Thermostability engineering of microbial transglutaminase using artificial intelligence and investigation of its underlying mechanisms

Abstract

The application of artificial intelligence in enzyme molecular evolution has emerged as a research hotspot. However, applying machine learning to enzyme molecular modification still presents many challenges. In particular, accelerating the integration of machine learning and rational design is one of the important development trends in the field of protein engineering. In this study, we experimentally validated key amino acid mutations (E164L, E164P, S199E, and S199Q) predicted by a lab-developed sparse convolutional neural network to enhance the thermostability of microbial transglutaminase. We further investigated the molecular basis of this enhanced stability using molecular dynamics simulations. Compared with the wild type MTG, the thermal stability of the four mutants was significantly improved, and S199E showed the most remarkable improvement. At 60 °C and 50 °C, the half-lives of S199E were 2.3 times and 5.8 times those of the wild type, respectively, and the enzyme activity was increased by 1.4-fold. Molecular dynamics simulations showed that the binding free energy of S199E was − 28.68 kcal/mol, slightly lower than that of the wild type (− 27.96 kcal/mol). The root mean square deviation and root mean square fluctuation of the S199E mutant were 0.25 nm and 0.0566 nm, respectively, showing no significant changes compared with the wild type. LigPlot analysis indicated that E199 formed one hydrogen bonds with A309 and three salt bridges with H201, which might enhance local stability. These findings indicate that the improved thermal stability of the S199E mutant arises from enhanced local structural stability, not from major changes in overall protein structure, and accounts for its slightly lower binding free energy compared to the wild type.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40643-026-01024-5.

Introduction

Transglutaminase (TG) is a highly efficient protein crosslinker that catalyzes acyl transfer reactions between γ-carboxamide groups of glutamine residues in peptide chains and various acyl acceptors (Chen, et al. 2013). This enzyme is widely distributed in humans, mammals, and microorganisms (Su, et al. 2020). Notably, microbial transglutaminase (MTG) exhibits superior properties for industrial applications, including lower molecular weight, calcium-independent activity, and broad substrate specificity (Gharibzahedi and Chronakis 2018). These characteristics make MTG particularly valuable in food processing, biomedical engineering, and tissue regeneration applications (Aryamanesh, et al. 2025, Dennler, et al. 2014, Korponay-Szabo and Gut 2004). In specialized applications such as site-specific protein conjugation and the construction of homogeneous antibody-drug conjugates (ADCs), MTG’s performance can be significantly enhanced through thermoprocessing techniques like extrusion molding and thermostable microencapsulation (Dennler, et al. 2014, El Alaoui, et al. 2024). These approaches dramatically improve crosslinking efficiency under elevated temperatures. Consequently, engineering MTG variants with enhanced thermostability not only optimizes existing manufacturing processes but also unlocks novel applications in high-temperature bioprocessing.

The predominant strategies for enhancing protein thermostability currently include directed evolution and rational design (Barreto, et al. 2025, Chen, et al. 2022, Wang, et al. 2021, Wei, et al. 2024). Directed evolution simulates natural selection. It uses methods such as error-prone PCR, staggered extension, and DNA shuffling to carry out multiple rounds of mutations on the target gene, and combines high-throughput screening to obtain excellent mutants (Barreto, et al. 2025, Chen, et al. 2022, Wang, et al. 2021, Wei, et al. 2024). Rational design is based on information about protein structure and function. It uses computer tools to analyze the impact of key amino acid changes on the molecular forces within the three-dimensional structure, and constructs a small-scale mutant library through precise modification of a few sites (Barreto, et al. 2025). Researchers have successfully obtained mutants with significantly improved thermal stability using directed evolution and rational design methods. Buettner K et al. prepared thermostable mutants using random mutagenesis, saturation mutagenesis, and DNA shuffling techniques, and obtained a triple mutant (S23V-Y24N-K294L) that showed a 12-fold higher half-life at 60 °C (t_1/2 = 24.3 min) than the wild-type MTG (Buettner, et al. 2012). Böhme et al. also constructed 31 MTG mutants. The most thermostable mutant (S2P-23Y-24 N-H289Y-294 L) had a 19-fold longer a half-life at 60 °C (t_1/2 = 38 min) than wild-type MTG (Böhme, et al. 2020, Suzuki, et al. 2024). Wang et al. reported a thermostable mutant, FRAPD-TGm2 (FRAPD-TGm1-E28T-A265P-A287P), created by modifying the mutant FRAP-TGm1 (S2P, S23V, Y24N, S199A, K294L) (Wang, et al. 2021). The half-life, melting temperature (Tm) and specific activity of FRAPD-TGm2 were 66.9 min (t_1/2 (60 °C)), 67.8 °C and 71.8 U/mg, respectively, a twofold, 2.6 °C and 43.8% increase compared to FRAPDTGm1. Further modification led to the discovery of FRAPTGm2A (FRAPD-TGm2-S116A-S179L) with a half-life at 60 °C of 132.38 min (t_1/2 (60 °C)) and specific activity of 79.15 U/mg: 84% and 21% higher than FRAPD-TGm2, respectively (Yang, et al. 2023).In the previous research conducted by our research team, Mut2 (S2P-S23V-Y24N-R215A-H289Y), with improved enzyme activity and thermal stability, was obtained. Its specific enzyme activity is 39.3 ± 1.6 U/mg, and its thermal stability (t_1/2 at 60 °C) is 27.6 min, representing 2.4-fold and 10.2-fold increases compared to the wild-type MTG, respectively (Song, et al. 2022) .

Although directed evolution and rational design methods play important roles in the field of enzyme engineering, both approaches heavily rely on high-throughput screening experiments (Suzuki, et al. 2024). This not only significantly increases the research workload but may also lead to the failure to effectively identify some mutants with excellent traits due to insufficient screening capacity (Böhme, et al. 2020, Li, et al. 2022). Therefore, screening capability has emerged as a critical technical bottleneck in enzyme engineering, severely constraining the discovery efficiency of advantageous variants and ultimately compromising the overall effectiveness of protein directed evolution (Jing, et al. 2021, Musil 2019, Wan, et al. 2021). The rapid development and widespread application of artificial intelligence (AI) in biological sciences has catalyzed significant progress in AI-assisted thermostability prediction. Researchers have developed numerous advanced predictive algorithms and computational models, providing novel tools and methodologies for protein engineering (Saito, et al. 2021, Wan, et al. 2021). Our research team has previously successfully implemented such models to establish a compact yet high-quality mutant library for MTG thermostability, laying a solid foundation for subsequent investigations.

Building upon our previous research utilizing artificial intelligence (AI) to predict thermostability enhancements in MTG, this study selected four mutant sequences demonstrating significant predicted thermostability improvements. These variants were successfully expressed and purified using Pichia pastoris, followed by a comprehensive evaluation of both enzymatic activity and thermal stability. To better reflect industrial application requirements, particularly for thermoprocessing techniques such as extrusion molding and thermostable microencapsulation, which typically require high-temperature conditions exceeding 50 °C, we specifically designated 50 °C and 60 °C as standard evaluation temperatures for assessing enzymatic thermostability. This systematic investigation aims to provide more accurate theoretical foundations and technical support for the industrial application of transglutaminase.

Materials and methods

Strains, plasmids, and media

Strains and vectors were described in Table S1(Additional file 1: Table S1). E. coli Top10 competent cells and P. pastoris‌ GS115 were both purchased from Invitrogen. The pPICZα-B, pPIC9k-pro, pPICZa-B-mtg, and the strains GS115/pPIC9k-pro and GS115(mtg/pro) were all constructed in our laboratory and stored at − 80 ℃. The E. coli was cultured in Luria-Bertani (LB) medium at 37 °C, whereas the P. pastoris was cultured in buffered glycerol-complex medium (BMGY) or buffered methanol-complex medium (BMMY) at 28–25 °C, respectively (Song, et al. 2024, Song, et al. 2019) .

AI-assisted mutant enzyme design

We employed the Sparse Convolutional Network driven by Self-Attention mechanism (SCSAddG) model developed by Xu et al. to predict the thermostability trends of MTG (Xu, et al. 2026).This model integrates four AAindex encodings (FASG890101, PRAM900101, LEVM760101, and PONP800104), which significantly contribute to thermostability prediction, enabling accurate forecasting of thermal stability variations for unknown proteins within the dataset. Specifically, a prediction output of 1 indicates enhanced thermostability at the mutated site. If all four encodings yield a prediction of 1, the corresponding mutant amino acid is selected as a candidate, thereby constructing an amino acid mutation library for MTG.

Obtaining the mutant gene

Based on AI-predicted thermostability enhancements, four mutant enzymes were selected from the amino acid mutant library for experimental validation. The gene (mtg-wt) encoding MTG-WT was selected from the cDNA of Streptomyces mobaraensis (GenBank accession No. DQ132977). The amino acid sequences of these mutants and their corresponding codon-optimized gene sequences (adapted for P. pastoris expression) are provided in the supplementary materials (Additional file 2). Detailed experimental procedures are shown below: The mutant genes (mtg1, mtg2, mtg3, and mtg4) were artificially synthesized by General Biosystems (Anhui) Co., Ltd.

At the 5’ end of the genes, the pPICZa-B vector sequence (1160–1183 bp), Xho I.

restriction site (ctcgag), and Kex2 peptide endoprotease recognition site (aaaaga) were introduced. At the 3’ end, the pPICZa-B vector sequence (1269–1293 bp), Not I restriction site (gcggccgc), and a 6×His tag sequence (CATCATCACCATCACCAT) were incorporated. The inclusion of the 6×His tag sequence was designed to facilitate the subsequent purification process of the recombinant protein. The primer information used for gene synthesis is detailed in Table S2 (Additional file 1: Table S2). A two-round PCR amplification was performed. The first-round PCR reaction system (50 µL): 2 µL of each primer, 2 µL of template, 10 µL of polymerase buffer, 1 µL of 10 mM dNTP, 1 µL of polymerase, and 32 µL of ddH₂O. Cycling parameters: Pre-denaturation at 96 °C for 5 min; 23 cycles of 95 °C for 25 s (denaturation), 58 °C for 25 s (annealing), and 72 °C for 30 s (extension); Final extension at 72 °C for 1 min. Second-round PCR reaction system (50 µL): 2 µL of each primer, 2 µL of the first-round PCR product, 10 µL of polymerase buffer, 1 µL of 10 mM dNTP, 1 µL of polymerase, and 82 µL of ddH₂O. Cycling parameters: Pre-denaturation at 96 °C for 5 min; 23 cycles of 95 °C for 25 s (denaturation), 58 °C for 25 s (annealing), and 72 °C for 30 s (extension); Final extension at 72 °C for 90 s. The target gene fragment length was 1083 bp.

Construction of expression strains

First, construct the recombinant vectors. The mutant genes and the pPICZa-B vector were double-digested with Xho I and Not I to generate gene fragments with sticky ends and a linearized plasmid, respectively. Subsequently, the digested products were ligated at a 3:1 molar ratio using T4 DNA ligase to construct the recombinant plasmid (pPICZa -mtg). The recombinant plasmid was then transformed into E. coli TOP10 competent cells and plated uniformly on LB solid medium containing 25 µg/mL Zeocin. The plates were incubated overnight at 37 °C until single colonies appeared(Song, et al. 2022). Single colonies were picked for culture, and recombinant plasmids were extracted and verified by double digestion with Xho I and Not I. The constructed recombinants were named Top10/pPICZα-mtg.

Then, construct the expression strain. For specific construction steps, refer to references (Song, et al. 2024, Yokoyama, et al. 2021). The recombinant vectors pPICZα-mtg was extracted from Top10/pPICZα-mtg and linearized using Sac I. Approximately 3000 ng of the linearized plasmid was added to GS115/pPIC9k-pro competent cells (stored in the laboratory) and transferred into an electroporation cuvette. Electroporation was performed at 2 kV with a pulse duration of 5.6 ms. Immediately after electroporation, 1 mL of ice-cold 1 mol/L D-sorbitol solution was added, and the cells were recovered at 28 °C with shaking at 200 rpm for 2 h. The recovered cells were then spread evenly on YPD plates containing 200 µg/mL Zeocin. After single colonies appeared, 20 clones were randomly selected for PCR amplification to confirm successful transformation. The constructed strain was named the yeast expression strain GS115(pro/mtg-mut).

Finally, three verified transformants of the yeast expression strain GS115(pro/mtg) were selected into test tubes containing 2 mL of BMGY medium and cultured at 28 °C with shaking at 200 rpm for 24 h. The cells were then harvested by centrifugation at 3000g for 5 min, and the supernatant was discarded. The pellet was resuspended in 2 mL of BMMY medium (BMGY with 1.0% methanol replacing glycerol) and induced for protein expression at 25 °C with shaking at 200 rpm for 70 h. After induction, the fermentation broth was centrifuged at 12,000 g for 10 min to collect the supernatant. The supernatant was subjected to Western blot analysis using an anti-MTG antibody to confirm protein expression. Additionally, the enzyme activity of the fermentation supernatant was measured (Song, et al. 2019). The positive clone with higher enzyme activity was stored at − 20 °C for further use in shake-flask fermentation and expression studies.

Expression and purification of wild-type and mutant enzymes

From the aforementioned mutants, the expression strain with the highest enzyme activity was selected for shake flask fermentation and purification. 600 µL of seed culture was inoculated into 100 mL of BMGY medium (pH 6.0) and cultured overnight at 28 °C and 200 r/min. After 24 h, the OD₆₀₀ value was measured. When the OD₆₀₀ reached 6, the cells were collected by centrifugation at 3000 g for 5 min. The cell pellet was resuspended in 200 mL of BMMY medium containing 1% methanol (pH 6.0), and the initial OD₆₀₀ was adjusted to 1.2–1.5. The induction expression conditions were set at 25 °C for 72 h, with methanol added every 24 h to a final concentration of 1% (V/V) (Li, et al. 2025). The purification process of the MTG mutant followed the method described in the literature(Song, et al. 2024). The pH of the fermentation supernatant was adjusted to 7.2, and the cell debris was removed by centrifugation at 12,000 g for 30 min, followed by collection of the supernatant. The supernatant was loaded onto a nickel column of the AKTA protein purification system, and the steps of equilibration, loading, re-equilibration, washing, and elution were performed sequentially to collect the eluate. The eluate was dialyzed using 0.2 M.

PBS (pH 7.2) to remove imidazole, and the dialyzed sample was collected. The dialyzed sample was analyzed by non-reducing SDS-PAGE (10% gel concentration), and the target protein was identified using Western blot technology (using Anti-MTG antibody) (Mu, et al. 2018, Song, et al. 2019, Yokoyama, et al. 2021). The expression and purification methods for the wild-type enzyme were the same as those for the mutant enzyme.

Enzymatic characterization

The enzymatic activity of MTG was determined using the hydroxamate colorimetric method, following the procedures described in references (Song, et al. 2019, Yokoyama, et al. 2021). The MTG enzyme activity unit was defined as the amount of enzyme required to catalyze the formation of 1 µmol of glutamic acid-monohydroxamate (hydroxamic acid) from the substrate N-CBZ-Gln-Gly per minute at 37 °C (Mu, et al. 2018, Yokoyama, et al. 2021).The protein concentration was measured using the Bradford method, with bovine serum albumin (BSA) as the standard (Mu, et al. 2018). All experiments were performed in triplicate to ensure the reliability of the data.

The methods for determining the core kinetic parameters of wild-type and mutant enzymes are as follows: Set different substrate concentration gradients within the range of 1–100 mmol/L, measure the MTG enzyme activity, plot a graph with substrate concentration [S] on the abscissa and reaction rate [V] on the ordinate, perform non-linear fitting using Origin 8.0 software to obtain the corresponding K_m and V_max values, and calculate k_cat using the formula kcat = Vmax/[E₀], where [E₀] is the initial enzyme concentration.

To identify the optimal reaction temperature for both wild-type and mutant MTG enzymes, their activities were initially measured at 20, 30, 37, 50, 60, and 70 °C to establish a preliminary temperature range. Subsequently, a finer gradient of 5 °C intervals was applied for precise determination. In the experiment, the final concentration of the substrate N-CBZ-Gln-Gly was set at 30 mmol/L, and the pH was adjusted to 6.0 using 20 mmol/L Tris buffer. By plotting the temperature-enzyme activity curve (with temperature as the x-axis and enzyme activity as the y-axis), the optimal reaction temperature was ultimately determined.

To further investigate the thermostability of the mutant enzyme, we studied its half-life (t_1/2) at 50 °C and 60 °C. The purified mutant and wild-type enzymes were diluted to the same concentration and incubated in a water bath at 50–60 °C for varying durations. After incubation, the samples were immediately cooled on ice for 10 min, followed by measurement of residual enzyme activity (with N-CBZ-Gln-Gly at a final concentration of 30 mmol/L). The relationship between incubation time (at 50–60 °C) and residual activity was plotted, with incubation time as the x-axis and residual activity as the y-axis. The first-order inactivation rate constant (k_d) at each temperature was obtained by linear fitting of the data. Subsequently, the half-life (t_1/2) was calculated using the equation t_1/2 = Ln2/k_d (Song, et al. 2022).

Analysis of the molecular mechanism of enhanced thermostability in mutant enzymes

Homology modeling was performed using the SWISS-MODEL online server, with the crystal structure of microbial transglutaminase (PDB ID: 3IU0A) derived from.

S. mobaraensis as the template, to predict the three-dimensional structure of the mutant MTGs. Further structural analysis was conducted using LigPlot software to investigate molecular interactions.

Molecular dynamics (MD) simulations were performed using the Gromacs 2021.5 open-source software package (Spoel, et al. 2005). The protein was protonated at pH 7.0, and the simulation system was set up under periodic boundary conditions (PBC) (Gao, et al. 2017). The temperature was maintained at 323 K (50 ℃), and the pressure was set to 1 bar in all systems (Gao, et al. 2017). After initial system construction, energy minimization was carried out for all atoms using the steepest descent method to eliminate steric clashes. Subsequently, the system was equilibrated under two conditions: NVT ensemble (constant number of particles, volume, and temperature) for 1000 ps and NPT ensemble (constant number of particles, pressure, and temperature) for 1000 ps. Following equilibration, a 100 ns production MD simulation was conducted with a time step of 2 fs, and trajectory snapshots were saved every 10 ps (Case, et al. 2010, Sarma, et al. 2024) Covalent bond lengths were constrained using the LINCS algorithm, while long-range electrostatic interactions were treated with the Particle Mesh Ewald (PME) method (Mark and Nilsson 2001, Sprenger, et al. 2015). After simulations, the gmx module was employed to calculate key properties, including: Root Mean Square Deviation (RMSD) and Root Mean Square Fluctuation (RMSF) (Britton, et al. 2024, Valdes-Tresanco, et al. 2021). Additionally, the MM/GBSA (Molecular Mechanics/Generalized Born Surface Area) model was applied to estimate the binding free energy ΔG bind between the substrate and protein (Bill, et al. 2012).

Statistical analysis

All measurements were conducted in triplicate, and the data were analyzed using IBM SPSS Statistics 24. The results are presented as the mean ± standard deviation (SD), with statistical significance determined by a p-value < 0.05.

Results

Identification of mutation sites and mutant amino acids

The SCSAddG model developed by Xu et al. was used to predict changes in the thermal stability of MTG and the results are shown in Table 1 The results revealed that nine mutation sites significantly influenced the enzyme’s thermostability, ranked in descending order of impact as follows: E164, S199, S2, H289, K294, Y24, P76, G97, and N297. Among these, mutations at E164, S199, S2, Y24, H289, and K294 have been previously reported in the literature. Specifically, the mutations E164L, S199A, S2P, S2M, S2Y, Y24N, H289Y, and K294L were shown to significantly enhance the enzyme’s thermostability. The predictions of our model align with these existing reports, further validating the accuracy of the model. Additionally, the study predicted novel beneficial mutation sites and their corresponding mutant amino acids, including P76, G97, and N297, with specific mutation types detailed in Table 1. Through a comprehensive analysis of the data in Table 3 and in-depth discussions with biologists, we identified E164 and S199 as key amino acid sites significantly impacting enzyme thermostability, along with their respective mutants. After rigorous evaluation, the following key mutation sites and corresponding mutant amino acids were selected for experimental validation: E164L, E164P, S199E, and S199Q. The amino acid sequences of the MTG mutants and their optimized genes are provided in the supplementary material (Additional file 2).

Table 1.

The ranking of the mutant sites and their influence degrees predicted by the model (from high to low)

Mutation position	Original amino acids	Mutated amino acid
		Literature reports Model prediction
164 199 2 289 294 24	E S S H K Y	L A P, M, Y Y L N	F, A, C, I, M, N, P, S C, E, K, Q A, G, N, Q K, M, N, P, T A, D, L, F, K, M, Q, S E, V
76 97	P G	– –	L, R, F, N, V C
297	N	–	F, R

Table 3.

Enzymatic properties of WT and its mutants

Mutation	Specific activity (U/mg)	t1/2 (50℃) (min)	t1/2 (60℃) (min)	km (mmol/L)	kcat (s− 1)
None E164L E164P S199E S199Q	15.7 ± 0.7 26.7 ± 0.8 21.6 ± 0.7 38.3 ± 1.1 33.5 ± 1.4	36.5 90.2 79.7 210.7 187.2	2.1 3.8 2.9 4.8 4.1	51.52 ± 1.85 46.79 ± 1.96 48.87 ± 2.05 42.35 ± 1.93 44.13 ± 1.39	0.24 ± 0.039 0.49 ± 0.037 0.38 ± 0.074 0.69 ± 0.086 0.61 ± 0.059

Construction of recombinant vectors, transformation, and high-expression screening

In previous studies, we successfully constructed the co-expression strain GS115/pro-mtg), which contains the leader peptide gene (pro) and the mature peptide gene (mtg) of microbial transglutaminase (laboratory stock) (Song, et al. 2022). Through this construction strategy, we achieved the direct expression of active microbial transglutaminase in the P. pastoris expression system, with the specific activity of the wild-type enzyme reaching 15.7 U/mg(Song, et al. 2022). This study continues to adopt this construction strategy.

Four recombinant plasmids (pPICzaB-mtg1, pPICzaB-mtg2, pPICzaB-mtg3, and pPICzaB-mtg4) were extracted and subjected to double digestion using Xho I and Not I restriction enzymes. Agarose gel electrophoresis confirmed two DNA fragments matching the expected sizes (1034 bp and 3512 bp), verifying successful construction of the expression vectors containing mutant genes (mtg1, mtg2, mtg3, or mtg4) (Additional File 3: Figures S1a, S2a, S3a, S4a). The linearized vectors were introduced into GS115/pPIC9K-pro (laboratory stock) via electroporation. transformants were selected on YPD plates supplemented with 100 µg/mL Zeocin and incubated at 28°C for 2–3 days. Twenty single colonies were randomly picked for colony PCR screening. PCR amplification using primers targeting the target gene and 3’-AOX region yielded fragments matching the predicted sizes (576 bp, 577 bp, 578 bp, and 576 bp), confirming successful integration of the mutant genes into the host genome (Additional file 3: Figure S1b, c; S2b, c, S3b, c and S4b, c). All tested clones were positive transformants, indicating high transformation efficiency.

Three PCR-verified positive clones were randomly selected and cultured in 2 mL of BMGY medium at 28℃ until reaching OD₆₀₀ = 2–6. Cultures were then transferred to BMMY induction medium (containing 1% methanol) and incubated at 25 °C for 70 h. Western blot analysis confirmed successful expression in all tested clones (detailed results not shown). Clones demonstrating higher expression levels were selected for 5 mL scale-up cultures under identical induction conditions. Enzyme activity assays (Table 2, Additional File 3: Figures S5) revealed all four mutants exhibited significantly higher activity than the wild-type enzyme (0.23 U/mL)(Song, et al. 2022, Song, et al. 2019). Mut3 (#2) has highest activity (0.62 U/mL), ~ 2.67-fold increase over wild-type. Mut4 (#1) has an activity of 0.54 U/mL, ~ 2.47-fold increase. Mut1 (#1) has an activity of 0.41 U/mL, ~ 1.78-fold increase. Mut2 (#2) has an activity of 0.32 U/mL, ~ 1.34-fold increase. Based on superior expression levels and enzymatic performance, Mut1 (#1), Mut2 (#2), Mut3 (#2), and Mut4 (#1) were selected as candidate strains for subsequent flask-scale fermentation experiments.

Table 2.

Enzyme activity of mutant MTG in 5mL culture system

Name	Mutation	Enzyme activity (U/mL)
		1	2	3
Mut1 Mut2 Mut3 Mut4	E164L E164P S199E S199Q	0.41 0.29 0.51 0.54	0.37 0.32 0.62 0.49	0.35 0.21 0.57 0.45

The data of enzyme activity are presented as mean ± SD of triplicate observations.

Expression and purification of mutant enzymes

From each mutant, clone with the highest enzyme activity was selected (Mut1 #1, Mut2 #2, Mut3 #2, and Mut4 #1) for shake-flask fermentation (200 mL culture system). The SDS-PAGE results of the purified proteins are shown in Fig. 1 (with the wild-type enzyme as the control). The results revealed that both the wild-type and mutant enzymes exhibited three specific protein bands: one non-glycosylated band and two glycosylated bands (the non-glycosylated band was active), with sizes ranging from 34 kDa to 43 kDa. This observation is consistent with previous studies and literature reports showing 1 to 3 bands (song 2021, Song, et al. 2019). The protein weights of the four purified mutant enzymes were 3.12, 2.96, 3.23, and 3.35 mg, respectively (the wild-type enzyme was 2.94 mg) (Additional file 4: Table S3).

Fig. 1.

SDS-PAGE electrophoretogram of purified mutant enzymes (loading volume 10µL). M: Protein molecular quality standards, Control: the commercial transglutaminase from E. coil.

Characterization of the enzymatic properties of mutant enzymes

To assess the enzymatic activity of mutant MTG in comparison to the wild-type enzyme, we employed the hydroxamate colorimetric assay for activity determination and the Bradford method for protein quantification, using bovine serum albumin (BSA) as the standard protein (Mu, et al. 2018, song 2021). As shown in Table 3, the specific activity of purified wild-type MTG was 15.7 ± 0.7 U/mg, which was lower than that of all four mutant enzymes (Table 3). Among the mutants, Mut3 exhibited the highest specific activity (38.3 ± 1.1 U/mg), approximately 2.4-fold that of the wild-type enzyme. Mut4 followed with a specific activity of 33.5 ± 1.4 U/mg (~ 2.1-fold increase), while Mut1 (26.7 ± 0.8 U/mg, ~ 1.7-fold) and Mut2 (21.6 ± 0.7 U/mg, ~ 1.3-fold) showed relatively lower but still significant enhancements compared to the wild-type enzyme (Table 3). These results demonstrate that the single-point mutation at S199 significantly improved the catalytic efficiency of MTG.

Using N-CBZ-Gln-Gly as the substrate, the Michaelis-Menten kinetic constants of WT and mutant enzymes were determined under the conditions of pH 6.0 and 37 °C. The kinetic parameters Km and kcat are shown in Table 3. As can be seen from Table 3, the Km values of S199E and S199Q were 42.35 mmol/L and 44.13 mmol/L, respectively, which were 17.8% and 14.3% lower than that of WT (51.52 mmol/L), indicating that these mutant enzymes had a higher affinity for the substrate, with S199E exhibiting the highest affinity. The Km values of E164L (46.79 mmol/L) and E164P (48.87 mmol/L) were lower than that of WT (51.52 mmol/L), indicating a slightly improved substrate affinity. The kcat values of the four mutant enzymes were higher than that of WT (0.24 s⁻¹), indicating that the number of substrate molecules converted per unit time by each catalytic center of these mutant enzymes was significantly higher than that of the wild-type enzyme. The kcat value of E164L was 0.69 s⁻¹, which was the highest compared to WT.

Assay of kinetic constants and specific activity were carried out as described in Materials and Methods. Data were expressed as mean ± SD (n = 3).

Thermal stability is a critical property of enzymes. To evaluate this, we investigated the optimal reaction temperature and thermal inactivation half-life (t_1/2) of mutant MTG enzymes at 60 °C and 50 °C. As shown in Fig. 2, all four mutant MTG enzymes exhibited maximum activity at 55 °C, whereas the wild-type MTG showed an optimum temperature of 50 °C, consistent with previous reports (Li, et al. 2024). This indicates that the mutant enzymes are less thermosensitive than the wild-type MTG. The wild-type MTG displayed a t_1/2 of 2.1 min at 60 °C and 36.5 min at 50 °C. In contrast, all mutants exhibited significantly improved thermostability. Among them, Mut3 exhibited the highest t_1/2 (60 °C) and t_1/2 (50 °C), which were 2.2-fold and 5.7-fold higher than those of wild-type MTG, respectively; followed by Mut4, with half-lives t_1/2 (60 °C) and t_1/2 (50 °C) that were 1.9-fold and 5.2-fold higher than those of wild-type MTG, respectively. Mut1 and Mut2 also showed enhanced stability, though to a lesser extent than Mut3 and Mut4.

Fig. 2.

Detection of thermal stability of MTG-WT and its mutant enzymes. a Relative enzyme activity of mutant enzymes at different reaction temperatures. b Relative activity of mutant enzymes at 50 °C. c Relative activity of mutant enzymes at 60 °C

Molecular mechanism analysis of enhanced mutant enzymes’ thermostability

To gain deeper insights into the interactions between mutated amino acids and their neighboring residues in the mutant enzymes, a detailed analysis of the Mutant MTG structures obtained through homology modeling was conducted using LigPlot software (Farahdina, et al. 2024, Laskowski, et al. 2011). For Mut3 (S199E), the substitution of the polar, uncharged surface residue S199 with the negatively charged acidic residue E199 resulted in notable structural enhancements. The new mutation site formed one new hydrogen bonds with A309, and its side-chain carboxyl group also established three ionic bonds (or salt bridges) with the imidazole ring of H201 (Fig. 3a, b,d, e). These interactions significantly improved the structural stability of this region. For Mut1 (E164L), the mutated residue L164 formed only one hydrogen bond with the adjacent L161, and no significant changes in chemical bond types were observed compared to the original residue E164 (Additional file 5: Figure S6a, b,d, e). For Mut2 (E164P), the mutated residue P164 failed to establish any interactions with surrounding amino acids, suggesting this mutation may not have significantly influenced the structure or function (Additional file 5: Figure S6 a, c,d, f). In Mut4 (S199Q), in addition to forming two hydrogen bonds with G257, Q199’s adjacent amino acid K200 forms one hydrogen bond each with D237, N239, Y256, and A309 through its side chain. Moreover, K200 also forms an ionic bond with D237 (Fig. 3a, c,d, f). These chemical bonds make the local structure of Mut4 more stable than that of S199.

Fig. 3.

Analysis of the interactions formed by mutant amino acids (using the wild type as a control). a,b, c Homology simulation on the structure of MTG-WT and variants (E199 and Q199) (Dark blue represents the catalytic triad of the enzyme -C64, D255, and H274; red represents the amino acid at position 199; and bright yellow indicates the amino acids that interact with the amino acid at position 199.). d, e, f The interactions between the amino acid at position 199 and the surrounding amino acids (The green lines represent hydrogen bond interactions between amino acids, and the red lines represent ionic bond interactions between amino acids)

To further investigate the molecular mechanism underlying the enhanced thermostability of the mutant, this study conducted a preliminary structural analysis of the E199 mutant. Surface electrostatic potential analysis revealed that the wild-type enzyme exhibited white or light blue potential distribution near the S199 site, whereas this region displayed a distinct shift to red (indicating increased negative potential) upon mutation to E199 (Fig. 4a, b). These results suggest that the mutation-induced increase in negative electrostatic potential may be a key factor contributing to the improved thermal stability of the enzyme. Additionally, combined with the aforementioned structural analysis, this study further elucidated the specific effects of the mutation on the local microstructure of the protein. The findings indicate that the enhanced local interactions and the formation of new salt bridges in the mutant protein are likely the primary reasons for the superior thermostability of the E199 mutant compared to the wild-type enzyme.

Fig. 4.

Structural analysis of the wild-type and mutant S199E. a, b Surface electrostatic potential maps for the wild-type and mutant S199E protein show positive, negative, and neutral values represented by blue, red, and white shading, respectively. c, d Comparison plots of RMSD and RMSF for the wild type and the mutant S199E, respectively

To further elucidate the mechanism underlying the enhanced thermostability of the E199 mutant, this study employed molecular dynamics (MD) simulations to systematically analyze the binding free energy, root-mean-square deviation (RMSD), and root-mean-square fluctuation (RMSF) values of both the wild-type (S199) and mutant (E199) enzymes at 323 K (Mark and Nilsson 2001, Sprenger, et al. 2015, Valdes-Tresanco, et al. 2021). The results revealed that the E199 mutant exhibited a marginally more favorable binding free energy (− 28.68 kcal/mol) compared to the wild type S199 (− 27.96 kcal/mol) during the final stabilized 10 ns trajectory. To further investigate the impact of the mutation on local residues and the global protein structure, this study conducted detailed analyses of root-mean-square deviation (RMSD) and root-mean-square fluctuation (RMSF) values for both wild-type and mutant proteins at 323 K using molecular dynamics (MD) simulations. At the 100 ns simulation time point, the mutant (E199) exhibited an RMSD value of 0.25 nm, showing no significant deviation from the wild-type (S199) RMSD value of 0.20 nm (Fig. 4c). Further analysis of residue-specific RMSF values throughout the 100 ns simulation revealed that the global RMSF of the mutant (0.0566 nm) remained comparable to that of the wild-type (0.0570 nm) (Fig. 4d). Furthermore, analysis of secondary structure evolution over simulation time revealed no significant structural transitions throughout the entire process (Fig. 5). The regions with slightly more obvious secondary structure changes are marked with red boxes, mainly showing transitions between α-helices and turns (Fig. 5a, b). The results of molecular dynamics simulations show that, compared with the wild-type enzyme, the mutant has a slightly lower binding free energy and slightly higher stability, with no significant differences in RMSD and RMSF values (Fig. 4c, d). This indicates that the single-point mutation has a minor impact on the overall structure and may mainly affect local stability. This speculation was further confirmed by homology modeling analysis of the mutant enzymes (Mutant MTGs). E199 is located on a β-sheet of the protein. Since β-sheets are inherently stable, this mutation does not cause obvious changes in the overall structure.

Fig. 5.

Analysis of the secondary structure. a, b There are differences in the overall secondary structures between the wild type and the mutant S199E. The regions where the secondary structures differ are highlighted in red boxes. c, d Differences in the secondary structures of the wild type and the mutant S199E during the simulated 100 ns

The molecular dynamics simulation results further indicate that the enhanced thermal stability is not caused by changes in the protein’s spatial structure, but rather by the formation of two new hydrogen bonds between the negatively charged acidic amino acid E199 and Y310 on the protein surface (Fig. 3d, f). Additionally, the carboxyl group of its side chain forms three ionic bonds (or salt bridges) with the imidazole ring of H201, thereby significantly improving the structural stability of this region.

Discussion

Rational design to rapidly and effectively enhance enzyme thermostability is a research hotspot in the field of protein engineering. It not only provides excellent new enzyme sources for industrial production but also helps analyze the relationship between protein structure and properties (Chen, et al. 2022, Grindel, et al. 2022, Kang, et al. 2021). Although researchers have attempted to improve enzyme thermostability through rational design, the results still fall short of fully meeting industrial requirements (Song, et al. 2022).

This study employed the SCSAddG model to assist in the thermostability modification of microbial transglutaminase (MTG). The model was used to predict potential mutation sites in the amino acid sequence of MTG, identifying nine key sites that significantly influence its thermostability. Ranked by their predicted impact from high to low, these sites are: E164, S199, S2, H289, K294, Y24, P76, G97, and N297. To experimentally validate the model’s predictions, we selected the top two most impactful sites (E164 and S199) and constructed four corresponding mutants: E164L, E164P, S199E, and S199Q. Compared with the wild-type enzyme (MTG-WT), all four mutants exhibited higher thermal stability. The order of their thermal stability from high to low is as follows: Mut3 (S199E) > Mut4 (S199Q) > Mut1 (E164L) > Mut2 (E164P), and their specific enzyme activities and kinetic parameters also showed the same trend. Specifically, the t_1/2 values at 60 °C and 50 °C for Mut3 were 2.2 times and 5.7 times those of the wild-type MTG, respectively, and its specific enzyme activity increased by 1.4-fold. Mut4 followed, with half-lives at 60 °C and 50 °C that were 1.9 times and 5.2 times those of the wild-type MTG, respectively. Mut1 and Mut2 showed slightly less improvement in thermal stability compared to Mut3 and Mut4. Among all the single-site mutations (S2, S23, K269, H289, K294) reported in the existing literature, K294L exhibits the highest thermal stability (t_1/2 (50 °C) = 188.9 min, t_1/2 (60 °C) = 4.6 min). Compared with K294L, Mut3 (S199E) in this study shows higher thermal stability; Mut4 (S199Q) has a half-life similar to that of K294L at 50 °C, while its half-life at 60 °C is slightly lower. Compared with other mutants (S2P, S2Y, S2M, S23L, S23V, K269E, H289Y, K294I), Mut3 (S199E), Mut4 (S199Q), and Mut1 (E164L) all exhibit better thermal stability (Böhme, et al. 2020, Buettner, et al. 2012, Yokoyama, et al. 2021).

To investigate the molecular mechanism underlying the enhanced thermal stability of the mutant Mut3, this study utilized LigPlot software to conduct a detailed analysis of the interactions between Mut3 (S199E) and its neighboring amino acids (Laskowski, et al. 2011) .The results demonstrated that after mutating the nonpolar amino acid S199, located on the protein surface, to the negatively charged acidic amino acid E199, the new mutation site not only formed one new hydrogen bonds with A309 (Fig. 4b) but also established three ionic bonds (or salt bridges) between its side-chain carboxyl group and the imidazole ring of H201. These interactions significantly improved the structural stability of this region.

To further elucidate the molecular mechanism underlying the enhanced thermal stability, this study employed Molecular Dynamics (MD) simulations to systematically analyze the binding free energy, root-mean-square deviation (RMSD), and root-mean-square fluctuation (RMSF) values of the wild-type (S199) and mutant (E199) at 323 K (Sprenger, et al. 2015, Valdes-Tresanco, et al. 2021). The computational results revealed that the binding free energy of the mutant (E199) was − 28.68 kcal/mol, slightly lower than that of the wild-type (S199) (-27.96 kcal/mol). Analysis of the binding free energy between the mutant/wild-type enzymes and the substrate indicated that electrostatic interactions played a dominant role in the binding process, while van der Waals forces contributed as a secondary factor. Specifically, in the mutant enzyme, Asn276 and Tyr302 each provided a binding free energy of less than − 2 kcal/mol, suggesting that these two residues significantly promote substrate binding (Case, et al. 2010, Sprenger, et al. 2015, Tiwari, et al. 2025). In contrast, in the wild-type enzyme, Asp255, His274, Trp298, and Tyr302 each contributed binding free energies below − 2 kcal/mol, further confirming their critical role in substrate binding. RMSD and RMSF are two important statistical metrics used to quantify structural differences between molecules and characterize the fluctuation of protein residues relative to their average structure during simulations. At 323 K, the mutant (E199) exhibited an RMSD value of 0.25 nm and an overall RMSF value of 0.0566, showing no significant difference compared to the wild-type (S199) (RMSD: 0.2 nm; overall RMSF: 0.057) (Fig. 4b). Structural analysis based on homology modeling indicated that the E199 mutation site is located on a β-sheet of the protein, and the mutation did not induce significant structural fluctuations.

This study integrated machine learning, molecular dynamics (MD) simulations, and sequence alignment to successfully construct four mutants of microbial transglutaminase (MTG), namely E164L, E164P, S199E, and S199Q. Experimental results demonstrated enhanced thermal stability in all mutants, with the S199E mutant exhibiting significantly superior thermostability at 60 °C compared to previously reported single-point MTG mutants.

A critical challenge in enzyme engineering lies in balancing the trade-off between enzymatic activity and thermal stability (Luo, et al. 2023, Wang, et al. 2021). To address this, the SCSAddG model was employed to predict potential mutation sites in the MTG sequence that significantly influence thermostability, followed by site-directed mutagenesis. All mutants were successfully expressed as active enzymes using the P. pastoris‌ system. Among them, the S199E mutant displayed the most pronounced improvement in both thermal stability and specific activity, followed by S199Q, while E164L and E164P showed relatively modest enhancements. LigPlot analysis revealed that the S199E mutant formed two new hydrogen bonds and three ionic bonds, substantially increasing the enzyme’s structural rigidity and thereby enhancing its thermostability. Further simulations confirmed that the improved stability was not due to global conformational changes but rather to localized interactions: the negatively charged E199 formed one hydrogen bond with A309, while its side-chain carboxyl group established three ionic bonds (or salt bridges) with the imidazole ring of H201. These interactions significantly stabilized the local structure (Fig. 3e).

In summary, this study verified the amino acid sites (E164 and S199) with higher thermal stability shown in the prediction results through biological experiments. The experimental results were consistent with the model predictions, confirming the accuracy and reliability of the model. In addition, according to the model predictions, three sites (P76, G97, and N297) that have not been reported in the literature were newly identified. The experimental results showed that the prediction based on the SCSAddG model not only achieved effective mutation and evolution of transglutaminase, providing high-quality candidate sequences for subsequent research, but also significantly reduced the labor cost required for the experiments.

Conclusion

This study employed an artificial intelligence-assisted rational design approach to engineer the thermal stability of transglutaminase (TG) derived from S. mobaraense. By constructing four single-point mutants (E164L, E164P, S199E, and S199Q), experimental results demonstrated that all mutants significantly enhanced the enzyme’s thermal stability. Among them, the S199E mutant exhibited the most outstanding performance, retaining 76.5% of its residual activity after 3 min of incubation at 60 °C–a 62.1% improvement compared to the wild-type MTG (which retained only 14.4% activity). These findings not only provide theoretical support for the broader application of MTG in high-temperature environments but also offer a valuable reference for the design and modification strategies of other (α/β) disc-fold enzymes. Additionally, this research lays the foundation for developing more efficient and stable enzyme preparations in fields such as the food industry, biomedicine, and tissue engineering.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

SCSAddG

Sparse convolutional network driven by self-attention mechanism

MTG

Microbial transglutaminase

ADC

Antibody-drug conjugates

MD

Molecular dynamics

PBC

Periodic boundary conditions

NVT

Number of particles, volume, and temperature

NPT

Number of particles, pressure, and temperature

RMSD

Root Mean Square Deviation

RMSF

Root Mean Square Fluctuation

Author contributions

Xiao-Ping Song: Writing—review & editing, Writing—original draft, Validation, Supervision, Resources, Project administration, Methodology, Funding acquisition, Conceptualization. Kai Han: Writing—review & editing, Methodology, Investigation. Pei Xu: original draft, Visualization, Methodology, Investigation, Formal analysis. Jia-Ni Zheng: Writing—review & editing, Methodology, Investigation. Jing-Jing Cai: Methodology. All authors read and approved the final manuscript.

Funding

This work was financially supported by the Anhui Provincial Natural Science Foundation (No. 2022AH052316, GXXT-2022-002) and key project for supporting outstanding young talents in colleges and universities (gxyqZD2022103).

Data availability

All data generated or analyzed during this study are included in this published article.

Declarations

Ethics approval and consent to participate

All authors have read and agreed the ethics for publishing the manuscript.

Consent for publication

All authors have read and approved the manuscript before submitting it to Bioresources and Bioprocessing.

Competing interests

The authors declare that they have no competing interests.