantified based on the standard curve. BLI assays BLI Octet® RED96 system (ForteBio) was used to assess the binding affinities between binders and the recombinant SzM and TNFR1. The Avi-tag labeled SzM and TNFR1 protein were biotinylated modified using a biotin labeling kit (Beyotime, China) and then were loaded onto streptavidin-coated sensor tips in binding buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 3 mM EDTA, 0.05% Tween20 and 1% BSA) for 30 min. The analyte proteins (binders) were diluted in binding buffer from stock solution to indicated concentrations. The BLI sensor tips were dipped into buffer-only wells for baseline, then into binder solution wells for association, and then back into buffer-only wells for dissociation. The kinetics constants (kon, koff) were obtained by global fitting of the association and dissociation curves for each binder under a 1:1 binding model. The affinity constants (KD) of the binders were calculated using Octet Analysis Studio v.13.0.1.35. RNA isolation and qRT-PCR Total RNA from different tissues was extracted using TRIzol reagent (Invitrogen, USA) and then were reversely transcribed into complementary DNA using ABScript III RT Master Mix (RK20429, ABclonal Technology, China). qRT-PCR was performed on CFX Connect real-time PCR system (Bio-Rad, USA) using Universal SYBR Green Master Mix (RK21203, ABclonal Technology, China). The delta-delta Ct method was used to calculate the relative fold gene expression. Primer sequences used were listed in Table S1. Western blot Cells or tissue homogenate were lysed in RIPA lysis buffer (Beyotime, China) supplemented with protease inhibitor and phosphatase inhibitor cocktail A (Beyotime, China). The lysates were collected by centrifugation, and the protein was quantified via BCA protein assay kit (Biosharp, China). Thirty micrograms of protein were loaded and separated by SDS-PAGE and then transferred onto PVDF membrane (Merck Millipore, USA). Thereafter, the membranes were blocked by 5% bovine serum albumin (BSA, Sigma, USA) that dissolved in tris-buffered saline with Tween (TBST) and incubated in sequence with the primary antibodies and HRP-conjugated secondary antibodies. The protein bands were visualized using SuperKine™ West Femto Maximum Sensitivity Substrate (BMU102, Abbkine, China), and the images were obtained using the ECL detection system (Tanon 5200, Tanon, China). The band intensities were measured by ImageJ software (National Institutes of Health, USA). ELISA The serum TNFα concentrations were determined by using Mouse TNFα ELISA Kit (ADS-0719M1, AIDISHENG, China) according to the manufacturer's instructions. The absorbance at 450 nm was recorded using microplate reader. Cell adhesion assay The adherence of SEZ to HEp-2 cells was determined as previously described [17]. Briefly, bacteria culture was mixed with various binders at different concentrations at room temperature for 30 min. Then the mixture of bacteria (5 × 107 CFU) and binder were co-incubated with HEp-2 cell monolayer at 37 °C for 2 h. Cells co-incubated with bacteria and 20 μM of BSA were served as control. The non-adherent bacteria were removed, and the cells were lysed using 1 mL PBS containing 0.2% (v/v) Triton X-100. The solution was serially diluted and plated on TSA to count bacterial numbers after incubation at 37 °C. Cell viability assay The viabilities of L929 cells under different treatment conditions were evaluated using the CCK-8 cell counting kit (A311, Vazyme, China) in accordance with the manufacturer's instructions. The absorbance at 450 nm was measured with a microplate reader and relative viability of cells was calculated using the following formula: relative cell viability (%) = [(ODtreatment – ODmodel)/(ODcontrol – ODmodel)] × 100%, where “treatment” referred to cells were stimulated with human-TNFα (H8196, Sigma-Aldrich, USA) followed by exposure to different binders, “model” referred to cells were treated with TNFα alone without binders, “control” represented cells were treated with PBS only without both TNFα and binders. Luciferase reporter assays Luciferase reporter assay was performed to evaluate the activation of NFκB signaling. HEK293T cells were co-transfected with NF-κB luciferase reporter plasmid (pNFκB-Luc, Beyotime, China) and Renilla luciferase plasmid (pRL-SV40-C, Beyotime, China) by using Lipofectamine™ 3000 transfection reagent (L3000001, Invitrogen, USA). Reporter activities were measured by using the Dual Luciferase Reporter Gene Assay Kit (RG207, Beyotime, China). Relative luciferase activities were calculated as the ratio of firefly luciferase activity to Renilla luciferase activity. Flow cytometry analysis The collected cells were resuspended in Flow Cytometry Staining Buffer (00-4222-26, Invitrogen, USA) followed by incubation with antibodies (diluted 1:200) according to the manufacturer's instructions. The samples were submitted to Cytoflex-LX Flow Cytometer (Beckman), and the data were analyzed and visualized using FlowJo software (BD Bioscience). H&E staining The tissues were collected and immediately fixed using 4% Paraformaldehyde Fix Solution (P1110, Solarbio, China) for 24 h. Then the samples were embedded in paraffin. Sections of paraffin blocks with 5 μm of thickness were cut and placed on slides, which were subsequently dehydrated using ethanol and stained with hematoxylin and eosin (H&E). The slides were observed under optical microscope (Olympus, Japan). Immunofluorescence assay For p65 immunofluorescence, lung tissues were fixed in 4% paraformaldehyde, followed by permeabilized with 0.3% Triton X 100, blocked with 5% BSA, and incubated overnight at 4 °C with anti p65 antibody (1:200). After washing, sections were incubated with Cy3 conjugated secondary antibody (1:400) for 1 h at room temperature, counterstained with DAPI, and mounted. Fluorescence images were acquired using a confocal microscope (Zeiss LSM800, Germany). Statistical analysis Statistical analysis was performed using GraphPad Prism 10, all data were presented as the means with SEM. Data distribution was assessed using the Shapiro-Wilk normality test. Parametric tests were applied for normally distributed data; otherwise, non-parametric tests were used. Results and discussion Combination of anti-TNFα and anti-adhesion treatment further improve the lethal outcome of SEZ infection TNFα is a pivotal early mediator in the inflammatory cascade, whose dysregulation is also a critical driver of cytokine storms. Mortality in severe infections is largely attributable to cytokine storms resulting from an exaggerated host immune response, which ultimately leads to multiple organ failure. Case studies have reported that patients undergoing anti-TNFα antibody therapy for rheumatic diseases exhibit a lower risk of hospitalization following SARS-CoV-2 infection [54]. Similarly, the TNF inhibitor Etanercept has been shown to provide protection against lethal H1N1 influenza infection [55]. These findings suggest that host might benefit from anti-TNFα treatment in face of severe infections. Accordingly, we investigated whether anti-TNFα treatment could mitigate lethal SEZ infection of mice (Fig. 1a).Fig. 1The therapeutic effects of TNFR1_s1b2 on SEZ-infected mice. a) Diagram of the bacteria challenge and binder treatment process. b) The serum TNFα concentration 48 h post-infection and binders treatment (n = 5). Statistical significance was analyzed by one-way ANOVA in Graphpad Prism 10 software. ∗P < 0.05. c) The tissue bacterial loads at 24 h and 48 h post SEZ infection and treatment (n = 5). d) Survival rates of mice (n = 11). Mice lethally infected with SEZ were treated with either binder alone (10 mg/kg) or a combination (10 mg/kg + 10 mg/kg). e) Histologic analysis of the brain. At 48 h post infection and treatment, the mice were euthanized, and the brains were harvested for H&E staining. The images were representative of three biological replicates, with the scale bar indicating a length of 0.1 mm. All data were presented as means ± SEM. The statistical significances in b-c were analyzed by one-way ANOVA and that in e by log[rank] (Mantel–Cox) test. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001, ns, no significance. The therapeutic effects of TNFR1_s1b2 on SEZ-infected mice. a) Diagram of the bacteria challenge and binder treatment process. b) The serum TNFα concentration 48 h post-infection and binders treatment (n = 5). Statistical significance was analyzed by one-way ANOVA in Graphpad Prism 10 software. ∗P < 0.05. c) The tissue bacterial loads at 24 h and 48 h post SEZ infection and treatment (n = 5). d) Survival rates of mice (n = 11). Mice lethally infected with SEZ were treated with either binder alone (10 mg/kg) or a combination (10 mg/kg + 10 mg/kg). e) Histologic analysis of the brain. At 48 h post infection and treatment, the mice were euthanized, and the brains were harvested for H&E staining. The images were representative of three biological replicates, with the scale bar indicating a length of 0.1 mm. All data were presented as means ± SEM. The statistical significances in b-c were analyzed by one-way ANOVA and that in e by log[rank] (Mantel–Cox) test. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001, ∗∗∗∗P < 0.0001, ns, no significance. We first measured the expression profiles of several critical pro-inflammatory cytokines after SEZ infection. Results revealed substantial upregulation of both tnfa gene and il6 gene across all major tissues at 24 h post infection. The increase in tnfa expression reached several hundred-fold in the liver, kidney, and spleen (Fig. S1), indicating that SEZ infection rapidly triggers a robust and wide inflammatory response. Consistently, serum levels of TNFα post SEZ infection were also significantly elevated (Fig. S1). To evaluate whether anti-TNFα treatment confers protection against SEZ-infected mice and whether combining anti-TNFα with anti-SzM treatment further improves the protective effect, we introduced the previously reported TNFR1 mini-binder (TNFR1_s1b2) and SzM mini-binder (SzM_binder3) [17,45]. The results showed that at 48 h post-infection, both binders administered alone or in combination significantly reduced serum TNFα levels in SEZ-infected mice (Fig. 1b). Regarding bacterial load in heart, liver, spleen, lung, kidney, and brain, SzM_binder3 alone, which inhibits bacterial adhesion, significantly reduced colony counts in these tissues, and its combination with TNFR1_s1b2 achieved a similar effect. In contrast, TNFR1_s1b2 alone had no significant impact on bacterial load in these tissues, although a modest reduction was observed at 48 h post-infection (Fig. 1c). We speculate that this observed reduction might be linked to a moderate upregulation of TNFα signaling potentially enhancing the phagocytic function of phagocytes such as neutrophils [56], underscoring the complex dual role of TNF in anti-infective immunity. In terms of 10-day survival, TNFR1_s1b2 monotherapy significantly improved survival from 0% to 36%. SzM_binder3 monotherapy increased survival to 64% (Fig. 1d). The combination therapy significantly improved survival compared to PBS control (82% vs. 0%) and to TNFR1_s1b2 monotherapy (82% vs. 36%). Although the survival difference between combination and SzM_binder3 monotherapy (82% vs. 64%) did not reach statistical significance, the survival improvement is meaningful (Fig. 1d). Moreover, bacterial loads in the organs of the combination group were significantly lower than those in the TNFR1_s1b2 monotherapy group (Fig. 1c). At 48 h post-infection, brains were collected from the mice. Upon gross examination, the brain tissues exhibited redness with dilated and congested blood vessels. Histopathological analysis by H&E staining revealed cerebral hemorrhage, accompanied by extensive neutrophils infiltration, indicative of meningitis. In contrast, both macroscopic and histological analyses demonstrated marked amelioration of these pathological features in the brain tissues of mice from the treatment group (Fig. 1e). Beyond the brain, SEZ infection induced significant inflammatory lesions in the lungs and spleens, characterized by hemorrhage, neutrophil infiltration and a certain degree of cellular necrosis upon H&E staining. Consistent with its effects in the brain, TNFR1_s1b2 alone or a combination with SzM_binder3 substantially mitigated pathological injury in both lung and spleen tissues (Fig. S5). Above data indicated that anti-TNFα treatment can further improve the outcome of SEZ-infected mice. Design and characterization of dual-target binders against the SzM and TNFR1 Computationally designed mini-binders have emerged as a promising novel therapeutic modality with broad application prospects. This approach enables the generation of protein molecules with precise shape- and chemical-complementary interactions to specific epitopes on target proteins [7]. Their application in agriculture holds significant potential for enhancing production safety and reducing costs. We aimed to design dual-target binders capable of simultaneously targeting SzM protein of SEZ and TNFR1. Based on the previously reported binders that individually target SzM and TNFR1 (Fig. S3), we developed two distinct strategies to generate dual-target binders: the Superimposition-optimization (SIO) method and the FusionDiffusion (FD) method. The SIO method generates triple-helical binders, while the FD method yields a penta-helical structure. In the SIO method, the design process initiated with the deconstruction of our previously developed binders against SzM (SzM_binder3) and TNFR1 (TNFR1_s1b2). Helical bundles exhibiting strong interactions with their respective targets were selected and connected via linkers designed using RFdiffusion to form a triple-helical bundle. Subsequently, ProteinMPNN was employed to optimize the amino acid sequences, minimizing the internal potential energy of the bundle while refining the binding interface residues. Finally, ColabFold was utilized to screen multiple candidates and identify binders capable of engaging both SzM and TNFR1 simultaneously (Fig. 2a). A key innovation in this design lies in the strategic assignment of helix functions. Unlike typical three-helix bundle binders, where the third helix primarily provides structural stability, we engineered one helix to exhibit dual-specificity, enabling interaction with both SzM and TNFR1. The other two helices were specifically dedicated to binding their respective targets. This ingenious arrangement ensures that each target is engaged by two helices, thereby enhancing binding avidity, while the canonical three-helix scaffold is retained to maintain overall structural stability.Fig. 2Overview of the design of dual-target binders. a) Schematic of the design of triple-helix bundle dual-target binders using the SIO method. This method comprised eight key steps: (1) calculating binding free energy of initial binders via MD simulations; (2) selecting and reorienting top-performing helices based on the binding energy and residue contribution; (3) designing linkers using RFdiffusion; (4) optimizing linker and core sequences with ProteinMPNN; (5) predicting and filtering structures via ColabFold based on pLDDT and RMSD; (6) refining interface sequences using ProteinMPNN and FastRelax within a ternary complex; (7) screening dual-target binding capacity with ColabFold and inspecting binding specificity; (8) evaluating final candidates through MD simulations and binding free energy calculations for experimental selection. b) Schematic of the design of penta-helix binders using the FD method. The FD method constructed a five-helix bundle using the non-binding C-terminal helix of TNFR1-s1b2 as a central scaffold. Binding helices from both parent binders were attached to this scaffold. The procedure involves: (1) assembling the helix structure; (2) designing linkers via RFdiffusion; (3) optimizing linker sequences using ProteinMPNN; (4) predicting and screening bundle structures with ColabFold based on pLDDT and ipTM scores; (5) evaluating binding affinity and stability through MD simulations. Overview of the design of dual-target binders. a) Schematic of the design of triple-helix bundle dual-target binders using the SIO method. This method comprised eight key steps: (1) calculating binding free energy of initial binders via MD simulations; (2) selecting and reorienting top-performing helices based on the binding energy and residue contribution; (3) designing linkers using RFdiffusion; (4) optimizing linker and core sequences with ProteinMPNN; (5) predicting and filtering structures via ColabFold based on pLDDT and RMSD; (6) refining interface sequences using ProteinMPNN and FastRelax within a ternary complex; (7) screening dual-target binding capacity with ColabFold and inspecting binding specificity; (8) evaluating final candidates through MD simulations and binding free energy calculations for experimental selection. b) Schematic of the design of penta-helix binders using the FD method. The FD method constructed a five-helix bundle using the non-binding C-terminal helix of TNFR1-s1b2 as a central scaffold. Binding helices from both parent binders were attached to this scaffold. The procedure involves: (1) assembling the helix structure; (2) designing linkers via RFdiffusion; (3) optimizing linker sequences using ProteinMPNN; (4) predicting and screening bundle structures with ColabFold based on pLDDT and ipTM scores; (5) evaluating binding affinity and stability through MD simulations. To explore more possibilities, we also implemented the FD strategy, in which the α-helices located at the binding interfaces of SzM_binder3 and TNFR1_s1b2 were connected via an additional central α-helix. The general procedure resembled that of the SIO method: RFdiffusion was used for linker design, ProteinMPNN for linker sequence optimization, and ColabFold for structural prediction and in silico screening. This approach resulted in a dual-target binder exhibiting a “dumbbell” conformation, with the two target-binding sites situated at opposite ends of the molecule (Fig. 2b). We hypothesized that this design might increase the spatial separation between the two targets, potentially reducing steric interference caused by simultaneous binding. To assess the efficiency and robustness of our design pipelines, we tracked the number of candidates progressing through each screening stage (Table S2 and Table S3). For the SIO method, starting from 100 initial linker backbones, 47 passed visual inspection. After ProteinMPNN sampling (3 sequences per scaffold), 141 sequences entered ColabFold prediction, of which 39 met quality thresholds (pLDDT > 80, Cα RMSD < 2.0 Å). Dual-target binding screening (pLDDT > 80, ipTM > 0.7 for both targets, correct helix orientation) retained 7 candidates. Finally, MD-based binding free energy calculations (ΔG < −10 kcal/mol for both targets) selected 2 candidates (SIO_b5 and SIO_b40) for experimental validation, representing an overall success rate of 2% (2/100). For the FD method, out of 1000 initial linker backbones and 3000 sequences, 327 passed initial filters, 81 survived visual inspection, and 3 were selected after MD simulations (FD_b1-b3), giving a success rate of 0.3% (3/1000). These success rates are consistent with established de novo protein design pipelines [11]. The in silico scores for the five final binders were provided in Table S4 and their amino acid sequences were listed in Table S5. The schematic representations of the binders and their targets were shown in Fig. 3a. Candidate molecules demonstrating favorable in silico interactions with both SzM and TNFR1 were expressed using an E. coli BL21 prokaryotic expression system and purified by fast protein liquid chromatography (FPLC). The molecular weight and purity of the purified proteins were confirmed by SDS-PAGE (Fig. S4a). We employed bio-layer interferometry (BLI) to assess the binding affinities of the expressed binders to SzM and TNFR1 individually. All binders exhibited micromolar affinity towards both targets, suggesting their potential to effectively inhibit the biological functions of SzM and TNFR1 (Fig. 3b and c). The BLI-measured affinities of the parental single-target binders were 4.98 μM for SzM_binder3 binding to SzM and 0.97 μM for TNFR1_s1b2 binding to TNFR1 (Fig. S3). In comparison, the dual-target binders exhibited no significant change in affinity for SzM, whereas their affinity for TNFR1 was reduced (approximately 8∼20-fold higher KD values). This may be because the two binding helices originally optimized for SzM and TNFR1, when fused into a single triple-helical bundle scaffold (SIO design), cannot simultaneously adopt their ideal binding geometries. Such spatial reorientation inevitably introduces small deviations at the binding interfaces, and the more pronounced affinity loss for TNFR1 suggests that its engagement geometry may be more constrained by the bundle architecture. Furthermore, while we used RFdiffusion and ProteinMPNN for optimization, the iterative refinement of a multi-target binder is inherently more challenging than optimizing a single interface. Balancing overall foldability with high-affinity binding at two separate sites may lead to suboptimal interface complementarity at one or both targets.Fig. 3Characterization of the dual-target binders. a) Schematic representations of the binders and their two targets. b, c) The binding affinities of each binder to SzM (b) and TNFR1 (c) were determined using BLI. The fitted curves were represented by black line in each panel and the corresponding binding affinities were indicated at the top. The colored curves represented experimental data at different concentrations as indicated. Characterization of the dual-target binders. a) Schematic representations of the binders and their two targets. b, c) The binding affinities of each binder to SzM (b) and TNFR1 (c) were determined using BLI. The fitted curves were represented by black line in each panel and the corresponding binding affinities were indicated at the top. The colored curves represented experimental data at different concentrations as indicated. The thermal stability of the binders was assessed by differential scanning fluorimetry (DSF), and the Tm values were determined from the derivative curve of the DSF assay. The results indicated that both SIO_b5 and SIO_b40 exhibited exceptional thermal stability, with Tm values exceeding 95 °C. In contrast, FD_b1, FD_b2, and FD_b3 showed moderate stability, with Tm values around 70 °C, which was notably lower than those of SIO_b5 and SIO_b40 (Fig. S4b). Dual-target binders inhibited SEZ adhesion in vitro The bioactivities of the dual-target binder were assessed via a series of in vitro assays. SzM is involved in binding to fibrinogen and facilitates SEZ survival in host blood [34]. We conducted an adhesion inhibition assay using HEp-2 cells, which are recognized as host cells for Streptococcus adhesion [[57], [58], [59], [60]]. After allowing bacterial attachment, the cells were lysed and plated for colony counting. The results revealed that treatment with each binder markedly reduced the number of bacterial colonies in a dose-dependent manner (Fig. 4), confirming their efficacy in inhibiting SEZ adhesion to host cells.Fig. 4Inhibition of SEZ adhesion in vitro. a-f) The effects of SIO_b5 (a), SIO_b40 (b), FD_b1 (c), FD_b2 (d), FD_b3 (e) and SzM_binder3 (f) on SEZ adhesion to HEp-2 cells. Bacterial culture was mixed with various binders in concentrations as indicated for 30 min, then the mixture was co-incubated with the cells for 2 h. BSA was served as a negative control. All data were obtained from three independent experiments and represented as means ± SEM (n = 3 per group). The statistical significances were analyzed by one-way ANOVA. ns, no significance; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. Inhibition of SEZ adhesion in vitro. a-f) The effects of SIO_b5 (a), SIO_b40 (b), FD_b1 (c), FD_b2 (d), FD_b3 (e) and SzM_binder3 (f) on SEZ adhesion to HEp-2 cells. Bacterial culture was mixed with various binders in concentrations as indicated for 30 min, then the mixture was co-incubated with the cells for 2 h. BSA was served as a negative control. All data were obtained from three independent experiments and represented as means ± SEM (n = 3 per group). The statistical significances were analyzed by one-way ANOVA. ns, no significance; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. Dual-target binders blocked TNFR1 signals and downstream functions in vitro and in vivo We first evaluated the effects of the binders on the biological functions of TNFα, and TNFR_s1b2 was used as a positive control. We used human TNFα since both murine TNFα and human TNFα can signal through murine TNFR1 [61]. The L929 cells, a TNFα sensitive cells line, was used to observe the protection of binders against TNFα-dependent cytotoxicity [62,63], and thereby compare the relative protective potency of different binders. As expected, TNFα treatment alone markedly reduced the cell viability. In contrast, co-incubation with the binders conferred a dose-dependent protective effect, significantly restoring cell survival. Notably, the half-maximal effective concentration (EC50) ranged from 30 nM to 130 nM, demonstrating nanomolar potency in counteracting TNFα-induced cell death (Fig. 5a). Of note, the BLI-measured binding affinities of these dual-target binders for TNFR1 (7-50 μM) were considerably weaker than their EC50 values, yielding a 170- to 250-fold discrepancy. A similar discrepancy was observed for the single-target TNFR1 binder in our previous work (∼200-fold difference) [45]. This discrepancy likely attributed to signal amplification in the TNFα pathway, enhanced functional avidity due to TNFR1 clustering on cell membranes [64], and the high integrative sensitivity of the cell survival assay.Fig. 5Inhibition of TNFα signaling. a) The designed dual-target binders reversed TNFα-mediated cell death of L929 cell line. The binder solution with concentrations ranged from 1000 nM to 0.46 nM were added to cells and incubated for 30 min at 37 °C. Then TNFα and actinomycin D at final concentration of 10 pM and 1 μg/ml were added to cells. After 24 h incubation, the cell viabilities were measured by CCK-8 kit. The dose-response curves and EC50 were fitted using GraphPad Prism 10 software. For each binder, experiments were repeated independently three times (n = 3). b) Dual luciferase assay was performed to determine TNFα-mediated NF-κB signaling. HEK293T cells were pretreated with indicated concentrations of the binders for 30 min, followed by co-incubation with 20 ng/mL TNFα for another 10 min. Then the luciferase activities were measured and normalized with the Renilla activity. c) Western blot was performed to determine the phosphorylation of NF-κB p65. HEK293T cells were pretreated with 10 μM binders for 30 min and then co-incubated with 20 ng/mL TNFα for another 10 min. The representative image of the expressions of phospho-p65, p65, and β-actin were displayed. The ratio of phospho-p65 to p65 was calculated to represent the relative of expression of phospho-p65. Western blot band intensities from three independent experiments were quantified using ImageJ. The data were expressed as fold induction relative to the control group which was set as 1. d) Dual-target binders treatment inhibited TNFα-induced neutrophil recruitment to the peritoneal compartment in mice (n = 3). Mice were intraperitoneally injected with TNFα (1 μg/kg) and the binder solution (10 mg/kg). After 3 h, peritoneal lavage fluid was collected and analyzed by flow cytometry to detect CD45+GR-1+ cells. The representative flow images were displayed. e) The proportions of CD45+GR-1+ cells from three individuals. All data were obtained from three independent experiments and represented as means ± SEM (n = 3 per group). The statistical significances were analyzed by one-way ANOVA. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗∗P < 0.0001. Inhibition of TNFα signaling. a) The designed dual-target binders reversed TNFα-mediated cell death of L929 cell line. The binder solution with concentrations ranged from 1000 nM to 0.46 nM were added to cells and incubated for 30 min at 37 °C. Then TNFα and actinomycin D at final concentration of 10 pM and 1 μg/ml were added to cells. After 24 h incubation, the cell viabilities were measured by CCK-8 kit. The dose-response curves and EC50 were fitted using GraphPad Prism 10 software. For each binder, experiments were repeated independently three times (n = 3). b) Dual luciferase assay was performed to determine TNFα-mediated NF-κB signaling. HEK293T cells were pretreated with indicated concentrations of the binders for 30 min, followed by co-incubation with 20 ng/mL TNFα for another 10 min. Then the luciferase activities were measured and normalized with the Renilla activity. c) Western blot was performed to determine the phosphorylation of NF-κB p65. HEK293T cells were pretreated with 10 μM binders for 30 min and then co-incubated with 20 ng/mL TNFα for another 10 min. The representative image of the expressions of phospho-p65, p65, and β-actin were displayed. The ratio of phospho-p65 to p65 was calculated to represent the relative of expression of phospho-p65. Western blot band intensities from three independent experiments were quantified using ImageJ. The data were expressed as fold induction relative to the control group which was set as 1. d) Dual-target binders treatment inhibited TNFα-induced neutrophil recruitment to the peritoneal compartment in mice (n = 3). Mice were intraperitoneally injected with TNFα (1 μg/kg) and the binder solution (10 mg/kg). After 3 h, peritoneal lavage fluid was collected and analyzed by flow cytometry to detect CD45+GR-1+ cells. The representative flow images were displayed. e) The proportions of CD45+GR-1+ cells from three individuals. All data were obtained from three independent experiments and represented as means ± SEM (n = 3 per group). The statistical significances were analyzed by one-way ANOVA. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗∗P < 0.0001. To further determine whether these binders inhibit TNFR1 downstream signaling, we examined the activation of NF-ĸB. A NF-κB luciferase reporter assay was performed, and the results showed that binders pre-treatment significantly suppressed the luciferase activity in a dose-dependent manner (Fig. 5b). Consistent with this, western blot analysis showed that stimulation with recombinant TNFα alone enhanced phosphorylation of p65 in HEK293T cells, whereas preincubation with the binders for 30 min markedly attenuated the phosphorylation (Fig. 5c and d). Together, these results confirm that the binders effectively inhibit TNFα/TNFR1-triggered signaling pathways. We also evaluated the in vivo efficacy of the binders in modulating the TNFα signaling using a mouse model. It has been previously established that intraperitoneal injection of exogenous TNFα induces robust recruitment of polymorphonuclear neutrophils (PMNs) into the peritoneal cavity [65]. Accordingly, mice were co-administered TNFα along with the binders via intraperitoneal injection. Peritoneal lavage was subsequently performed, and recruited neutrophils were identified as CD45+GR-1+ cells by flow cytometry. Treatment with the binders resulted in a significant reduction in the proportion of CD45+GR-1+ cells (Fig. 5e and f), demonstrating effective suppression of TNFα-induced neutrophil recruitment in vivo. Dual-target binders offered optimal protection against the SEZ-infected mice To evaluate the protective efficacy of the dual-target binders against a lethal SEZ challenge, mice were infected intraperitoneally and simultaneously treated with the respective binders via the same route (Fig. 6a). Significantly, administration of the two triple-helical binders (SIO_b5 and SIO_b40) conferred complete (100%) protection, whereas the three penta-helical binders provided lower survival rates, ranging from 50% to 80% (Fig. 6b), despite their favorable binding affinity and bioactivity in vitro (Fig. 3). We speculate that the compact triple-helix bundle conformation of the SIO design likely facilitates more efficient transperitoneal absorption and tissue penetration.Fig. 6Protective effects of dual-target binders to SEZ-infected mice. a) Diagram of the bacteria challenge and binder treatment in the survival study. b) Survival rates of mice (n = 11) infected with a lethal dose of SEZ (2 × 107 CFU) followed by binder (10 mg/kg) or PBS treatments. c) Diagram of the bacteria challenge and binder treatment in the colony counting and histopathological examination study. d) The bacterial load in various tissues (n = 5). At 48 h post infection and treatments, the mice were euthanized and the tissues were harvested and homogenized for bacterial load counting. e) The serum concentrations of TNFα (n = 5). At 48 h post infection and treatments, blood was collected and serum was isolated. The concentrations of TNFα were measured by ELISA assay. f) Histologic analysis of the brain. At 48 h post infection and treatment, the mice were euthanized, and the brains were harvested for H&E staining. The images were representative of five biological replicates, with the scale bar indicating a length of 0.1 mm. All data were represented as means ± SEM. The statistical significance in b was analyzed by log[rank] (Mantel–Cox) test, and those in d-e by one-way ANOVA. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. Protective effects of dual-target binders to SEZ-infected mice. a) Diagram of the bacteria challenge and binder treatment in the survival study. b) Survival rates of mice (n = 11) infected with a lethal dose of SEZ (2 × 107 CFU) followed by binder (10 mg/kg) or PBS treatments. c) Diagram of the bacteria challenge and binder treatment in the colony counting and histopathological examination study. d) The bacterial load in various tissues (n = 5). At 48 h post infection and treatments, the mice were euthanized and the tissues were harvested and homogenized for bacterial load counting. e) The serum concentrations of TNFα (n = 5). At 48 h post infection and treatments, blood was collected and serum was isolated. The concentrations of TNFα were measured by ELISA assay. f) Histologic analysis of the brain. At 48 h post infection and treatment, the mice were euthanized, and the brains were harvested for H&E staining. The images were representative of five biological replicates, with the scale bar indicating a length of 0.1 mm. All data were represented as means ± SEM. The statistical significance in b was analyzed by log[rank] (Mantel–Cox) test, and those in d-e by one-way ANOVA. ∗P < 0.05; ∗∗P < 0.01; ∗∗∗P < 0.001; ∗∗∗∗P < 0.0001. Compared to the combination of SzM_binder3 and TNFR1_s1b2, although the data come from different experiments, the PBS control groups were virtually identical across all batches (100% death within 96 h), allowing meaningful cross-experiment comparison. The dual-target binder achieved complete protection, a level not reached by single-target binders or their combination. This indicated that the dual-target design provided a meaningful advantage over both the individual single-target binders and their combination. To investigate the effects of binders on bacterial adhesion and TNF signaling, at 48 h post infection and administration, the major tissues including lung, liver, kidney, spleen, brain and heart were collected for bacterial colonies counting and histopathological analysis, and the mice serum was collected for TNFα concentrations measurement (Fig. 6c). The results showed that binders administration dramatically decreased the bacterial load in all collected tissues (Fig. 6d), indicated that the binders, especially the two triple-helical binders, effectively inhibited bacterial adhesion. Correspondingly, the levels of serum TNFα were also significantly decreased, and SIO_b5 and SIO_b40 treatment groups exhibited lower serum concentrations (Fig. 6e). Consistent with this trend, mechanistic analyses revealed that the dual-target binder also suppressed SEZ-induced upregulation of Il6 and Il1b mRNA, p65 phosphorylation, and nuclear translocation of p65 in representative lung tissues, with the SIO-designed binders exhibiting more potent inhibitory effects (Fig. S6). These data confirmed effective blockade of TNFα signaling in the infectious context, corroborating the superior efficacy of the SIO design. Histopathological analysis by H&E staining further revealed marked alleviation of SEZ-induced pathological damage, including inflammatory cell infiltration and tissue congestion, in the brains, lungs, and spleens of treated mice (Fig. 6f and Fig. S5). Consistent with the drastic clearance of bacteria, bacterial growth was barely detected in the organs of any surviving mice treated with the two triple-helical binders at the experimental endpoint (Fig. S7). The collective in vivo evidence highlights the potent therapeutic efficacy of the triple-helical binders and underscores the advantage of the dual-targeting strategy over single-target therapies. In the groups treated with the three penta-helical binders, the bacterial loads in the organs of non-surviving mice were comparable to those in the untreated controls. Notably, even the surviving individuals in these groups still harbored a detectable, albeit low, bacterial burden, which was likely maintained at a sublethal level and thus not lethal (Fig. S7). We also measured the pharmacokinetics of the binders. After a single intraperitoneal injection, the maximum serum concentrations reached 63.20 nM for SIO_b5, 79.68 nM for SIO_b40, 50.83 nM for FD_b1, 41.89 nM for FD_b2, and 29.57 nM for FD_b3, with SIO_b5 exhibiting the longest half-life of 0.73 h (Fig. S8). Rapid clearance is a general phenomenon among de novo designed mini-binders, yet these binders have demonstrated significant therapeutic effects [2,66]. In this study, the acute lethal SEZ infection model exhibited 100% mortality within 96 h, indicating that the bacterial dissemination and cytokine storm occurred within the first 24∼48 h. Our daily dosing regimen ensured sufficient exposure during this period. Moreover, intraperitoneal co-injection created high local concentration immediately, enabling rapid inhibition of bacterial adhesion and TNFα signaling. By the time the binder is cleared (within hours), the initial bacterial load and inflammation have already been reduced, and the subsequent dose cleared residual infection. Similar effects have been demonstrated by reltecimod (a CD28 mimetic peptide with a plasma half life of only 1-8 min), which provides sustained survival benefit in sepsis models through rapid tissue redistribution rather than prolonged plasma residence [67]. In fact, a short-half-life can be an advantage as long-half-life anti-inflammatory biologics (such as tocilizumab against IL-6R and anakinra against IL-1R) may induce sustained systemic anti-inflammatory effects and thus be less suitable for acute or localized treatments [4]. Substantial clinical evidence indicates that conventional anti-TNF therapies increase the risk of bacterial infections [68], particularly during concurrent severe bacterial exposure. Complete blockade of TNF function is undesirable given its essential roles in anti-infection defense, anti-tumor immunity, and immune homeostasis [69]. The short half-life and micromolar (μM) affinity of the binders for TNFR1 may reduce the potential risks of high-affinity antibody-based TNF inhibitors like Infliximab, Adalimumab [68]. Although the long-term safety of the binders requires further investigation, our data demonstrate their exceptional efficacy in the acute infection model, highlighting their potential for acute therapeutic use. The short-half-life could be suboptimal for chronic use yet advantageous for acute infections like SEZ. For applications requiring extended exposure, established strategies such as Fc fusion could be employed to modulate their pharmacokinetic profile [7]. Although both dual-target mini-binders and a combination of single-target binders demonstrate significant therapeutic effects, there are distinct advantages of the dual-target design. First, in this study, co-administration of the two single-target binders never achieved complete protection, whereas the dual-target binders SIO_b5 and SIO_b40 consistently conferred 100% survival. Second, a single dual-target molecule avoids the unpredictable concentration ratios that arise from independent pharmacokinetics of two separate binders. Moreover, a single molecule may distribute more uniformly into deep tissues. Third, from a translational standpoint, developing one molecular entity is substantially less complex than two independent biologics, offering lower production costs, simplified dosing, and easier quality control. These advantages support the dual-target mini-binder strategy as a promising approach for acute infection therapy and broader biotherapeutic applications. Conclusion In this study, we have established and validated the RFdiffusion-based computational strategy for the de novo design of dual-target mini-binders. By developing and comparing two distinct methods, Superimposition-Optimization (SIO) and Fusion-Diffusion (FD), we successfully generated single molecular entities capable of concurrently engaging the bacterial virulence factor SzM and the host inflammation receptor TNFR1. The lead candidates, SIO_b5 and SIO_b40, demonstrated that this dual-targeting approach is not only feasible but can achieve synergistic therapeutic outcomes, as evidenced by complete protection in a lethal SEZ infection model. Beyond providing a potent therapeutic candidate against SEZ, this study delivers a robust and adaptable framework. The developed method can be directly applied to the design of mini-binder therapeutics for other complex diseases where multi-pathway intervention is desirable. Future efforts will focus on optimizing the pharmacokinetic profiles of these binders and exploring their application against a broader range of infectious and inflammatory diseases, paving the way for a new class of smart, multi-specific protein drugs. CRediT authorship contribution statement Ke Ming: Data curation, Formal analysis, Funding acquisition, Methodology, Visualization, Writing – original draft. Yang Hu: Data curation, Investigation, Validation, Writing – review & editing. Banbin Xing: Data curation, Methodology, Software, Validation, Writing – review & editing. Meiting Chen: Investigation, Validation. Yuhua Chen: Investigation, Validation. Xiujuan Chen: Investigation, Validation. Miaomiao Geng: Investigation, Validation. Zhiyong Wu: Investigation, Validation. Yanxia Liu: Investigation, Validation. Lin Wei: Methodology, Writing – review & editing. Jun Weng: Conceptualization, Methodology, Writing – review & editing. Zigong Wei: Conceptualization, Funding acquisition, Project administration, Resources, Supervision, Writing – review & editing. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.