Deep Learning‐Enabled Engineering of a Hyper‐Stable and Soluble MPB70‐83 Antigen for Sensitive Bovine Tuberculosis Surveillance - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Official websites use .gov A .gov website belongs to an official government organization in the United States. Secure .gov websites use HTTPS A lock ( Lock Locked padlock icon ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites. Search Log in Dashboard Publications Account settings Log out Search… Search NCBI Primary site navigation Search Logged in as: Dashboard Publications Account settings Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide PERMALINK Copy As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Microb Biotechnol . 2026 Apr 16;19(4):e70348. doi: 10.1111/1751-7915.70348 Search in PMC Search in PubMed View in NLM Catalog Add to search Deep Learning‐Enabled Engineering of a Hyper‐Stable and Soluble MPB70‐83 Antigen for Sensitive Bovine Tuberculosis Surveillance Wen‐Hao Wang Wen‐Hao Wang 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 2 College of Animal Science and Technology, Shihezi University, Shihezi, China Find articles by Wen‐Hao Wang 1, 2 , Jia‐Zhen Ge Jia‐Zhen Ge 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 3 College of Veterinary Medicine, Gansu Agricultural University, Lanzhou, China Find articles by Jia‐Zhen Ge 1, 3 , Ying‐Ying Xie Ying‐Ying Xie 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 4 College of Veterinary Medicine, South China Agricultural University, Guangzhou, China Find articles by Ying‐Ying Xie 1, 4 , Guo‐Dong Song Guo‐Dong Song 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Guo‐Dong Song 1 , Yijian Liu Yijian Liu 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Yijian Liu 1 , Yin‐juan Song Yin‐juan Song 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Yin‐juan Song 1 , Ren‐ge Li Ren‐ge Li 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 3 College of Veterinary Medicine, Gansu Agricultural University, Lanzhou, China Find articles by Ren‐ge Li 1, 3 , Rui‐Shuang Li Rui‐Shuang Li 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Rui‐Shuang Li 1 , Zi‐qing Wang Zi‐qing Wang 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 5 College of Animal Medicine, Xinjiang Agricultural University, Urumqi, China Find articles by Zi‐qing Wang 1, 5 , Xin‐Miao Liu Xin‐Miao Liu 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Xin‐Miao Liu 1 , Shuang‐Shuang Guo Shuang‐Shuang Guo 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Shuang‐Shuang Guo 1 , Jie Li Jie Li 2 College of Animal Science and Technology, Shihezi University, Shihezi, China Find articles by Jie Li 2 , Shengli Chen Shengli Chen 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Shengli Chen 1 , Na Li Na Li 5 College of Animal Medicine, Xinjiang Agricultural University, Urumqi, China Find articles by Na Li 5 , Fuying Zheng Fuying Zheng 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China Find articles by Fuying Zheng 1, ✉ , Yuefeng Chu Yuefeng Chu 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 2 College of Animal Science and Technology, Shihezi University, Shihezi, China 6 Gansu Province Research Center for Basic Disciplines of Pathogen Biology, Lanzhou, China 7 Key Laboratory of Veterinary Etiological Biology, Key Laboratory of Ruminant Disease Prevention and Control (West), Ministry of Agricultural and Rural Affairs, Lanzhou, China Find articles by Yuefeng Chu 1, 2, 6, 7, ✉ Author information Article notes Copyright and License information 1 State Key Laboratory for Animal Disease Control and Prevention, College of Veterinary Medicine, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences, Lanzhou University, Lanzhou, China 2 College of Animal Science and Technology, Shihezi University, Shihezi, China 3 College of Veterinary Medicine, Gansu Agricultural University, Lanzhou, China 4 College of Veterinary Medicine, South China Agricultural University, Guangzhou, China 5 College of Animal Medicine, Xinjiang Agricultural University, Urumqi, China 6 Gansu Province Research Center for Basic Disciplines of Pathogen Biology, Lanzhou, China 7 Key Laboratory of Veterinary Etiological Biology, Key Laboratory of Ruminant Disease Prevention and Control (West), Ministry of Agricultural and Rural Affairs, Lanzhou, China * Correspondence: Fuying Zheng ( [email protected] ), Yuefeng Chu ( [email protected] ) ✉ Corresponding author. Revised 2026 Mar 26; Received 2025 Dec 10; Accepted 2026 Mar 30; Collection date 2026 Apr. © 2026 The Author(s). Microbial Biotechnology published by John Wiley & Sons Ltd. This is an open access article under the terms of the http://creativecommons.org/licenses/by-nc/4.0/ License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. PMC Copyright notice PMCID: PMC13084532 PMID: 41988819 ABSTRACT Bovine tuberculosis (bTB) diagnostics are frequently hindered by the structural instability and insolubility of recombinant multi‐epitope proteins in prokaryotic expression systems. To address this, we developed an integrated strategy combining immunoinformatics with the ProteinMPNN deep learning framework. We constructed a fusion protein utilizing B‐cell epitopes from MPB70 and MPB83, where ProteinMPNN was employed to redesign non‐epitope scaffolds for optimized thermodynamic stability. The structural integrity was verified via 200‐ns molecular dynamics (MD) simulations. MD analysis revealed the redesigned construct transitions into a rigid, compact native state with a stable hydrophobic core. Validating this in silico design, the optimized protein achieved high‐yield soluble expression in Escherichia coli , eliminating inclusion body formation. The resulting indirect ELISA demonstrated 96.30% sensitivity and 98.61% specificity, showing excellent concordance (Cohen's κ > 0.9) with standard assays. This study demonstrates that deep learning‐based sequence redesign effectively resolves solubility bottlenecks in antigen engineering, providing a robust, scalable tool for precision bTB surveillance. Keywords: bovine tuberculosis diagnosis, molecular dynamics simulation, multi‐epitope fusion protein, Mycobacterium bovis , ProteinMPNN, soluble expression Schematic representation of the research workflow for the efficient diagnosis of bovine tuberculosis employing peptide fusion proteins. Abbreviations BSA bovine serum albumin bTB bovine tuberculosis CSAD commercialized secondary antibody diluen FSG fish skin glue GRAVY grand average of hydropathicity IgG immunoglobulin G Mb M. bovis NCBI National Center for Biotechnology Information OD opticaldelnsity PBST phosphate—buffered saline with Tween—20 SMP skim milk powder WHO World Health Organization 1. Introduction Mycobacterium bovis ( M. bovis ), the etiological agent of bovine tuberculosis (bTB), possesses a broad host spectrum including cattle, wildlife, and humans, posing severe challenges to the livestock industry and public health under the One Health framework (Brorson et al. 2003 ; Buddle et al. 2010 ; Borham et al. 2022 ; Bagcchi 2023 ). Despite long‐standing eradication efforts, bTB persists due to the pathogen's ability to establish chronic, subclinical infections and the limitations of current diagnostic tools (Camussone et al. 2009 ; Firdessa et al. 2012 ; Enayatkhani et al. 2021 ; Collins et al. 2022 ). The gold standard, the tuberculin skin test (TST), is hampered by operational complexity and false positives due to environmental mycobacteria (Freigassner et al. 2009 ). While interferon‐gamma (IFN‐γ) release assays improve specificity, their high cost and logistical requirements restrict widespread application in resource‐limited settings (Garnier et al. 2003 ; Galdino et al. 2016 ; Goncalves et al. 2024 ). Consequently, there is an urgent need for accurate, cost‐effective, and high‐throughput serological diagnostic methods (Griffin et al. 1994 ; Green et al. 2009 ). Serological assays rely on high‐quality antigens. The secreted proteins MPB70 and MPB83 are primary immunodominant targets during M. bovis infection, eliciting strong humoral and cellular immune responses, respectively (Qin‐Wang et al. 2010 ; Hebditch et al. 2017 ). However, the application of these native antigens is constrained by production bottlenecks. In prokaryotic systems like Escherichia coli , MPB70 and MPB83 are prone to misfolding and aggregation into inclusion bodies. Refolding processes are inefficient and often fail to restore the native conformational epitopes required for antibody recognition, leading to compromised diagnostic sensitivity (Jin et al. 2010 ; Jespersen et al. 2017 ; Jones et al. 2024 ). Recombinant multi‐epitope proteins (RMPs) represent a strategic advancement, theoretically allowing the fusion of conserved, high‐affinity epitopes from multiple antigens into a single molecule to broaden diagnostic coverage (Kanagavel et al. 2014 ; Kesidis et al. 2020 ). Immunoinformatics has accelerated the identification of such epitopes, reducing reliance on live pathogen handling (Koo et al. 2005 ; Kumar et al. 2024 ). However, a critical translational gap remains: the direct fusion of heterogeneous peptide fragments often disrupts protein folding landscapes, resulting in structural instability, intrinsic disorder, and insolubility (Kunst 2006 ). Traditional solubility tags often fail to address the underlying thermodynamic instability of the fusion core. Recent breakthroughs in artificial intelligence, specifically deep learning‐based protein design, offer a transformative solution to these ‘foldability’ problems. Algorithms like ProteinMPNN can solve the inverse protein folding problem by redesigning sequences to fit a target backbone structure with high compatibility (Leenaars and Hendriksen 2005 ; Lv et al. 2016 ; Kyro et al. 2025 ). By optimizing residue interactions within the scaffold regions while preserving key epitopes, this approach can theoretically stabilize the tertiary structure and enhance solubility. In this study, we integrated advanced immunoinformatics with the ProteinMPNN deep learning framework to engineer a novel MPB83‐MPB70 multi‐epitope fusion protein. We utilized all‐atom molecular dynamics (MD) simulations to mechanistically elucidate how sequence optimization remodels the protein's energy landscape towards a stable, compact state. Furthermore, we experimentally validated that the computationally redesigned protein achieves soluble expression and superior immunoreactivity, establishing a highly sensitive and specific indirect ELISA for the rapid detection of bTB. 2. Materials and Methods 2.1. Strains, Culture and Serum Samples Single colonies of recombinant Escherichia coli ( E. coli ) were cultured on solid LB agar plates (Solarbio, China) and subsequently cultured in LB broth (Solarbio, China) supplemented with 0.1 mg/mL ampicillin (BBI, China). E. coli Top10 competent cells were used for plasmid cloning, whereas E. coli BL21(DE3) competent cells served as the host for recombinant protein expression. All serum samples used in this study were preserved in the serum bank of the Herbivorous Animal Bacterial Disease Research Team, Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Sciences. Serum samples with well‐defined background originating from cattle herds in Xinjiang, Gansu, Qinghai, Ningxia, Inner Mongolia and Shandong provinces were selected from the serum bank. A total of 243 bovine tuberculosis (bTB)‐positive serum samples were obtained from cattle definitively diagnosed as infected based on a combination of the single intradermal comparative cervical tuberculin (SICCT) test, interferon‐gamma release assays (IGRAs), pathological examination, bacteriological culture, and a commercial M. bovis antibody detection kit. In addition, 184 bTB‐negative serum samples were collected from cattle confirmed to be healthy by all of the above diagnostic methods. 2.1.1. Ethics Statement The animal care procedures and experiments were approved by the Committee for the Ethics of Animal Experiments of the Lanzhou Veterinary Research Institute at the Chinese Academy of Agricultural Sciences (LVRIAEC‐2024‐003). 2.2. Design and Cloning of the Directly Fused (Direct) MPB70–MPB83 Protein The sequence of MPB70 gene (GenBank ID: M33916 ) and the sequence of MPB83 gene (GenBank ID: D64165 ) were retrieved from the NCBI ( https://ncbi.nlm.nih.gov/ ) database. A flexible linker comprising 16 amino acid residues, arranged as four tandem repeats of the sequence Gly‐Gly‐Gly‐Ser (GGGS 4 ), was introduced between MPB70 and MPB83 to construct the fusion expression plasmid pGEX‐6p‐1‐MPB70‐83 (Direct). The entire recombinant plasmid was de novo synthesized by Wuhan GeneCreate Biological Engineering Co. Ltd. (Wuhan, China). 2.3. Multiepitope Chimeric Antigens Design and Cloning Protein sequence scans of MPB70 and MPB83 were performed using the CLBTope and BepiPred prediction servers to identify potential linear B‐cell epitopes (Jespersen et al. 2017 , Kumar et al. 2024 ) Amino acid fragments with higher immunogenicity scores were preferentially selected based on the prediction results to ensure that the designed fusion protein possessed robust antigenic properties. Subsequently, solubility was evaluated using the grand average of hydropathicity (GRAVY) index. Hydrophilic fragments, particularly those with lower (more negative) GRAVY values, were preferentially selected, while sequences with excessively high hydrophobic content were avoided. This strategy was applied to enhance the potential for soluble expression of the fusion protein in the Escherichia coli expression system (Kyte and Doolittle 1982 ). Based on the selected epitope sequences, flexible linkers consisting of the Gly‐Gly‐Gly‐Ser (GGGS) motif were incorporated between adjacent epitopes to preserve structural flexibility and maintain epitope independence. The resulting initial fusion protein construct was designated as MPB70‐83 (Original). The three‐dimensional structure of the fusion protein was subsequently modelled using the Boltz‐2 server (Passaro et al. 2025 ). The complete recombinant gene was de novo synthesized by Wuhan GeneCreate Biological Engineering Co. Ltd. (Wuhan, China). 2.4. Optimization of the Non‐Epitope Scaffold Region Using ProteinMPNN To enhance the overall stability and solubility of the MPB70‐83 (Original) protein, the non‐epitope scaffold region was subjected to sequence optimization using the ProteinMPNN deep learning framework, while the core epitope sequences were preserved (Dauparas et al. 2022 ). The procedure was carried out as follows: first, the protein structures were predicted using the Boltz‐2 algorithm to compare conformational differences between the two constructs. Subsequently, missing residues were repaired with PDBFixer, and the protonation states were determined using PROPKA3. Subsequently, directed evolution sampling was performed at 0.15°C using ProteinMPNN with weights trained on soluble proteins, generating a total of 10 variant sequences. The final expression plasmid was designated pGEX‐6p‐1‐MPB70‐83 (ProteinMPNN). Subsequently, the Protein‐Sol platform was used to predict and visualize the solubility and energy distribution of both proteins, in order to verify the improvements in stability and solubility conferred by the optimized sequence. Subsequently, full‐length gene synthesis was commissioned from Wuhan GeneCreate Biological Engineering Co. Ltd. 2.5. Molecular Dynamics Simulation 2.5.1. System Construction The structures of the initial protein and the optimized protein were selected as the starting models for MD simulations to compare their dynamic behaviours. The simulation systems were built using GROMACS 2025.3 with the AMBER99SB‐ILDN force field (Abraham et al. 2015 ). Each protein structure was solvated in a cubic box using the TIP3P water model, ensuring a minimum distance of 1.0 nm (10 Å) between the protein surface and the box boundaries. To mimic the physiological environment (0.01 M PBS, pH 7.4), the protonation states of ionizable residues were assigned at pH 7.4. Na + and Cl − ions were added to neutralize the system charge and maintain a physiological salt concentration of 0.15 M. 2.5.2. Simulation Parameters and Procedure Energy minimization was performed using the steepest descent algorithm until the maximum force was less than 1000 kJ/mol/nm to remove steric clashes within the protein structures. System equilibration was conducted in two consecutive steps for each system: first, a 100 ps equilibration in the NVT ensemble was performed with the temperature maintained at 300 K using the V‐rescale thermostat; followed by a 100 ps equilibration in the NPT ensemble with pressure maintained at 1.0 bar using the Parrinello‐Rahman barostat. The production simulations were performed in the NPT ensemble for a duration of 200 ns for both the initial and optimized proteins. All bond lengths were constrained using the LINCS algorithm, allowing for a time step of 2 fs. Long‐range electrostatic interactions were calculated using the Particle Mesh Ewald (PME) method with a cutoff of 1.0 nm. 2.5.3. Trajectory Analysis Methods Trajectory analysis was performed using GROMACS built‐in tools to conduct a comparative analysis between the initial and optimized proteins. The structural stability and residual flexibility were assessed by calculating the root mean square deviation (RMSD) and root mean square fluctuation (RMSF), respectively. Protein compactness was evaluated using the radius of gyration (Rg). The solvent accessible surface area (SASA) was calculated to analyse the exposure of the hydrophobic/hydrophilic surfaces. Additionally, intramolecular hydrogen bond analysis was performed to evaluate the contribution of hydrogen bonding (Hbond) networks to the structural stability of the optimized protein compared to the initial state. In addition, a free energy landscape (FEL) was constructed based on the Rg and the RMSD to identify the most probable stable energy basins sampled by the system during the simulation. 2.6. Protein Expression The recombinant plasmids pGEX‐6p‐1‐MPB70‐83 (Direct), pGEX‐6p‐1‐MPB70‐83 (Original), and pGEX‐6p‐1‐MPB70‐83 (ProteinMPNN) were transformed into Escherichia coli BL21(DE3) competent cells (Weidi Biotechnology, China) using the heat‐shock transformation method. The transformed cells were cultured in LB medium supplemented with ampicillin at 37°C with shaking at 180 rpm in a temperature‐controlled incubator. When the optical density at 600 nm (OD 600 ) reached 0.6, protein expression was induced with isopropyl β‐D‐1‐thiogalactopyranoside (IPTG) at a final concentration of 0.5 mmol/L for 4 h. Cells were then harvested by centrifugation and resuspended in phosphate‐buffered saline (PBS). Cells were disrupted by low‐temperature ultrasonication and centrifuged at 10,000 rpm to separately collect the supernatant and bacterial sediment. Protein concentrations in both fractions were adjusted to 1 mg/mL using a BCA Protein Assay Kit (Abbkine, China). Subsequently, the samples were separated by 12% stain‐free SDS‐PAGE (Coolaber, China). Following ultraviolet excitation, images were captured using a ChemiDoc MP Imaging System (Bio‐Rad, USA) to assess the expression of pGEX‐6p‐1‐MPB70‐83 (Direct), pGEX‐6p‐1‐MPB70‐83 (Original), and pGEX‐6p‐1‐MPB70‐83 (ProteinMPNN). 2.7. Protein Purification First, E. coli cultures were collected into centrifuge tubes and centrifuged at 4000 g for 20 min at 4°C. The supernatant was discarded, and the cell pellet was retained for subsequent processing. Cells were lysed by ultrasonication on ice at a power output of 200–300 W, with six cycles of 10 s sonication followed by 10 s pauses. The supernatant was incubated with GST agarose beads (Beyotime, China) at a specified ratio. The mixture was incubated at room temperature for 30–60 min on a tilting shaker or a rotary mixer. A total of 2 mL of PBS buffer (pH 7.4) was added, and the GST agarose magnetic beads were gently resuspended by pipetting. The tube was placed on a magnetic stand for 10 s to separate the beads, and the supernatant was discarded. This washing step was repeated three times. The bound proteins were subsequently eluted stepwise with 5, 10, and 20 mM reduced glutathione. For each elution step, the same concentration of elution buffer was added, and the tube was gently inverted several times to resuspend the magnetic beads. After incubation for 5 min, the beads were separated on a magnetic stand, and the eluate was collected into a new centrifuge tube. The collected eluates were considered the purified GST‐tagged proteins. The purity was verified by 12% stain‐free SDS‐PAGE. 2.8. Western Blot Protein concentrations of pGEX‐6P‐1‐MPB70‐83 (Original) and pGEX‐6P‐1‐MPB70‐83 (ProteinMPNN) were determined by BCA assay and adjusted to 1 mg/mL. The original MPB70‐83 was predominantly expressed as inclusion bodies. Although we attempted to solubilize and refold this inclusion‐body fraction, we were unable to obtain a stable soluble preparation. Therefore, the inclusion‐body preparation of MPB70‐83 (Original) was used for Western blotting, and antigen loading was normalized by total protein concentration to match that of MPB70‐83 (ProteinMPNN). We note that inclusion‐body‐derived protein may be partially misfolded, which can reduce epitope accessibility and may introduce bias in a direct comparison of Western blot reactivity with the optimized soluble antigen. Western blot analysis was performed using M. bovis ‐positive bovine serum (1:5000 dilution) as the primary antibody and HRP‐conjugated rabbit anti‐bovine IgG polyclonal antibody (ImmunoWay, USA; 1:10,000 dilution) as the secondary antibody. 2.9. Establishment of an Indirect ELISA for MPB70‐83 (ProteinMPNN) 2.9.1. Determination of the Optimal Antigen Coating Concentration and Serum Working Dilution The preliminary indirect ELISA procedure was performed as follows: (1) Antigen coating—100 μL of antigen solution was added to each well and incubated at 37°C for 2 h, followed by three washes with PBST. (2) Blocking—100 μL of 1% BSA in PBST was added to each well and incubated overnight at 4°C. The wells were washed as described above. (3) Serum incubation—100 μL of diluted serum was added and incubated at 37°C for 1 h. The washing step was repeated as above. (4) Secondary antibody incubation—100 μL of diluted HRP‐conjugated secondary antibody was added and incubated at 37°C for 1 h, followed by the same washing step. (5) Substrate development—100 μL of TMB substrate solution was added and incubated in the dark at 37°C for 15 min. (6) Reaction termination—50 μL of stop solution was added, and the absorbance was measured at 450 nm using a microplate reader for subsequent data analysis. The optimal antigen coating concentration and serum dilution were determined by a checkerboard titration assay. For the checkerboard titration assay, MPB70‐83 (ProteinMPNN) was coated onto microplate wells at concentrations of 1800, 900, 450, 225, 112.5 and 56.25 ng per well, with each concentration tested in duplicate columns. Positive control serum and negative control serum were serially diluted at 1:50, 1:100, 1:200 and 1:400, with each dilution tested in a single row. The P/N value (ratio of mean OD450 nm (OD 450 ) of positive control to that of negative control) was calculated for each antigen‐serum combination. The antigen coating concentration and serum dilution corresponding to the highest P/N value were selected as the optimal conditions. 2.9.2. Determination of the Optimal Coating Conditions Based on the previously determined optimal reaction system, the coating conditions of the MPB70‐83 (ProteinMPNN) protein were further optimized. The following four conditions were tested: (1) coating at 37°C for 2 h, followed by blocking overnight at 4°C; (2) coating at 37°C for 1 h and then blocking the coating process overnight at 4°C; (3) coating at 4°C overnight; and (4) coating at 37°C for 2 h followed by continued blocking overnight at 4°C. 2.9.3. Screening of the Optimal Blocking Buffer and Blocking Conditions Following the determination of the optimal antigen coating concentration and serum dilution, four different blocking buffers were evaluated, including 1% betaine, 2% trehalose, 1% fish gelatin, and 1% gum arabic. For each buffer, the P/N value—defined as the ratio of the mean OD 450 of the positive control to that of the negative control—was calculated. The blocking buffer yielding the highest P/N value was selected as the optimal blocking buffer. Based on this optimal buffer, different blocking conditions were further assessed, including incubation at 37°C for 1, 1.5, 2 or 3 h, and at 4°C for 12 h. The blocking condition producing the highest P/N value was determined to be optimal. 2.9.4. Screening of the Optimal Serum Diluent and Incubation Conditions Based on the previously established optimal conditions, positive and negative control sera were evaluated to determine the ideal serum incubation time. Incubation periods of 1, 1.5 and 2 h at 37°C were tested, and the incubation duration yielding the highest P/N value was regarded as optimal. In parallel, various serum diluents—1% gelatin, 0.5% gelatin, 2% BSA, 1% BSA and PBST—were screened. The combination of serum incubation time and diluent that produced the maximum P/N value was selected for subsequent assays. 2.9.5. Screening of the Optimal HRP‐Conjugated Antibody Diluent and Dilution Ratio Four diluents—2% BSA, 1% BSA, 5% skim milk solution, and a commercially available HRP‐conjugated antibody diluent—were tested with dilution ratios of 1:5000, 1:8000, 1:10 000 and 1:15 000. The combination generating the highest P/N value was selected. 2.9.6. Screening of the Optimal Incubation Conditions for the HRP‐Conjugated Antibody Using the optimal diluent and dilution ratio determined above, incubation times of 0.5, 1 and 1.5 h at 37°C were compared. The time yielding the maximum P/N value was designated as optimal. 2.9.7. Screening of the Optimal TMB Colour Development Time All other procedures were performed as previously described, with TMB incubation times set at 5, 10, 15 and 20 min. The incubation time yielding the highest P/N value was identified as the optimal colour development time. 2.10. Determination of the Cut‐Off Value by ROC Analysis Based on the preliminarily established indirect ELISA, serum samples with well‐defined background from the serum bank described in Section 2.1 were tested, and the P/N value for each sample was calculated. A receiver operating characteristic (ROC) curve was then constructed from the results, and the P/N value corresponding to the maximum Youden's index was selected as the cut‐off value. 2.11. Assessment of the Analytical Performance of the Indirect ELISA 2.11.1. Evaluation of Sensitivity Positive bovine M. bovis sera were serially diluted to 1:100, 1:200, 1:400, 1:1600, 1:3200, 1:6400, and 1:12 800. The dilutions were tested using the established indirect ELISA to determine the assay sensitivity. 2.11.2. Specificity Analysis Sera positive for Brucella spp., Mycobacterium avium subsp. paratuberculosis, Mycoplasma bovis , Pasteurella multocida , Mycobacterium smegmatis , Mycobacterium kansasii , and Escherichia coli were tested using the established indirect ELISA to evaluate the assay specificity. 2.11.3. Assessment of Repeatability To evaluate the repeatability of the established indirect ELISA, both intra‐assay and inter‐assay repeatability tests were performed. 2.11.3.1. Intra‐Assay Repeatability ELISA plates coated in the same batch were used. Four bovine tuberculosis (bTB)‐positive serum samples and four bTB‐negative serum samples were selected, with each sample tested in triplicate. The assay was performed according to the established indirect ELISA procedure, and the coefficients of variation (CV) were calculated to assess intra‐assay variability. 2.11.3.2. Inter‐Assay Repeatability ELISA plates from different coating batches were used. The same eight serum samples (four positive and four negative) as in the intra‐assay test were tested in triplicate. The assay was performed using the same indirect ELISA procedure, and the CVs were calculated to assess inter‐assay variability. 2.12. Detection and Analysis of Clinical Samples To evaluate the performance of the established indirect ELISA in detecting clinical samples, a total of 100 bovine serum samples were tested, including 70 positive and 30 negative specimens. All samples were simultaneously analysed using the established indirect ELISA, a commercial M. bovis ELISA kit, and a commercial colloidal gold strip assay kit, following the respective manufacturers' instructions. The detection results obtained by the established assay and the commercial methods were compared, and the agreement rate was calculated. Cohen's kappa ( κ ) statistic was used to assess the level of agreement between the methods. 2.13. Statistical Analysis All experiments were independently repeated three times with each sample analysed intriplicate. Statistical analyses were performed using GraphPad Prism 9.5.0 software with non‐paired t ‐tests and one‐way or two‐way ANOVA. Significant differences were marked asfollows: * p < 0.05, ** p < 0.01, *** p < 0.001 and**** p < 0.0001: ‘ns’ indicates no significance. The ROC curve was generated by plotting the true positive rate (sensitivity) against the false positive rate (1−specificity) at various cutoff points. The area under the curve (AUC) was calculated to provide a quantitative measure of the overall discriminative ability of the test. The optimal cutoff value was determined using the Youden index (sensitivity + specificity − 1), which maximizes both sensitivity and specificity. The Kappa statistic ( κ ) is calculated according to the following formula: κ = ( P o − P e )/(1 − P e ) where P ₒ (observed agreement) represents the proportion of cases in which the results obtained by the two methods are identical, and P ₑ (expected agreement by chance) denotes the theoretical proportion of agreement that would occur if the results of the two methods were randomly distributed. 3. Results and Analysis 3.1. Design and Optimization of MPB70‐83 Fusion Constructs The structural schematic of MPB70‐83 (Direct) is shown in Figure 1 . The construct consists of MPB70 followed by MPB83, linked via a flexible (GGGS) 4 linker. FIGURE 1. Open in a new tab Schematic representation of the MPB70‐83 (Direct) construct, showing MPB70 linked to MPB83 via a flexible (GGGS) 4 linker. As shown in Table 1 , the design of MPB70‐83 (Original) was based on BepiPred scoring, with preference given to peptide regions predicted to display higher immunogenicity, in order to ensure that the resulting fusion protein possessed strong antigenic potential. In addition, GRAVY values were considered, and sequences with more negative scores (indicating higher hydrophilicity) were prioritized to improve overall solubility and to avoid hydrophobic regions that could hinder protein expression. Finally, the selected MPB70 and MPB83 fragments were linked via a flexible (GGGS) 4 linker to generate the recombinant fusion protein MPB70‐83 (Original). TABLE 1. B‐cell epitope prediction and physicochemical property analysis of MPB70 and MPB83 peptides. Serial number Epitope sequence CLBTope BepiPred Grand average of hydropathicity (GRAVY) MPB70‐1 SKLPASTIDELKTNSS B‐cell 0.176 −0.115 MPB70‐2 TSPANVVGTRQTLQGASVT B‐cell 0.257 −0.095 MPB70‐3 TANATVYMIDSV B‐cell 0.223 0.546 MPB83‐1 SSTKPVSQDTSPKPATSP B‐cell 0.256 −1.183 MPB83‐2 DLIGRGCAQYAAQNPTGPGSVAG B‐cell 0.251 −0.187 MPB83‐3 TAASNNPMLST B‐cell 0.223 −0.209 MPB83‐4 AAFDKLPAATIDQLKTDAK B‐cell 0.194 0.182 MPB83‐5 AGQASPSRIDGTHQTLQGAD B‐cell 0.260 −0.865 MPB83‐6 GARDDLMVNN B‐cell 0.188 0.179 MPB83‐7 HTANATVYMIDTV B‐cell 0.228 0.369 Open in a new tab The specific arrangement of MPB70‐83 (Original) is illustrated in Figure 2A . The comparison results of these two sequences are shown in Figure S1 . From the N‐terminus, the construct begins with a highly immunogenic and strongly hydrophilic epitope (MPB83‐5), followed by alternating insertion of MPB70 epitopes (e.g., MPB70‐2) to enhance epitope diversity. The first half of the sequence is dominated by hydrophilic epitopes to facilitate initial folding and solubility in the Escherichia coli expression system. In contrast, the latter half progressively incorporates more hydrophobic regions, with the strongest hydrophobic epitope (MPB70‐3) positioned at the C‐terminus. This design aims to mimic the stabilizing hydrophobic core found in native protein structures while avoiding N‐terminal hydrophobicity that could lead to aggregation, thereby ensuring optimal solubility and immunoreactivity of the recombinant protein. FIGURE 2. Open in a new tab Sequence design, structural modelling, and solubility prediction analysis of the MPB70‐83 fusion proteins. (A) Design strategy and schematic arrangement of the MPB70‐83 (Original) fusion protein sequence. (B) Design schematic of the MPB70‐83 (ProteinMPNN) construct. (C) Three‐dimensional structural models of MPB70‐83 (Original) and MPB70‐83 (ProteinMPNN) predicted using Boltz‐2. (D) Predicted solubility scores for prokaryotic expression. (E) Protein‐Sol energy heatmap showing predicted stability and hydrophilicity. The specific arrangement of MPB70‐83 (ProteinMPNN) is illustrated in Figure 2B . Following fixation of the epitope scaffold, targeted optimization of linker regions was performed to enhance overall protein stability. After computational redesign using the ProteinMPNN algorithm, a novel sequence was obtained, which shares 43.24% sequence identity with the original protein. Three‐dimensional structural models of MPB70‐83 (ProteinMPNN) and MPB70‐83 (Original) were constructed using Boltz‐2 (Figure 2C ). After optimization, the predicted solubility for prokaryotic expression in Escherichia coli increased from 0.690 to 0.708 (Figure 2D ). The Protein‐Sol energy heatmap indicated a marked improvement in both stability and solubility of the optimized protein compared with the original construct. At pH 6.0–6.5, which approximates physiological neutrality, the optimized protein exhibited the lowest predicted energy (−194) (Figure 2E ). From an experimental application perspective, the optimized construct is theoretically more stable in solution and capable of achieving a higher soluble concentration. 3.2. Structural Stability Analysis Based on MD Simulation Root Mean Square Deviation (RMSD) Analysis During the 200 ns NPT ensemble simulation, exceptional convergence was observed for the optimized protein. The RMSD rapidly stabilized within the 0.4–0.6 nm range, with no significant conformational drift detected throughout the trajectory. In sharp contrast, the initial sequence failed to reach thermodynamic equilibrium; the RMSD diverged continuously to 4–6 nm, accompanied by drastic fluctuations of up to 2 nm, indicating severe unfolding (Figure 3A ). These features suggest the initial sequence likely behaves as an intrinsically disordered protein (IDP) or possesses a non‐foldable character. Quantitative analysis revealed that structural optimization reduced backbone deviation by approximately 90% and attenuated structural oscillation by 25‐fold. These results demonstrate that sequence redesign significantly enhanced overall rigidity, successfully constraining the conformational space within a stable, native‐like energy basin. FIGURE 3. Open in a new tab Structural stability and conformational dynamics analysis of the initial sequence versus the optimized protein during 100 ns MD simulations. (A) Time evolution of backbone Root Mean Square Deviation (RMSD). The optimized protein (solid line) rapidly achieves thermodynamic equilibrium within the 0.3–0.5 nm range, exhibiting high structural rigidity. In contrast, the initial sequence (dashed line) shows continuous divergence (reaching 5–6 nm) and severe unfolding, characteristic of an intrinsically disordered protein (IDP). (B) Residue‐wise Root Mean Square Fluctuation (RMSF) profiles. The optimized structure demonstrates suppressed atomic fluctuations (mean ≈0.14 nm), indicating a stable tertiary fold, whereas the initial sequence displays aberrant flexibility (1.0–2.0 nm) across most residues. (C) Dynamics of the intramolecular hydrogen bond network. The optimized design maintains a significantly denser and more rigid interaction network (average 127 ± 6 bonds), representing a 35% increase in bond quantity and a 50% reduction in fluctuation compared to the initial structure. (D) Radius of Gyration (Rg) distributions reflecting structural compactness. The optimized protein converges to a highly compact state (Rg≈1.57 nm) with minimal deviation, while the initial structure fails to form a stable core, exhibiting high conformational heterogeneity. (E) Solvent Accessible Surface Area (SASA) trajectory and (F) Residue‐SASA (ResSASA) comparison. The reduction in total SASA and the burial of specific hydrophobic residues (e.g., Res 34, Res 110–111) in the optimized structure confirm the formation of a stable hydrophobic core and the elimination of high‐energy exposed surfaces. (F) FEL of the MPB70‐83 (original) projected onto RMSD and radius of gyration (Rg). The low‐energy region forms an elongated ridge‐like basin spanning a broad RMSD range and coupled to gradual compaction (decreasing Rg). The global minimum occurs at relatively high RMSD and connects smoothly to neighbouring low‐energy regions with weak barriers, indicating a shallow, weakly funnelled landscape with multiple near‐degenerate minima and pronounced conformational heterogeneity. (G) FEL of the MPB70‐83 (ProteinMPNN) variant in the RMSD–Rg space. In contrast to the original protein, the optimized design exhibits a more funnelled landscape dominated by a compact, well‐localized basin at low RMSD within a narrow Rg window, with steeper free‐energy gradients away from the minimum. This concentrated minimum suggests a predominantly single native‐like ensemble with fewer accessible alternative conformations and reduced ruggedness, consistent with enhanced thermodynamic stability. Root Mean Square Fluctuation (RMSF) analysis RMSF calculations further highlighted the disparity in stability. The unoptimized structure exhibited aberrant atomic fluctuations, with most residues showing RMSF values between 1.0 and 2.0 nm, indicative of extreme instability. Conversely, following MD optimization, the system's RMSF values significantly decreased and converged to a rational range (mean~0.19 nm), confirming that the protein had successfully folded and maintained a stable tertiary structure in thermodynamic equilibrium (Figure 3B ). Intramolecular Hydrogen Bond (Hbond) Network To elucidate the atomistic mechanism underlying the enhanced stability, the evolution of intramolecular hydrogen bonds was quantified (Figure 3C ). The optimized protein established a significantly denser hydrogen bond network, maintaining an average of 124 ± 8 bonds—a 18% net increase compared to the initial structure. Dynamically, the fluctuation (standard deviation) in hydrogen bond count was reduced by approximately 50%, reflecting a highly rigid interaction network. Time‐evolution analysis revealed that this dense network reorganized and reached equilibrium at the onset of simulation (~0.1 ns) and remained above a threshold of 104 bonds throughout. This high‐density, low‐fluctuation profile correlates strongly with the observed RMSD plateau, identifying the reinforced hydrogen bond network as a primary physical driver for the transition from flexibility to rigidity. To evaluate the impact of optimization on global compactness and folding stability, we analysed the radius of gyration (Rg) over the 200‐ns simulations (Figure 3D ). The optimized protein exhibited a markedly more compact and stable conformational ensemble, with an average Rg of 1.287 ± 0.067 snm (CV = 5.22%; RMSD to the initial frame = 0.069 nm). In contrast, the pre‐optimization model showed a substantially larger and more heterogeneous Rg distribution (mean 1.830 ± 0.595 nm; CV = 32.50%; RMSD to the initial frame = 3.127 nm). Overall, optimization reduced the mean Rg by~29.7% (from 1.830 to 1.287 nm) and decreased the standard deviation by ~8.9‐fold (from 0.595 to 0.067 nm), indicating a pronounced gain in structural compactness and dynamic restraint. From a temporal perspective, the optimized trajectory remained near its compact basin with limited dispersion, consistent with rapid attainment and maintenance of a stable folded‐like state. Conversely, the pre‐optimization trajectory displayed large‐amplitude excursions and persistent non‐equilibrium‐like variability across the simulation window, suggestive of conformational heterogeneity and incomplete consolidation into a single stable compact state. These distributional characteristics (high CV and large RMSD relative to the starting frame in the pre‐optimization system) support that optimization substantially improved packing coherence and reduced “breathing” motions, thereby stabilizing the global fold. SASA analysis further distinguished the solvent exposure and compactness of the two systems (Figure 3E ). For the initial model, the SASA remained high and poorly converged over the trajectory, with a mean of 119.1 nm 2 and substantial variability (SD = 16.3 nm 2 ; CV = 13.7%). Although a rapid structural rearrangement occurred early in the simulation, the trace did not reach a clear plateau and instead exhibited persistent, slow contraction as evidenced by a non‐negligible positive drift in the terminal segment (last‐window slope = 0.2120 nm 2 /ns) together with a broad fluctuation amplitude (last‐window range = 6.02 nm 2 ; last‐window SD = 1.66 nm 2 ). Collectively, these features indicate continued relaxation and incomplete stabilization of solvent‐exposed surface area. In contrast, the optimized design displayed a markedly more compact and stable solvent‐exposure profile. Its SASA converged to a substantially lower mean value of 100.5 nm 2 with tight dispersion (SD = 3.55 nm 2 ; CV = 3.54%). In the terminal window, fluctuations remained bounded (range = 5.50 nm 2 ; SD = 1.66 nm 2 ) and the trajectory showed no evidence of large‐amplitude excursions, consistent with maintenance of a stable globular state. Overall, the optimized protein reduces solvent‐accessible surface area by ~16% relative to the initial structure (100.5 vs. 119.1 nm 2 ) and exhibits ~4‐fold lower relative variability (CV 3.54% vs. 13.72%), supporting effective shielding of exposed surfaces and improved structural stability in solution. Furthermore, residue‐based SASA (ResSASA) analysis quantified the thermodynamic impact of optimization (Figure 3F ). The mean ResSASA decreased from 0.667 ± 0.474 nm 2 (initial) to 0.563 ± 0.419 nm 2 (optimized), corresponding to a 15.6% reduction. This shift indicates an overall compaction of the conformational ensemble, consistent with redistribution of solvent exposure from surface‐accessible regions towards a more buried core. Notably, several hotspots showed pronounced burial upon optimization: residue 34 exhibited a large decrease in ResSASA (2.372 → 0.466 nm 2 ; −80.4%), supporting its transition from a highly exposed position to a more core‐integrated environment. In contrast to the expectation of a globally strengthened ‘hydrophobic collapse’, the 110–111 segment displayed only a moderate reduction (Res 110: 0.670 → 0.178 nm 2 ; Res 111: 0.394 → 0.213 nm 2 ), suggesting partial burial and/or persistent local breathing rather than a fully occluded state. Collectively, these residue‐resolved changes highlight localized structural rearrangements that accompany optimization and contribute to the observed increase in compactness and stability. FEL analysis highlights a clear thermodynamic divergence between the original MPB70‐83 and the ProteinMPNN‐optimized design (Figure 3G,H ). For the original protein, the RMSD–Rg landscape forms an elongated, ridge‐like low‐energy basin spanning a wide RMSD range and coupled to gradual compaction. The global minimum occurs at relatively high RMSD and connects smoothly to neighbouring low‐energy regions with weak barriers, indicating a shallow, weakly funnelled landscape with multiple near‐degenerate minima and pronounced conformational heterogeneity—consistent with ongoing relaxation and structural drift. In contrast, the optimized variant exhibits a markedly more funnelled landscape, dominated by a compact and well‐localized basin at low RMSD and a narrow Rg window, with steeper energy gradients away from the minimum. This concentrated minimum suggests that the ensemble predominantly occupies a single native‐like state, with fewer accessible alternative conformations and reduced ruggedness. Collectively, these features support enhanced thermodynamic stability of the optimized design, whereas the original protein remains distributed across broader metastable states with incomplete convergence. Conclusion Collectively, multi‐dimensional trajectory analyses (RMSD, RMSF, H‐bonds, Rg, SASA and FEL) consistently demonstrate that the computational optimization strategy successfully overcame the intrinsic disorder of the initial sequence. Through the synergistic effects of a high‐density hydrogen bond network and a compact hydrophobic core, the protein conformation was strictly locked into a low‐energy, compact, and thermodynamically stable native‐like basin. 3.3. MPB70‐83 Optimized by ProteinMPNN Exhibits Soluble Expression and Improved Solubility When induced at 37°C with agitation at 180 rpm using 0.5 mM IPTG for 4 h, both MPB70‐83 (Direct) and MPB70‐83 (Original) were predominantly detected in the insoluble fraction as inclusion bodies after ultrasonication and centrifugation of the cell pellets. In contrast, the MPB70‐83 (ProteinMPNN) protein was detected in the supernatant after ultrasonication, indicating successful soluble expression (Figure 4A,B ). Total bacterial proteins were adjusted to the same concentration (1 mg/mL), separated by SDS–PAGE, and subsequently transferred onto NC membranes. Western blotting was then performed using M. bovis ‐positive and ‐negative bovine sera as the primary antibodies, followed by detection with horseradish peroxidase (HRP)‐conjugated sheep anti‐bovine IgG secondary antibody. The results showed that both MPB70‐83 (Original) and MPB70‐83 (ProteinMPNN) reacted specifically with M. bovis ‐positive bovine serum as the primary antibody; however, at the same protein concentration, MPB70‐83 (ProteinMPNN) exhibited stronger specific reactivity (Figure 4C ). When M. bovis ‐negative bovine serum was used as the primary antibody, neither protein produced any non‐specific bands (Figure 4D ). As demonstrated in Section 3.1 , the fusion expression proteins MPB70‐83 (Direct) and MPB70‐83 (Original) formed inclusion bodies after cell disruption, requiring denaturation and refolding for purification, with the latter exhibiting low efficiency. Therefore, the recombinant MPB70‐83 protein (ProteinMPNN) expressed in the soluble fraction was purified. Elution with 5 mM reduced glutathione (pH 7.8) yielded a target band of the expected molecular weight with high purity (Figure 3E ). FIGURE 4. Open in a new tab Expression, purification and Western blot verification of recombinant proteins. (A) Solubility analysis of MPB70‐83 (Original) and MPB70‐83 (ProteinMPNN). M, protein marker; lane 1, MPB70‐83 (ProteinMPNN) pellet; lane 2, MPB70‐83 (Direct) pellet; lane 3, MPB70‐83 (Direct) supernatant; lane 4, MPB70‐83 (ProteinMPNN) supernatant. m. (B) Solubility analysis of MPB70‐83 (Direct) and MPB70‐83 (ProteinMPNN). M, protein marker; lane 1, MPB70‐83 (Original) pellet; lane 2, MPB70‐83 (Original) supernatant; lane 3, MPB70‐83 (ProteinMPNN) pellet; lane 4, MPB70‐83 (ProteinMPNN) supernatant. (C) Western blot identification of MPB70‐83 (ProteinMPNN) and MPB70‐83 (Original) using M. bovis ‐positive bovine serum as the primary antibody. M, protein marker; lanes 1–2, recombinant MPB70‐83 (ProteinMPNN) and MPB70‐83 (Original) proteins (~50 kDa), respectively. (D) Western blot analysis of MPB70‐83 (ProteinMPNN) and MPB70‐83 (Original) using M. bovis ‐negative bovine serum as the primary antibody. M, protein marker; lanes 1–2, MPB70‐83 (ProteinMPNN) and MPB70‐83 (Original) proteins (~50 kDa), respectively. (E) Purification of MPB70‐83 (ProteinMPNN). M, protein marker; lane 1, flow‐through; lanes 2–3, wash fractions; lanes 4–5, elution fractions obtained with 5 mM reduced glutathione (pH 7.8). 3.4. Establishment and Optimization of an Indirect ELISA for Detection The results showed that when the coating antigen amount was 1800 ng per well and the serum dilution was 1:50, the P/N value reached its maximum (Figure 5A ). Considering that a coating antigen amount of 900 ng per well yielded a comparable P/N value and was more cost‐effective, 900 ng per well was ultimately selected as the optimal coating concentration, while a serum dilution of 1:50 was determined to be the optimal working concentration. During the optimization of coating and blocking conditions, it was found that incubation at 37°C for 2 h resulted in the highest P/N value (Figure 5B ). In addition, using 1% betaine as the blocking buffer also yielded the maximum P/N value (Figure 5C ). Meanwhile, the optimal blocking condition was also determined to be incubation at 37°C for 2 h (Figure 5D ). Based on the above results, the optimal coating condition was determined to be incubation at 37°C for 2 h, the optimal blocking buffer was 1% betaine, and the optimal blocking condition was incubation at 37°C for 2 h. Under these optimized parameters, further evaluations were conducted to determine the optimal serum incubation time and the composition of the serum diluent. The results indicated that a serum incubation period of 1 h at 37°C, combined with the use of 1% fish gelatin as the serum diluent, yielded the highest P/N value. Therefore, the optimal serum incubation time was determined to be 1 h at 37°C (Figure 5E ), and 1% fish gelatin was identified as the optimal serum diluent (Figure 5F ). On this basis, parameters related to the secondary antibody were further optimized. The results showed that when the dilution of goat anti‐bovine IgG secondary antibody was 1:10,000 (Figure 5G ), and a commercialized secondary antibody diluent (sigma, America) was used as the secondary antibody diluent (Figure 5H ), the P/N value reached its maximum in both cases. In addition, the optimal incubation time for HRP‐conjugated goat anti‐bovine IgG was determined to be 1 h (Figure 5I ), and the highest P/N value was obtained when the TMB substrate was developed for 15 min (Figure 5J ). FIGURE 5. Open in a new tab Optimization of iELISA assay conditions. The iELISA method was optimized by altering the following parameters: (A) Heat map of P/N values for different antigen coating amounts and serum dilutions. (B) Effect of coating temperature and time on P/N values. (C) Optimization of blocking buffer composition. (D) Effect of blocking time on P/N values. (E) Effect of serum diluent composition on P/N values. (F) Effect of serum incubation time on P/N values. (G) Effect of secondary antibody dilution on P/N values. (H) Optimization of secondary antibody diluent. (I) Effect of secondary antibody incubation time on P/N values. (J) Effect of TMB development time on P/N values. P/N value represents the ratio of the optical density (OD) of positive samples (P) to that of negative samples (N). * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001. 3.5. Evaluation of the Diagnostic Efficacy of the Indirect ELISA Under optimized reaction conditions, 427 clinical serum specimens with clearly defined background information, collected from various regions as detailed in Section 2.1 , were analysed by indirect ELISA(Figure 6A ). Based on the 450 nm (OD450) values, a distribution plot of serum OD values (Figure 6B ) and a receiver operating characteristic (ROC) curve (Figure 6C ) were generated. By analysing the sensitivity and specificity corresponding to different values of Youden's index, the cut‐off optical density (OD) value for discriminating between positive and negative samples was determined to be 0.4181. Specifically, samples with 450 nm (OD450) ≥ 0.4181 were classified as antibody‐positive, whereas those with 450 nm (OD450) < 0.4181 were classified as antibody‐negative. At this cut‐off value, the Youden's index was 0.944, with a sensitivity of 96.30% and a specificity of 98.61%. Positive sera against M. bovis were serially diluted two‐fold from 1:50 to 1:12,800 and tested using the established indirect enzyme‐linked immunosorbent assay (iELISA). According to the positivity criterion of 450 nm (OD450) ≥ 0.4181, the results indicated that the maximum dilution at which M. bovis ‐positive sera could still be detected was 1:6400 (Figure 6D ). In addition, positive sera against Mycobacterium avium subsp. paratuberculosis , Brucella abortus , Mycoplasma bovis , Salmonella spp., Mycobacterium smegmatis, Mycobacterium kansasii , and Escherichia coli were tested using the established indirect ELISA method. The results showed that all OD values were below the cut‐off value, and thus were determined to be negative (Figure 6E ). These findings indicate that the assay has good specificity and exhibits no cross‐reactivity with positive sera against common bacterial pathogens. FIGURE 6. Open in a new tab Sensitivity and specificity of the indirect ELISA (iELISA). (A) Distribution map of the tested serum samples. (B) Receiver operating characteristic (ROC) curve analysis of the iELISA showing a maximum Youden's index of 0.944, with a sensitivity of 96.30% and a specificity of 98.61%. (C) Determination of the cut‐off value (0.4181) based on the analysis of 126 clinical serum samples. (D) Serial two‐fold dilution of M. bovis ‐positive serum from 1:50 to 1:12,800 tested by iELISA; using the cut‐off value (0.4181), the maximum detectable dilution was 1:6400. (E) Specificity test of the iELISA showing no cross‐reactivity with positive sera against Mycobacterium avium subsp. paratuberculosis (MAP), Salmonella spp., Brucella abortus , Mycoplasma bovis , Mycobacterium smegmatis , Mycobacterium kansasii . 3.6. The Established Method Exhibited Good Reproducibility To evaluate the reproducibility of the indirect ELISA developed in this study, eight serum samples were tested using antigen‐coated microplates from the same batch and from different batches. The results showed that the intra‐assay coefficients of variation (CVs) ranged from 2.35% to 5.67% (Table 2 ), while the inter‐assay CVs ranged from 1.86% to 6.35% (Table 2 ). These findings indicate that the established indirect ELISA method has good intra‐ and inter‐assay reproducibility. TABLE 2. Intra‐ and inter‐assay reproducibility of the indirect ELISA developed in this study. Sera Intra‐batch repeatability Inter‐batch repeatability Average OD values Standard deviation Coefficient of variation/% CV Average OD values Standard deviation Coefficientof Variation/% CV P1 2.441 0.059 2.40 2.487 0.046 1.86 P2 2.146 0.052 2.41 2.390 0.090 3.77 P 3 2.300 0.056 2.45 2.294 0.053 2.29 P4 2.361 0.134 5.67 2.385 0.083 3.48 N1 0.249 0.006 2.59 0.263 0.017 6.35 N2 0.223 0.005 2.35 0.252 0.010 3.93 N3 0.223 0.009 4.11 0.226 0.011 4.66 N4 0.322 0.016 4.97 0.251 0.009 3.56 Open in a new tab Note: CVs were calculated from OD values of eight serum samples. Intra‐assay reproducibility: triplicate measurements within one run; Inter‐assay reproducibility: three runs with different batches of antigen‐coated microplates. 3.7. Concordance Between the Established iELISA and a Commercial ELISA Kit To further evaluate the diagnostic performance of the indirect ELISA (iELISA) developed in this study, a total of 427 clinical serum samples were tested, including 257 tuberculin skin test (TST)‐positive sera and 170 TST‐negative sera. The results obtained with the established iELISA were compared with those of the TST, a commercial M. bovis antibody rapid test strip, and a commercial indirect ELISA kit. Agreement between methods was assessed using Cohen's kappa ( κ ) analysis. As shown in Table 3 , the newly developed indirect ELISA demonstrated a positive concordance rate of 94.55%, a negative concordance rate of 100%, and an overall concordance rate of 90.63% with the tuberculin skin test (TST), yielding a κ value of 0.81. When compared with a commercial M. bovis antibody detection strip, the assay achieved positive, negative, and overall concordance rates of 100%, 97.87%, and 96.72%, respectively, with a κ value of 0.94. In comparison with a commercial indirect ELISA kit, positive, negative, and overall concordance rates were 100%, 94.84%, and 95.78%, respectively, with a κ value of 0.91. In all evaluations, κ values exceeded 0.80, indicating excellent agreement between the developed indirect ELISA and the reference diagnostic methods. TABLE 3. Agreement rates between the established indirect ELISA and other diagnostic methods based on Cohen's kappa analysis. Detection assays MPB70‐83 (ProteinMPNN) indirect ELISA Coincidence rate/% κ values Positive no. Negative no. Total TST Tuberculin skin test Positive no. 230 27 257 94.55 0.81 Negative no. 13 157 170 100 Total 243 184 427 90.63 Antibody test strip Positive N o. 234 5 239 100 0.94 Negative no. 9 179 188 97.87 Total 243 184 427 96.72 Indirect ELISA kit Positive no. 229 4 233 100 0.91 Negative no. 14 180 194 94.84 Total 243 184 427 95.78 Open in a new tab 4. Discussion While serological testing holds promise for diagnosing bovine tuberculosis ( M. bovis ), assays based on single antigens often suffer from suboptimal sensitivity due to the considerable inter‐individual variability in antibody responses (Moens et al. 2023 ; O'Brien et al. 2023 ). Consequently, Recombinant Multiepitope Proteins (RMPs) have emerged as a superior alternative. By integrating multiple immunodominant epitopes into a single molecular entity, RMPs not only broaden the detection spectrum but also overcome the batch inconsistency and antigen interference inherent in simple multi‐antigen mixtures (Anandarao et al. 2006 ; de Souza et al. 2013 ; Faria et al. 2015 ; Taherkhani et al. 2015 ; Thomasini et al. 2018 ). However, the construction of RMPs faces a critical bottleneck: the direct fusion of heterogeneous epitopes frequently leads to structural instability and insolubility (Rosano and Ceccarelli 2014 ). Unlike traditional strategies that rely on solubility‐enhancing fusion tags—which only partially mitigate aggregation without addressing the underlying structural defects—our study employs a Deep Learning‐driven approach using ProteinMPNN to fundamentally optimize the protein sequence (Fast et al. 2009 ; Tyurin et al. 2018 ). As evidenced by our Molecular Dynamics (MD) simulations, this sequence redesign significantly improves the physicochemical properties and thermodynamic stability of the construct (Rouhani et al. 2018 ). This highlights that rational sequence redesign via deep learning is a transformative strategy for developing next‐generation diagnostic antigens. In this study, the novel and high‐accuracy B‐cell epitope prediction tools CLBTope and BepiPred were utilized to prioritize sequences with high predicted immunogenicity according to their prediction scores. GRAVY values were subsequently incorporated into the selection process to identify highly hydrophilic fragments, thereby enhancing solubility and mitigating the potential adverse effects of highly hydrophobic sequences on protein expression. The selected epitopes were ultimately concatenated via a flexible (GGGS) linker to generate the MPB70‐83 (Original) recombinant multiepitope fusion protein. Previous studies have demonstrated that MPB70 and MPB83 are frequently expressed as inclusion bodies in prokaryotic systems, potentially resulting in distortion of epitope conformations. Such expression necessitates denaturation–renaturation procedures, which typically exhibit low refolding efficiencies (often < 30%), and may inadvertently expose cryptic epitopes, thereby increasing the risk of cross‐reactivity with non‐target pathogens (Carr et al. 2003 ; Singh and Panda 2005 ; Jiang et al. 2006 ). Although natural proteins exhibit favourable functional properties, their expression levels and solubility in prokaryotic systems are often limited, and they display temperature sensitivity, thereby constraining their suitability for large‐scale production. Conversely, eukaryotic expression systems can enhance protein quality, but their high cultivation costs substantially limit their feasibility for extensive industrial application. Therefore, the development of a universal strategy capable of enhancing the physicochemical properties of natural proteins while retaining their biological activity holds substantial promise for broad applications in protein‐based technologies. To overcome this challenge, we employed the deep neural network algorithm ProteinMPNN to optimize the linear epitope sequences of the MPB70‐83 multiepitope fusion protein, with the goal of enhancing its solubility, stability, and expression efficiency. Soluble expression performance was comparatively evaluated in two configurations—70–83 Direct and 70–83 RMPs—and the optimized construct demonstrated a marked improvement in soluble expression relative to the original design. A 200 ns all‐atom MD simulation was employed to elucidate the mechanisms by which sequence redesign remodels the protein folding landscape. The results demonstrate that the optimization strategy induced a global phase transition from a disordered, molten globule‐like state to a compact native state, rather than merely fine‐tuning local structural elements (Porebski et al. 2016 ). First, backbone stability metrics demonstrate a qualitative transition in conformational behaviour. Over 200 ns, the optimized protein rapidly converged and maintained a stable RMSD plateau (0.4–0.6 nm) with no detectable long‐term drift, whereas the original sequence failed to reach equilibrium and exhibited continuous RMSD divergence to 4–6 nm with large‐amplitude fluctuations (up to ~2 nm), consistent with severe unfolding and an IDP‐like/non‐foldable character. In agreement, RMSF values for the original protein were broadly elevated (typically 1.0–2.0 nm), while the optimized variant showed substantially reduced and well‐behaved residue fluctuations (mean~0.19 nm), supporting formation of a stable tertiary structure. Second, analysis of the intramolecular hydrogen‐bond network identifies a key physical basis for the enhanced rigidity. The optimized sequence maintained a denser and more stable hydrogen‐bonding pattern (124 ± 8), corresponding to an ~18% increase over the original and ~50% lower fluctuation in bond count. Notably, this network reorganized and reached equilibrium at the onset of simulation (~0.1 ns) and remained above a stable threshold thereafter, closely tracking the rapid RMSD stabilization and indicating that reinforced polar interactions are a primary driver of the shift from flexibility to rigidity. Third, global compactness and solvent exposure further support improved folding and packing. The optimized protein displayed a significantly smaller and more narrowly distributed Rg (1.287 ± 0.067 nm; CV = 5.22%) relative to the original (1.830 ± 0.595 nm; CV = 32.50%), reflecting ~29.7% reduction in mean Rg and ~8.9‐fold reduction in dispersion. Consistently, SASA decreased from 119.1 nm 2 (original) to 100.5 nm 2 (optimized), with markedly improved convergence and ~4‐fold lower relative variability (CV 13.7% vs. 3.54%). Residue‐resolved SASA corroborated this compaction trend (15.6% reduction in mean ResSASA), with pronounced burial at specific hotspots (e.g., residue 34), while residues 110–111 showed only moderate decreases, suggesting partial burial and residual local breathing rather than complete occlusion. Together, these data indicate that optimization improves packing coherence and reduces large‐scale ‘breathing’ motions, thereby stabilizing the globular fold. Finally, FEL analysis provides a thermodynamic summary of these kinetic and structural observations. The original protein exhibits an extended, weakly funnelled basin spanning broad RMSD and Rg ranges, with shallow barriers and multiple near‐degenerate minima, consistent with conformational heterogeneity and continued relaxation. In contrast, the optimized design shows a more funnelled and localized minimum at low RMSD and a narrow Rg interval, indicating a dominant native‐like ensemble with reduced ruggedness and fewer accessible alternative states. In conclusion, multi‐dimensional MD metrics (RMSD, RMSF, H‐bonds, Rg, SASA/ResSASA, and FEL) consistently support that ProteinMPNN redesign converts the original MPB70‐83 from a non‐convergent, disorder/misfolding‐prone conformational regime into a compact, thermodynamically stable, native‐like basin. The enhanced stability is primarily attributable to a denser and less fluctuating intramolecular hydrogen‐bond network, together with a reduced solvent‐exposed surface that promotes tighter global packing and more coherent folding. Consistent with the computational predictions, subsequent expression experiments showed that the sequence‐optimized multiepitope fusion protein exhibited a markedly increased soluble expression fraction in Escherichia coli and a substantially reduced propensity to form inclusion bodies. In contrast, the unoptimized original construct was predominantly detected in the insoluble fraction, indicating that it fails to attain a properly folded conformation in this prokaryotic expression system. To compare the immunoreactivity of the two constructs, we performed Western blot immunoblotting. Importantly, however, directly comparing WB signal intensities between an inclusion body‐dominant protein and a soluble, properly folded protein is inherently biased due to differences in sample state and epitope accessibility. Therefore, the observed differences more plausibly reflect improved antigen presentation arising from enhanced solubility/correct folding, rather than a genuine increase in intrinsic epitope‐based antigenicity. Given that our design strategy redesigned mainly the scaffold/backbone rather than the epitope core sequences, a more parsimonious interpretation is that deep learning–enabled sequence optimization substantially improved foldability and solubility, thereby increasing apparent reactivity in immunodetection; whether immunogenicity is enhanced at the epitope level requires further validation under isoform‐ and conformation‐matched conditions using cellular and/or animal immunological assays. Overall, the major contribution of this study is the establishment of a generalizable sequence‐reengineering strategy that markedly improves the soluble expression and structural foldability of multi‐epitope proteins while preserving predefined epitope information. In addition, molecular dynamics simulations provide mechanistic insights into how these redesigns confer the observed improvements. In recent years, substantial research efforts have been directed towards the rational design and development of multi‐epitope functional proteins against diverse pathogens—with epitope‐focused vaccine candidates emerging as the most extensively investigated and clinically promising subclass. Notably, such vaccines have demonstrated robust and balanced immunogenicity across multiple delivery modalities, including parenteral, mucosal, and particulate antigen‐display systems, underscoring their versatility and translational potential(Rubio‐Reyes et al. 2017 ; Vasquez et al. 2025 ). However, translating the multi‐epitope antigen strategy into practical diagnostics and control tools for bovine tuberculosis remains hampered by two interrelated bottlenecks: first, empirical antigen selection and epitope concatenation often lack structural and immunological rationale—leading to suboptimal coverage of immunodominant B‐cell epitopes and compromised assay sensitivity; second, multi‐epitope fusion proteins frequently suffer from poor soluble expression in prokaryotic systems, conformational heterogeneity, and inadequate surface accessibility of key epitopes—collectively undermining their performance in serological assays with respect to both analytical sensitivity and inter‐batch reproducibility. For instance, an indirect ELISA assay constructed using the M. bovis antigens CFP10 and ESAT6 demonstrated a sensitivity of 45% and a specificity of 93% (Aagaard et al. 2006 ), both of which fall substantially short of the desired diagnostic standards. The indirect ELISA developed in this study was evaluated against a commercial M. bovis indirect ELISA antibody detection kit. Comparative analysis revealed a positive agreement rate of 100% and a negative agreement rate of 94.84%, with a sensitivity of 96.30% and a specificity of 98.61%. The concordance between the developed MPB70‐83 (ProteinMPNN) indirect ELISA and existing diagnostic methods was assessed using the Kappa ( κ ) statistic. κ values of 0.81, 0.94, and 0.91 were obtained when compared with the tuberculin skin test (TST), an antibody detection strip, and a commercial indirect ELISA kit, respectively, indicating an excellent level of agreement with the established methods. Furthermore, the performance of the assay was further compared with that of the tuberculin skin test (TST) and a commercially available antibody detection strip. The findings demonstrated that the indirect ELISA developed in this study exhibited high specificity (97.8%) and was able to detect seven TST‐confirmed positive samples that were missed by the commercial antibody detection kit, thereby suggesting its potential for earlier identification of M. bovis infection. In conclusion, leveraging the soluble expression of the 70–83 ProteinMPNN multi‐epitope fusion protein, we successfully developed an indirect ELISA for M. bovis antibody detection, which demonstrated excellent specificity, sensitivity, and reproducibility. This assay exhibited distinct advantages in the detection of early‐stage bovine tuberculosis infection, offering a novel diagnostic approach for the early detection of bTB and demonstrating great potential for broad application in the field. In summary, by leveraging ProteinMPNN optimization, the 70–83 multi‐epitope fusion protein was efficiently and solubly expressed, enabling the development of an indirect ELISA for bovine tuberculosis antibody detection that exhibited high specificity, sensitivity, and reproducibility. This assay demonstrated marked advantages in the detection of early‐stage infections and holds broad potential for application. Future work will focus on validating its stability and adaptability using larger sample sizes across diverse geographical settings, as well as exploring its feasibility for on‐site rapid testing and cross‐species detection. 5. Conclusion The aggregation and insolubility of recombinant antigens have severely limited the sensitivity of serological diagnostics for bovine tuberculosis. By leveraging ProteinMPNN for targeted sequence optimization, this study presents a transformative solution that ensures the soluble expression of the MPB70–83 fusion protein while preserving its critical immunodominant epitopes. Multi‐dimensional validation confirmed that the optimized antigen possesses superior structural rigidity and thermodynamic stability, which are essential for consistent assay performance. Consequently, the newly developed indirect ELISA demonstrated high concordance (\kappa > 0.90 κ > 0.90) with gold‐standard methods and exhibited enhanced sensitivity in identifying early‐stage infections often missed by conventional assays. In summary, this study offers a highly reliable, cost‐effective diagnostic candidate for M. bovis surveillance and validates deep learning‐based protein design as a powerful strategy to upgrade next‐generation veterinary diagnostics. Author Contributions Ying‐Ying Xie: data curation, investigation, methodology. Xin‐Miao Liu: methodology. Jie Li: formal analysis. Wen‐Hao Wang: writing – review and editing, writing – original draft, formal analysis, resources, visualization, validation, methodology, investigation, conceptualization. Yijian Liu: methodology, investigation. Zi‐qing Wang: formal analysis, investigation. Rui‐Shuang Li: formal analysis. Shuang‐Shuang Guo: methodology. Ren‐ge Li: methodology. Guo‐Dong Song: formal analysis, methodology. Yin‐juan Song: formal analysis, supervision. Na Li: formal analysis. Shengli Chen: funding acquisition, project administration. Fuying Zheng: resources, supervision, formal analysis, project administration, funding acquisition. Yuefeng Chu: resources, supervision, formal analysis, project administration, funding acquisition. Funding This work was supported by the Joint Scientific Research Fund Major Project of Gansu Province (25JRRA1085), the Major Science and Technology Project of Gansu Province (24ZD13NA008, 23ZDNA007), the Innovation Program of the Chinese Academy of Agricultural Sciences (CAAS‐ZDRW‐202410), and the Key Research and Development Program of Xinjiang (2025B02022‐3). Disclosure The authors have nothing to report. Conflicts of Interest The authors declare no conflicts of interest. Supporting information Figure S1: The sequence comparison results before and after optimization. MBT2-19-e70348-s001.png (35.8KB, png) Acknowledgements This work was supported by the Joint Scientific Research Fund Major Project of Gansu Province (25JRRA1085), the Major Science and Technology Project of Gansu Province (24ZD13NA008, 23ZDNA007), the Innovation Program of the Chinese Academy of Agricultural Sciences (CAAS‐ZDRW‐202410), and the Key Research and Development Program of Xinjiang (2025B02022‐3). We would like to thank all the participants who took part in this study. All authors gratefully acknowledge WeMol, Wecomput Technology Co. Ltd. ( https://wemol.wecomput.com , Room 2104, No. 9, North 4th Ring West Road, Haidian District, Beijing, China) for providing computational server support used in this study. The authors also wish to express sincere gratitude to Wenhao's wife, Liu Fule, for her unwavering support throughout this work. Contributor Information Fuying Zheng, Email: [email protected]. Yuefeng Chu, Email: [email protected]. Data Availability Statement The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding authors. References Aagaard, C. , Govaerts M., Meikle V., et al. 2006. “Optimizing Antigen Cocktails for Detection of Mycobacterium bovis in Herds With Different Prevalences of Bovine Tuberculosis: ESAT6‐CFP10 Mixture Shows Optimal Sensitivity and Specificity.” Journal of Clinical Microbiology 44: 4326–4335. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Abraham, M. J. , Murtola T., Schulz R., et al. 2015. “GROMACS: High Performance Molecular Simulations Through Multi‐Level Parallelism From Laptops to Supercomputers.” SoftwareX 1: 19–25. [ Google Scholar ] Anandarao, R. , Swaminathan S., Fernando S., Jana A. M., and Khanna N.. 2006. “Recombinant Multiepitope Protein for Early Detection of Dengue Infections.” Clinical and Vaccine Immunology: CVI 13: 59–67. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Bagcchi, S.
2023. “WHO'S Global Tuberculosis Report 2022.” Lancet Microbe 4: e20. [ DOI ] [ PubMed ] [ Google Scholar ] Borham, M. , Oreiby A., El‐Gedawy A., et al. 2022. “Review on Bovine Tuberculosis: An Emerging Disease Associated With Multidrug‐Resistant Mycobacterium Species.” Pathogens 11: 715. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Brorson, K. , Krejci S., Lee K., Hamilton E., Stein K., and Xu Y.. 2003. “Bracketed Generic Inactivation of Rodent Retroviruses by Low pH Treatment for Monoclonal Antibodies and Recombinant Proteins.” Biotechnology and Bioengineering 82: 321–329. [ DOI ] [ PubMed ] [ Google Scholar ] Buddle, B. M. , Wilson T., Denis M., et al. 2010. “Sensitivity, Specificity, and Confounding Factors of Novel Serological Tests Used for the Rapid Diagnosis of Bovine Tuberculosis in Farmed Red Deer ( Cervus elaphus ).” Clinical and Vaccine Immunology 17: 626–630. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Camussone, C. , Gonzalez V., Belluzo M. S., et al. 2009. “Comparison of Recombinant Trypanosoma cruzi Peptide Mixtures Versus Multiepitope Chimeric Proteins as Sensitizing Antigens for Immunodiagnosis.” Clinical and Vaccine Immunology 16: 899–905. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Carr, M. D. , Bloemink M. J., Dentten E., et al. 2003. “Solution Structure of the Mycobacterium tuberculosis Complex Protein MPB70: From Tuberculosis Pathogenesis to Inherited Human Corneal Desease.” Journal of Biological Chemistry 278: 43736–43743. [ DOI ] [ PubMed ] [ Google Scholar ] Collins, A. B. , Floyd S., Gordon S. V., and More S. J.. 2022. “Prevalence of Mycobacterium bovis in Milk on Dairy Cattle Farms: An International Systematic Literature Review and Meta‐Analysis.” Tuberculosis 132: 102166. [ DOI ] [ PubMed ] [ Google Scholar ] Dauparas, J. , Anishchenko I., Bennett N., et al. 2022. “Robust Deep Learning‐Based Protein Sequence Design Using ProteinMPNN.” Science (New York, N.Y.) 378: 49–56. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] de Souza, M. Q. , Galdino A. S., Dos Santos J. C., et al. 2013. “A Recombinant Multiepitope Protein for Hepatitis B Diagnosis.” BioMed Research International 2013: 148317. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Enayatkhani, M. , Hasaniazad M., Faezi S., et al. 2021. “Reverse Vaccinology Approach to Design a Novel Multi‐Epitope Vaccine Candidate Against COVID‐19: An In Silico Study.” Journal of Biomolecular Structure & Dynamics 39: 2857–2872. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Faria, A. R. , de Castro Veloso L., Coura‐Vital W., et al. 2015. “Novel Recombinant Multiepitope Proteins for the Diagnosis of Asymptomatic Leishmania infantum ‐Infected Dogs.” PLoS Neglected Tropical Diseases 9: e3429. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Fast, J. L. , Cordes A. A., Carpenter J. F., and Randolph T. W.. 2009. “Physical Instability of a Therapeutic Fc Fusion Protein: Domain Contributions to Conformational and Colloidal Stability.” Biochemistry‐Us 48: 11724–11736. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Firdessa, R. , Tschopp R., Wubete A., et al. 2012. “High Prevalence of Bovine Tuberculosis in Dairy Cattle in Central Ethiopia: Implications for the Dairy Industry and Public Health.” PLoS One 7: e52851. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Freigassner, M. , Pichler H., and Glieder A.. 2009. “Tuning Microbial Hosts for Membrane Protein Production.” Microbial Cell Factories 8: 69. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Galdino, A. S. , Santos J. C., Souza M. Q., et al. 2016. “A Novel Structurally Stable Multiepitope Protein for Detection of HCV.” Hepatology Research and Treatment 2016: 6592143. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Garnier, T. , Eiglmeier K., Camus J. C., et al. 2003. “The Complete Genome Sequence of Mycobacterium bovis .” Proceedings of the National Academy of Sciences of the United States of America 100: 7877–7882. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Goncalves, A. A. M. , Ribeiro A. J., Resende C. A. A., et al. 2024. “Recombinant Multiepitope Proteins Expressed in Escherichia coli Cells and Their Potential for Immunodiagnosis.” Microbial Cell Factories 23: 145. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Green, L. R. , Jones C. C., Sherwood A. L., et al. 2009. “Single‐Antigen Serological Testing for Bovine Tuberculosis.” Clinical and Vaccine Immunology 16: 1309–1313. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Griffin, J. , Cross J. P., Chinn D. N., Rodgers C. R., and Buchan G. S.. 1994. “Diagnosis of Tuberculosis due to Mycobacterium bovis in New Zealand Red Deer ( Cervus elaphus ) Using a Composite Blood Test and Antibody Assays.” New Zealand Veterinary Journal 42: 173–179. [ DOI ] [ PubMed ] [ Google Scholar ] Hebditch, M. , Carballo‐Amador M. A., Charonis S., Curtis R., and Warwicker J.. 2017. “Protein‐Sol: A Web Tool for Predicting Protein Solubility From Sequence.” Bioinformatics 33: 3098–3100. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Jespersen, M. C. , Peters B., Nielsen M., and Marcatili P.. 2017. “BepiPred‐2.0: Improving Sequence‐Based B‐Cell Epitope Prediction Using Conformational Epitopes.” Nucleic Acids Research 45: W24–W29. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Jiang, X. , Wang C., Wang C., Zhang P., and He Z.. 2006. “Cloning and Expression of Mycobacterium bovis Secreted Protein MPB83 in Escherichia coli .” Journal of Biochemistry and Molecular Biology 39: 22–25. [ DOI ] [ PubMed ] [ Google Scholar ] Jin, Y. , Lu K., Zhou W., et al. 2010. “Comparison of Recombinant Proteins From Schistosoma japonicum for Schistosomiasis Diagnosis.” Clinical and Vaccine Immunology: CVI 17: 476–480. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Jones, C. H. , Androsavich J. R., So N., et al. 2024. “Breaking the Mold With RNA – A “RNAissance” of Life Science.” NPJ Genomic Medicine 9: 2. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Kanagavel, M. , Shanmughapriya S., Anbarasu K., and Natarajaseenivasan K.. 2014. “B‐Cell‐Specific Peptides of leptospira interrogans LigA for Diagnosis of Patients With Acute Leptospirosis.” Clinical and Vaccine Immunology 21: 354–359. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Kesidis, A. , Depping P., Lode A., et al. 2020. “Expression of Eukaryotic Membrane Proteins in Eukaryotic and Prokaryotic Hosts.” Methods 180: 3–18. [ DOI ] [ PubMed ] [ Google Scholar ] Koo, H. C. , Park Y. H., Ahn J., et al. 2005. “Use of rMPB70 Protein and ESAT‐6 Peptide as Antigens for Comparison of the Enzyme‐Linked Immunosorbent, Immunochromatographic, and Latex Bead Agglutination Assays for Serodiagnosis of Bovine Tuberculosis.” Journal of Clinical Microbiology 43: 4498–4506. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Kumar, N. , Tripathi S., Sharma N., Patiyal S., Devi N. L., and Raghava G. P. S.. 2024. “A Method for Predicting Linear and Conformational B‐Cell Epitopes in an Antigen From Its Primary Sequence.” Computers in Biology and Medicine 170: 108083. [ DOI ] [ PubMed ] [ Google Scholar ] Kunst, H.
2006. “Diagnosis of Latent Tuberculosis Infection: The Potential Role of New Technologies.” Respiratory Medicine 100: 2098–2106. [ DOI ] [ PubMed ] [ Google Scholar ] Kyro, G. W. , Qiu T., and Batista V. S.. 2025. “A Model‐Centric Review of Deep Learning for Protein Design.” arXiv preprint arXiv:2502.19173. 10.48550/arXiv.2502.19173. [ DOI ] Kyte, J. , and Doolittle R. F.. 1982. “A Simple Method for Displaying the Hydropathic Character of a Protein.” Journal of Molecular Biology 157: 105–132. [ DOI ] [ PubMed ] [ Google Scholar ] Leenaars, M. , and Hendriksen C. F. M.. 2005. “Critical Steps in the Production of Polyclonal and Monoclonal Antibodies: Evaluation and Recommendations.” ILAR Journal 46: 269–279. [ DOI ] [ PubMed ] [ Google Scholar ] Lv, C. , Hong Y., Fu Z., et al. 2016. “Evaluation of Recombinant Multi‐Epitope Proteins for Diagnosis of Goat Schistosomiasis by Enzyme‐Linked Immunosorbent Assay.” Parasites & Vectors 9: 135. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Moens, C. , Filée P., Boes A., et al. 2023. “Identification of New Mycobacterium bovis Antigens and Development of a Multiplexed Serological Bead‐Immunoassay for the Diagnosis of Bovine Tuberculosis in Cattle.” PLoS One 18: e0292590. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] O'Brien, A. , Clarke J., Hayton A., et al. 2023. “Diagnostic Accuracy of the Enferplex Bovine Tuberculosis Antibody Test in Cattle Sera.” Scientific Reports 13: 1875. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Passaro, S. , Corso G., Wohlwend J., et al. 2025. “Boltz‐2: Towards Accurate and Efficient Binding Affinity Prediction.” bioRxiv the Preprint Server for Biology. 10.1101/2025.06.14.659707. [ DOI ] Porebski, B. T. , Keleher S., Hollins J. J., et al. 2016. “Smoothing a Rugged Protein Folding Landscape by Sequence‐Based Redesign.” Scientific Reports 6: 33958. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Qin‐Wang, N. , Hong‐Xia L. I., Lin‐Zi Z., et al. 2010. “Detection of Mycobacterium Tuberculosis by ELISA Using 3 Specific Antibodies.” Journal of Tropical Medicine 10, no. 10: 1163–1165. +1186. (Chinese). [ Google Scholar ] Rosano, G. L. , and Ceccarelli E. A.. 2014. “Recombinant Protein Expression in Escherichia coli : Advances and Challenges.” Frontiers in Microbiology 5: 172. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Rouhani, M. , Khodabakhsh F., Norouzian D., Cohan R. A., and Valizadeh V.. 2018. “Molecular Dynamics Simulation for Rational Protein Engineering: Present and Future Prospectus.” Journal of Molecular Graphics & Modelling 84: 43–53. [ DOI ] [ PubMed ] [ Google Scholar ] Rubio‐Reyes, P. , Parlane N. A., Buddle B. M., Wedlock D. N., and Rehm B. H. A.. 2017. “Immunological Properties and Protective Efficacy of a Single Mycobacterial Antigen Displayed on Polyhydroxybutyrate Beads.” Microbial Biotechnology 10: 1434–1440. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Singh, S. M. , and Panda A. K.. 2005. “Solubilization and Refolding of Bacterial Inclusion Body Proteins.” Journal of Bioscience and Bioengineering 99: 303–310. [ DOI ] [ PubMed ] [ Google Scholar ] Taherkhani, R. , Farshadpour F., and Makvandi M.. 2015. “Design and Production of a Multiepitope Construct Derived From Hepatitis E Virus Capsid Protein.” Journal of Medical Virology 87: 1225–1234. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Thomasini, R. L. , Souza H. G. A., Bruna‐Romero O., et al. 2018. “Evaluation of a Recombinant Multiepitope Antigen for Diagnosis of Hepatitis C Virus : A Lower Cost Alternative for Antigen Production.” Journal of Clinical Laboratory Analysis 32: e22410. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Tyurin, A. A. , Kabardaeva K. V., Mustafaev O. N., et al. 2018. “Expression of Soluble Active Interferon αA in Escherichia coli Periplasm by Fusion With Thermostable Lichenase Using the Domain Insertion Approach, Biochemistry .” Biokhimiia 83: 259–269. [ DOI ] [ PubMed ] [ Google Scholar ] Vasquez, R. , Song J. H., Mendoza R. M., et al. 2025. “Oral Immunisation With Non‐GMO Surface Displayed SARS‐CoV‐2 Spike Epitopes on Bacteria‐Like Particles Provokes Robust Humoral and Cellular Immune Responses, and Modulated the Gut Microbiome in Mice.” Microbial Biotechnology 18: e70073. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Associated Data This section collects any data citations, data availability statements, or supplementary materials included in this article. Supplementary Materials Figure S1: The sequence comparison results before and after optimization. MBT2-19-e70348-s001.png (35.8KB, png) Data Availability Statement The original contributions presented in the study are included in the article. Further inquiries can be directed to the corresponding authors. Articles from Microbial Biotechnology are provided here courtesy of Wiley ACTIONS View on publisher site PDF (6.2 MB) Cite Collections Permalink PERMALINK Copy RESOURCES Similar articles Cited by other articles Links to NCBI Databases Cite Copy Download .nbib .nbib Format: AMA APA MLA NLM Add to Collections Create a new collection Add to an existing collection Name your collection * Choose a collection Unable to load your collection due to an error Please try again Add Cancel Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top