A combined strategy for high-efficiency expression of alkaline protease PrtA in Komagataella phaffii (syn. Pichia pastoris)

Abstract

Background

Microbial serine proteases are valuable for industrial applications due to broad substrate specificity and stability. However, heterologous overexpression in microbial hosts is often limited by cytotoxicity and poor secretion. This study developed an integrated strategy combining protein engineering and signal peptide optimization to enhance extracellular production of PrtA—a key acid-stable alkaline serine protease—in Komagataella phaffii.

Results

Directed evolution generated the Q245K variant, showing 1.35-fold higher extracellular expression than wild-type PrtA. A machine learning model, MPEPE (Mutation Predictor for Enhanced Protein Expression), was used to identify critical residues involved in protein secretion; saturation mutagenesis at the top-predicted site generated the I342D mutant with 1.48-fold improved productivity. The double mutant PrtA-Q245K/I342D achieved synergistic enhancement (1.84-fold higher secretion) without altering enzymatic properties. Evaluation of nine signal peptides revealed that serum albumin, α-factor (without pro-region), and PrtA’s native signal peptides each doubled the combinatorial mutant’s secretion, yielding 4.98-fold higher expression than the wild-type. In contrast, α-factor pro-region inclusion drastically reduced yields. In a 15-L fed-batch bioreactor, the optimized strain produced PrtA-Q245K/I342D at 4807.5 U/mL, equivalent to 1.5 g/L protein.

Conclusions

The combined approach of directed evolution, machine learning-guided mutagenesis, and signal peptide engineering significantly boosted PrtA secretion while maintaining functional integrity. This strategy demonstrates strong potential for scalable industrial production of challenging heterologous proteases.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12934-026-02978-z.

Background

As the dominant class of peptidases, serine proteases constitute more than one-third of all characterized proteolytic enzymes. According to the MEROPS protease database, serine proteases are categorized into 15 clans and 54 families based on catalytic mechanism and common ancestry (https://www.ebi.ac.uk/merops/). The S8 family, also known as the subtilase family, is divided into the S8A and S8B two subfamily and represents the second largest group of serine proteases. Although serine is universally conserved as the catalytic residue across this diverse enzyme family, the composition of the remaining catalytic triad residues varies significantly between subfamilies [1].

S8A family proteases are characterized by non-specific cleavage and robust catalytic activity [2]. In terms of pH adaptation, S8 family proteases maintain favorable stability over a pH range from weakly acidic to strongly alkaline [3], and some members exhibit high stability in both acidic and alkaline environments [4, 5]. Serine proteases of this family generally have an optimal temperature above 55℃ [6–8] and outstanding thermostability, with little loss of enzyme activity after high-temperature incubation [9, 10]. Owing to these advantageous properties, the S8A subfamily serine proteases are widely utilized in food processing [11], animal feed [12], detergent [13], leather [14], pharmaceuticals and biotechnology. The functional potential of S8A serine proteases continue to motivate discovery and characterization novel microbial sources yielding enzymes with desirable properties from bacteria and fungi [15]. Major bacterial and fungal strains utilized for S8 serine proteases production are from the Bacillus spp. and Aspergillus spp [16]. Among the fungal sources, A. niger [17], A. fumigatus [18], A. oryzae [19], A. nidulans [20], A. sojae [21], and A. terreus [22] are significant serine protease producers. However, persistent bottlenecks including low enzyme activity yields during industrial production and insufficient in vitro stability of the produced protease remain major obstacles to their practical application [16].

Heterologous expression in suitable host systems is an efficient strategy for enhancing the yield of serine proteases. Microbial expression systems such as Escherichia coli (E. coli), Bacillus spp., Komagataella phaffii (K. phaffii) and A. niger have all demonstrated feasibility for expressing heterologous S8 famliy serine protease. However, fungal-derived serine proteases often face challenges in achieving proper folding and secretion in prokaryotic hosts. The K. phaffii system offers advantages, including a simplified endogenous secretory background and robust post-translational modification capabilities, which facilitate the scalable fermentation of fungal serine proteases [23]. Successful heterologous expression in K. phaffii has been achieved for proteases derived from A. niger [17], A. sojae [21] A. oryzae [24] with yields reaching up to 331.5 U/ml, 400.4 ± 40.5 U/mL or 513 mg/L. Despite these successes, the heterologous expression of protease PrtA from A. nidulans in K. phaffii was not successful, even though PrtA exhibits significant biotechnological potential due to its thermostability and broad pH tolerance [20].

The efficiency of heterologous protease expression is influenced by multiple factors, including protein structure, misfolding propensity, stability, and inherent toxicity [25, 26]. Common strategies for optimizing expression involve codon optimization [27], promoter engineering [28], signal peptides selection [29], co-expression of molecular chaperones [30], and fermentation conditions optimization [31]. The heterologous expression capability of enzymes has been significantly enhanced by the development of directed evolution and high-throughput screening technologies [32]. Meanwhile, machine learning (ML) has introduced transformative approaches to protein engineering. By integrating deep learning with multi-omics data, researchers can perform predictive analyses to guide the engineering of protein sequences for improved expression. For instance, our team developed MPEPE (Mutation Predictor for Enhanced Protein Expression), an expression prediction model that leverages a protein language model to predict expression propensity across 88 host organisms and facilitate the rational design of protein sequences with enhanced expression levels [33].

Additionally, signal peptide selection is crucial in secretory expression. Signal peptides are located at the N-terminus of proteins and guide them into the endoplasmic reticulum for subsequent processing and transport. Their classic structure comprises: a positively charged N-terminal region (N-domain), a hydrophobic core region (H-domain), and a C-terminal region (C-domain) containing the signal peptidase cleavage site [34]. In K. phaffii, the most commonly used heterologous secretion signals are the α-mating factor (α-MF) signal peptide from Saccharomyces cerevisiae and the native acid phosphatase (PHO1) signal peptide [35]. The signal peptide from the heterologous protein itself can sometimes function effectively [36]. The α-MF signal peptide (85 amino acids) comprises a leader peptide (first 19 amino acids, responsible for targeting the endoplasmic reticulum) and a pro-peptide (remaining 66 amino acids, containing glycosylation sites that assist protein folding and maturation) [37]. Although α-MF signal peptides are widely used, they may not be the optimal or most efficient choice for certain specific proteins (e.g. PrtA) [36, 38].

Previous studies have reported that subtilisin-like proteases of bacterial origin can reach an expression level of 16 g/L in Bacillus expression systems, whereas only 0.2 g/L was achieved in K. phaffii [39]. Serine proteases from A. niger, A. oryzae, and the thermophilic fungus Thermoascus aurantiacus var. levisporus have been successfully heterologously expressed in K. phaffii. Specifically, the expression level of the serine protease from A. oryzae was 513 mg/L [40], the specific activities of serine proteases from A. niger were 331.5 U/ml [17] and 115.58 U/mg [41], respectively, and the yield of serine protease from T. aurantiacus var. levisporus reached 3.2 g/L [42].

This work represents the first successful heterologous expression of PrtA in K. phaffii. The recombinant PrtA inherited the excellent properties of the S8 protease family: the purified enzyme showed a high specific activity of 3519 U/mg, an optimal pH of 7.0, and broad pH stability, retaining more than 80% of its activity after incubation at pH 2.0–11.0 for 1 h. PrtA exhibited an optimal temperature of 60 °C and favorable stability at 55 ℃. Additionally, we employed a combination of directed evolution and machine learning-guided site-saturation mutagenesis to identify and engineer key residues limiting expression. And we systematically screened secretory signal peptides to improve translocation and processing efficiency. This comprehensive approach outperform conventional single-factor modifications, achieving a substantial increase in proteolytic yield.

Methods

Strains, plasmid, chemicals, and media

E. coli Trans1-T1 and K. phaffii GS115 were used as the hosts for cloning and expression. The plasmid pPIC9 and pPICZαA used gene expression vectors. Casein sodium salt from bovine milk (C8654) was purchased from Sigma-Aldrich (St. Louis, MO, USA) and used as the substrate. Restriction endonucleases (EcoRI, NotI and DraI) were purchased from Thermo Fisher Scientific (Waltham, MA, USA). All other chemiscals were of analytical grade and were commercially available. Buffered glycerol complex medium (BMGY), buffered methanol complex medium (BMMY), minimal dextrose medium (MD), and minimal methanol medium (MM) were prepared as described in the manual of the EasySelect™ Pichia Expression Kit (Invitrogen).

Plasmid construction

The gene encoding serine protease PrtA (XP_663162) was optimized according to K. phaffii codon usage bias and synthesized by BGI Genomics. The gene fragment coding for mature PrtA was amplified with specific oligonucleotide primers PrtA-MF and Prt-MR (Table S1) and cloned into pPIC9 plasmid in-frame fusion with the α-factor via homologous recombination to generate pPIC9- prtA-M using the Basic Seamless Cloning and Assembly Kit (Transgen, Beijing). The resulting construction was sequenced to verify the correct insertion.

Protein expression and purification

The recombinant plasmids containing target gene fragments were linearized with Bgl II restriction endonuclease and subsequently electro transformed into K. phaffii GS115 competent cells. Protein expression was induced following the manufacturer’s protocol (EasySelect™ Pichia Expression Kit, Invitrogen). His⁺ transformants were primarily selected by transferring to MD agar plates and MM agar plates for 48 h at 30 °C. Using sterile toothpicks, individual colonies were transferred onto skim milk agar screening plates (pH 7.0) [40]. The colonies exhibiting distinct hydrolysis halos were selected for subsequent liquid cultivation and induction expression. The transformant exhibiting the largest hydrolysis halo was propagated in 400 mL BMGY for biomass accumulation, followed by induction in 200 mL BMMY supplemented with 0.5% v/v methanol. Cultivation proceeded at 30 °C with 200 rpm agitation for 48 h. Culture supernatants were recovered by centrifugation (12,000×g, 10 min, 4 °C), followed by concentration through a Vivaflow 200 ultrafiltration membrane with a molecular weight cut-off of 5 kDa (Vivascience, Hannover, Germany). Secreted proteins were purified via a HiTrap Q HP Cation Exchange Column (#17115301, Cytiva, USA). Following equilibration with 10 mM citrate-phosphate buffer (pH 6.6), a linear NaCl gradient (0–1.0 M) was applied for elution. Purity and molecular weight were analyzed by SDS-PAGE.Protein concentrations were determined using the Easy Protein Quantitative Kit (Bradford, Transgen, Beijing) following the manufacturer’s instructions, with bovine serum albumin (BSA, ≥ 98% purity) as the calibration standard.

Enzyme activity assay

The protease activity was determined using casein as the substrate according to the methods of Wang et al. (2023) [5]. Briefly, 200 mM phosphate buffer (pH 7.0) was used, and the reaction system consisted of 500 µL of casein substrate (2%, w/v) and 500 µL of appropriately diluted enzyme solution. The reaction was performed at 60 °C for 20 min and terminated by 1 mL of 0.4 M trichloroacetic acid, followed by centrifugation at 12,000×g for 3 min. 500 µL supernatant of the reaction mixture was transfered to 2.5 mL of 0.4 M sodium carbonate solution, followed by the addition of 500 µL Folin-phenol reagent. The mixture was incubated at 40 °C for 20 min and the absorbance measurements were recorded at 680 nm following equilibration of the samples to room temperature. One unit of protease activity (U) is defined as the amount of enzyme required to liberate 1 µg of tyrosine per minute under specified reaction conditions. All results are presented as the mean of three assays.

Characterization of purified enzymes

The optimal temperature of PrtA and mutants were identified within 30–70 °C under optimal pH conditions. The thermal stability was tested by the enzymes at 50, 55 and 60 °C for specified durations, and the residual activity assessed under optimal conditions. The pH adaptive profiles for PrtA and mutants were determined across pH 3.0–12.0. using 50 mM lactate-sodium lactate buffer (pH 3.0–5.0), 200 mM phosphate buffer (pH 5.0–8.0), and 100 mM borax-sodium hydroxide buffer (pH 8.0–12.0) at the optimal temperature for 20 min. The pH stability was analyzed by incubating the enzymes in above buffers at 37 °C for 1 h, and residual activity was assessed under standard conditions.

The kinetic parameters were determined using 0.5–25.0 mg/mL casein as substrate under optimal conditions for 5 min. The kinetic parameters Km and Vmax were derived via the Michaelis–Menten equations in GraphPad Prism 9.0 (GraphPad Software, San Diego, USA).

Directed evolution for screening high-expression mutants

The pPICZαA- prtA plasmid served as template for random mutagenesis using the Gene Morph II Random Mutagenesis Kit (Agilent) according to manufacturer specifications. Mutation frequency (target range: 1–3 bp/kb) was modulated through template gradient dilution. An EZClone seamless cloning strategy was employed with gene-specific primers (F: 5′-CGAGAAAAGAGAGGCTGAAGCTGAATTC-3′; R: 5′-TTCTAGAAAGCTGGCGGCCGC-3′). PCR products were subjected to DMT enzyme digestion (37 °C, 2 h) to eliminate parental plasmids, followed by electro transformation into E. coli TOP10 competent cells. Transformed cells were plated on LB agar containing zeocin (25 µg/mL) and incubated at 37 °C for 16 h. Fifty randomly selected clones were verified by sequencing, confirming an average mutation frequency.

Library plasmids were linearized with DraⅠ (37 °C, 2.5 h) and electroporated into K. phaffii GS115 competent cells. Transformants were plated on YPD agar containing zeocin and cultured at 30 °C for 48 h. Individual colonies were spot-inoculated onto skim milk double-layer agar plates. After 12 h incubation at 30 °C, mutants exhibiting hydrolysis were selected. Positive clones with the largest transparent zone were cultured in 50 mL BMGY medium at 30 °C for 48 h. And induction was performed in BMMY medium supplemented with 0.5% methanol for 48 h. Crude enzyme extracts were analyzed by SDS-PAGE and enzymatic assay.

Expression modeling-guided saturation mutagenesis for enhanced protein production

In order to predict mutations that could enhance the heterologous expression of PrtA, the mutation predictor for enhanced protein expression (MPEPE) strategy [43] was employed. Whole-sequence scanning saturation mutagenesis was performed on the mature peptide region of PrtA, constructing an in silico mutant library. Amino acid sequences of these virtual mutants were used as inputs for the MPEPE model to predict all single-point mutations that could improve their expression. The prediction model generated quadruplicate expression scores for each variant, enabling robust identification of expression-enhancing mutations. High score virtual mutants within each group were selected and a Venn analysis was performed on these four sets of top-ranked virtual mutants. Furthermore, evolutionary analysis was utilized to identify evolutionarily nonconserved residues predicted to maintain protein function for subsequent mutagenesis.

Evaluation of different secretion signals on the effect of protein secretion

To enhance PrtA secretion in K. phaffii, we evaluated multiple secretory signal peptides by replacing the α-MF. Tested signal peptides included: native PrtA signal peptide, the α-MF pre-peptide, PHO1 signal sequence from K. phaffii, glucoamylase signal sequence from Aspergillus niger, serum albumin signal sequence from Homo sapiens, inulinase presequence from Kluyveromyces maxianus, invertase signal sequence from Saccharomyces cerevisiae, killer protein signal sequence from Saccharomyces cerevisiae and lysozyme signal sequence from Gallus gallus (reference: PichiaPink™ Manual, Cat. A11150-A11154). All signal peptide DNA sequences were synthesized by BGI Genomics and cloned between the BamHI and EcoRI sites of pPIC9-PrtA. This maintained translational reading frame continuity between each signal peptide and the PrtA coding sequence.

Copy number calculation

To rule out the possibility that the observed increase in protease production was caused by an elevated gene copy number, the copy number of the PrtA gene was quantified. Genomic DNA was extracted from the following three strains using the Fungal DNA Kit (Omega): GS115-PrtA, GS115-Q245K/I342D-PrtA, and GS115-Q245K/I342D-PrtA with its native signal peptide. A standard curve was constructed using the pPIC9-PrtA plasmid, and qPCR analysis demonstrated that the PrtA gene was present as a single copy in all tested strains. This evidence indicates that the enhanced protease expression resulted from the introduced sequence modifications (Q245K/I342D) and the signal peptide substitution.

High-cell-density fermentation of recombinant K. phaffii

Recombinant K. phaffii was cultivated in a 15-L bioreactor using a fed-batch strategy to achieve high recombinant protease yields. The fermentation protocol comprised three phases: biomass growth, carbon-limited fed-batch and methanol-induced expression. A 7 L working volume was used, and parameters were set as follows: 0.8 m³/h aeration, 30 °C, pH 4.5–5.0, and 800 rpm. After 24 h of cultivation, methanol induction was initiated at a cell wet weight of 200 g/L. Samples were collected at 0, 24, 48, 72, 96, and 108 h during induction for analysis of protease activity and protein expression level. The process was performed according to the PichiaPink™ Expression System manual (Invitrogen). Protease activity in cell-free culture supernatant was measured daily throughout the induction phase.

Results

Heterologous expression of PrtA in K. phaffii and its enzymatic properties

PrtA from A. nidulans initially exists as a 403-residue protein, including a putative signal peptide of 20 residues and a pro-peptide of 101 residues at the N-terminus, which are subsequently cleaved to produce the 282-residue active mature protease. This study successfully expressed PrtA with an enzyme activity of 5.08 U/mL after cloning a prtA gene fragment (1152 bp) encoding both the propeptide and mature protease domains into a K. phaffii strain and inducing it with methanol. SDS-PAGE revealed purified PrtA at 35 kDa, exceeding its theoretical mass (28 kDa). Endo H treatment confirmed glycosylation as the cause, reducing the mass to 28 kDa (Fig. 1A).

Fig. 1.

Heterologous expression of PrtA in K. phaffii and its enzymatic properties. A SDS-PAGE analysis of recombinant PrtA. M: Protein molecular weight marker, lane 1: culture supernatant from the induced transformant, lane 2: Purified PrtA, lane 3: purified PrtA after deglycosylation with Endo H. B Optimal pH at 40℃ and 60 ℃. C Optimal temperature. D pH stability. E Thermostability. F Enzyme activity curve of PrtA treated with pepsin at 37℃ for different durations

Enzymatic characterization indicated that the optimal pH of PrtA was 7.0 at 60 °C and 9.0 at 40 °C (Fig. 1B). According to standard classification, it is defined as an alkaline serine protease due to its alkaline optimal pH at 40 °C. The optimal temperature of PrtA was determined to be 65 °C (Fig. 1C). The enzyme retained over 90% of its activity within a broad neutral pH range (6.0–7.0). Notably, PrtA exhibited exceptional stability across an extensive pH spectrum (2.0–11.0), retaining more than 80% of its initial activity after pre-incubation (Fig. 1D). Its stability under highly acidic conditions (pH 2.0) is a distinctive trait rarely reported for serine proteases. This property is particularly advantageous for animal feed applications, as it allows PrtA to remain intact in the acidic stomach environment while remaining active in the neutral intestines [44]. The enzyme also demonstrated notable thermostability, maintaining over 90% and 60% of its activity after incubation at 50 °C for 1 h and 55 °C for 30 min, respectively (Fig. 1E). Furthermore, PrtA exhibited significant resistance to pepsin, retaining 61% of its activity after a 120-minute digestion at 37 °C (Fig. 1F).

The specific activity of purified PrtA measured at optimal temperature and pH conditions was 3519 ± 18 U/mg. The Michaelis constant (Km) of PrtA was 0.117 ± 0.005 mM, and the maximum reaction rate (Vmax) was 4573 ± 77 µmol/min·mg.

Directed evolution for screening high-expression mutants

Directed evolution is a commonly used protein engineering approach that can effectively enhance the catalytic activity and stability of enzymes. In this study, the EZClone method was employed to construct a directed evolution mutant library, mutation frequency ranging from 6 to 10 bp/1000 bp. Based on screening the hydrolysis zone on milk agar plates (Fig. 2A), a total of 10,000 mutants were evaluated. Among these, 302 mutants exhibiting distinctly enlarged hydrolysis zones (indicating higher protease activity) were selected for shake flask cultivation, followed by determination of protease activity. The optimal mutant obtained from primary screening based on enzymatic activity assays exhibited an activity of 27.6 U/mL, 38% increase over the wild-type (WT) (20.0 U/mL). During library screening, mutants retaining proteolytic activity were isolated using milk-based agar plates, significantly reducing workload. Crucially, enzyme activity/OD600 was employed to evaluate mutational effects on enzymatic function, eliminating biases from cell density variations during protein expression. Nevertheless, we observed that hydrolysis zone size fails to quantitatively correlate with either protein expression levels or enzymatic activity. Consequently, establishing high-throughput screening methods capable of accurately quantifying expression changes remains a critical focus for future protease engineering endeavors.

Fig. 2.

Expression analysis of mutants with directed evolution of PrtA. A Hydrolysis of the mutants. B Enzymatic activity of PrtA and its mutants (red dots of ΔA680/OD600 represent the enzyme production ability of PrtA and mutant strains). C SDS-PAGE analysis

Through genomic extraction and identification from these mutants, two effective mutation sites were identified, namely L102M and Q245K. The variants PrtA-L102M, PrtA-Q245K, and PrtA-L102M/Q245K were constructed and characterized. Comparative analysis of expression levels revealed that the Q245K mutant exhibited 1.35-fold higher protein yield than WT PrtA, whereas the L102M mutation showed no significant impact on expression (Fig. 2B). Mutant Q245K exhibited a specific activity of 3446.71 U/mg, while maintaining enzymatic characteristics comparable to the WT enzyme (Fig. S2). SDS-PAGE results showed that the Q245K mutant exhibited higher expression levels than the WT (Fig. 2C). These results identify Q245 as a critical residue modulating PrtA expression.

Deep learning expression modeling-guided mutagenesis for enhanced protein production

Following comprehensive single-site saturation mutagenesis across the entire 283-residue mature peptide of PrtA, an in silico mutant library comprising 5377 variants was constructed. Each mutant sequence was processed using expression prediction model MPEPE to quantify expression levels, generating four replicate output datasets for subsequent analysis. As summarized in Table 1, the MPEPE model scores were ranked in descending order with the top 30 highest-scoring virtual mutants selected per group. A venn analysis was performed on these four sets of 30 highest-scoring virtual mutants (Fig. 3A). The results revealed 16 virtual mutants commonly identified as high-expression variants across all four output sets (I342D, L339N, L337D, Y340D, I342N, L339E, L339D, L341D, L337E, L339Q, V335D, I342E, S338D, L341N, L341E, and V335E).

Table 1.

Virtual mutants prediction results

Sort by score (high → low)	List1		List2		List3		List4
1	L339D	0.56782	L337D	0.50954	L337D	0.69171	L339D	0.93202
2	L341D	0.56594	L339D	0.50722	L339D	0.69155	L337D	0.92637
3	L337D	0.56471	L341D	0.50644	I342D	0.68957	L341D	0.91789
4	I342D	0.56359	Y340D	0.50421	L341D	0.68669	I342D	0.91693
5	Y340D	0.56161	I342D	0.50320	V335D	0.68600	L376D	0.91197
6	L341P	0.55923	Y185P	0.50174	Y340D	0.68553	L339E	0.90624
7	L341N	0.55897	L341N	0.50146	L339E	0.68516	I342N	0.90507
8	V335D	0.55798	L337E	0.50143	L337E	0.68432	L339N	0.90244
9	S338D	0.55643	G336D	0.50043	I342N	0.68263	L341N	0.89921
10	L339E	0.55544	L339N	0.50012	V335E	0.68199	Y340D	0.89841
11	G336D	0.55409	L339E	0.50003	L339N	0.68194	L337E	0.89091
12	L339N	0.55408	V335D	0.49979	L339Q	0.68077	L339Q	0.88774
13	I342N	0.55382	L337N	0.49875	L341N	0.67992	V335D	0.88253
14	L337E	0.55366	L337K	0.49872	Y185D	0.67950	L337N	0.88028
15	L339K	0.55310	I342E	0.49847	I208E	0.67886	L341P	0.87941
16	I342P	0.55277	L341P	0.49832	P332D	0.67886	L339K	0.87611
17	L339Q	0.55172	Y185K	0.49820	I342E	0.67875	I342P	0.87435
18	L341E	0.55136	L341E	0.49817	H333D	0.67865	K358N	0.87246
19	I342E	0.55109	S338D	0.49802	I208D	0.67833	I342E	0.87224
20	L341K	0.55049	H166P	0.49799	L376D	0.67833	S338D	0.87126
21	L341Q	0.54984	L337Q	0.49792	G336D	0.67795	L376A	0.87071
22	I342G	0.54895	L339K	0.49762	Y340E	0.67790	L375A	0.86872
23	Y340N	0.54867	G163P	0.49756	L337Q	0.67774	L341E	0.86653
24	V335E	0.54840	G182P	0.49750	S338D	0.67774	L337Q	0.86465
25	I342K	0.54813	I342K	0.49747	L341E	0.67747	V334Q	0.86437
26	L337N	0.54793	S188P	0.49707	W215D	0.67742	L344D	0.86270
27	Y340E	0.54789	I342N	0.49692	I208Q	0.67646	Y340N	0.86049
28	I342Q	0.54728	Y340E	0.49689	W215Q	0.67635	V335E	0.85807
29	L341S	0.54719	V335E	0.49683	Y340N	0.67635	A343D	0.85639
30	I342S	0.54719	L339Q	0.49680	P373A	0.67635	L339R	0.85403

Fig. 3.

Deep learning expression modeling-guided mutagenesis for enhanced PrtA protein production. A Venn analysis of predicted high-scoring mutants. B Conservation analysis of mutation regions. C Expression analysis of PrtA and mutants (red dots of ΔA680/OD600 represent the enzyme production ability of PrtA and mutant strains). D Expression analysis of the i342 saturated mutant

These mutation sites were concentrated within the V335–I342 region. Based on evolutionary conservation analysis of this amino acid region (Fig. 3B), we rationally selected less conserved sites (V335D, V335E, S338D, I342D, I342E and I342N) for site-directed mutagenesis. SDS-PAGE analysis revealed significantly enhanced protein expression levels of the I342D, I342N, and I342E variants relative to the WT (Fig. S1). These variants were therefore selected for subsequent enzymatic activity assays. The results showed that the enzyme production capacity of the I342D, I342N, and I342E mutants was significantly higher than that of the WT PrtA (Fig. 3C). Among them, I342D exhibited the highest activity (U/OD600 = 29.6 U/mL), reaching 1.48-fold that of WT PrtA. And the activities of the I342E and I342N mutants were 1.40 times that of the WT level. Saturated mutagenesis was conducted at the I342 site, and the obtained results were consistent with the predictions of the constructed model. Specifically, the I342D mutant exhibited the highest enzyme production capacity, followed by I342N and I342E (Fig. 3D). The specific activities of the I342D, I342N, and I342E mutants were 3,564.0 U/mg, 3,488.4 U/mg, and 3,510.2 U/mg, respectively, which were comparable to that of the wild type (WT). Additionally, the enzymatic properties of the I342D, I342E, and I342N mutants were similar to those of the WT (Fig. S3). This demonstrates that the mutation sites predicted by the model are highly accurate, and the rational application of the model can enable more efficient prediction of high-expression sites.

Combinatorial mutagenesis

To synergistically enhance PrtA expression, we constructed a panel of double mutants, including: Q245K/I342D, Q245K/I342E, Q245K/I342N. Enzymatic characterization was performed to determine their optimal pH, optimal temperature, pH stability, and thermal stability, and the results showed that all three mutants exhibited enzymatic properties identical to those of the WT enzyme (Fig. S4). Notably, among these variants, the Q245K/I342D double mutant displayed the highest volumetric activity (U/OD600 = 36.8U/mL), achieving 1.84-fold of WT (U/OD600 = 20U/mL) (Fig. 4A). As shown by SDS-PAGE, the expression level of the double mutant Q245K/I342D was higher than that of the WT (Fig. 4B).

Fig. 4.

Protein expression analysis of PrtA double mutants. A Enzymatic activity of PrtA and its mutants (red dots of ΔA680/OD600 represent the enzyme production ability of PrtA and mutant strains). B SDS-PAGE analysis

Evaluation of different secretion signals on the effect of Q245K/I342D-PrtA secretion

To increase the titer of recombinant Q245K/I342D-PrtA proteins in K. phaffii, 10 signal peptides (Table 2) were evaluated as alternatives to the native α-MF signal peptide. The results showed that following replacement with signal peptides from invertase, lysozyme, and killer toxin signal peptides, protease expression levels increased by 1.48-fold, 1.21-fold, and 1.44-fold, respectively. However, when using glucose oxidase, inulinase, or PHO1 as signal peptides, expression levels decreased to varying degrees, with fold changes of 0.3, 0.48, and 0.43, respectively. (Fig. 5A).

Table 2.

Different secretory signal peptides

Signal	Source	Accession numbers
Inulinase presequence	Kluyveromyces maxianus	QGN13683.1
Glucoamylase signal sequence	Aspergillus awamori	AAS91802.1
Phosphate metabolism gene 1	Saccharomyces cerevisiae	KAI0461015.1
Invertase signal sequence	Saccharomyces cerevisiae	AAA35129.1
Lysozyme signal sequence	Gallus gallus	NP_001001470.1
Killer Protein signal sequence	Saccharomyces cerevisiae	AAA46254.1
Serum albumin signal sequence	Homo sapiens	NP_000468.1
α-MF pre-prptide	Saccharomyces cerevisiae	NP_015137.1
PrtA native signal sequence	Aspergillus nidulans	AAA67705.1

Fig. 5.

Expression of PrtA-Q245K/I342D using different signal peptides. A Expression analysis of PrtA with different signal peptides (red dots of ΔA680/OD600 represent the enzyme production ability of PrtA and mutant strains). B Enzymatic activity of the strain in a 15-L fed-batch bioreactor. C SDS-PAGE analysis of fermentation supernatant

The final protease activity of the GS115-PrtA strain reached 965.2 U/mL, while the GS115-Q245K/I342D-PrtA strain with its native signal peptide exhibited a markedly enhanced activity of 4807.5 U/mL, with a protein expression level of 15 g/L, representing 4.8-fold increase (Fig. 5B). SDS-PAGE results also indicated that using the native signal peptide of PrtA is more favorable for its production by fermentation (Fig. 5C). These results demonstrate that replacing the signal peptide with the native version significantly enhance the expression efficiency of the protease in recombinant K. phaffii.

Analysis of the hydrophilicity and hydrophobicity of the revealed the signal peptides of glucose oxidase, inulinase, and PHO1 demonstrated lower hydrophobicity values compared to the α-MF. These lower-value signal peptides thus represent more hydrophilic sequences (Fig. S3A). The signal peptides of invertase, lysozyme, and killer toxin exhibited significantly higher hydrophobicity in their H-regions compared to the S. cerevisiae α-MF signal sequence (Fig. S3B). Compared to the α-MF, the signal peptides of serum albumin, the α-MF pre-peptide, and the native PrtA signal peptide maintained a relatively balanced level between their hydrophobic and hydrophilic regions.

Discussion

K.phaffii has been validated to be an effective heterologous host for serine proteases, as demonstrated in studies of Thermomonospora fusca YX serine protease TfpA [45], Aspergillus oryzae proteases [40], Thermoascus aurantiacus var. levisporus proteases [41], and Trichoderma proteases [46]. These serine proteases comprise a secretion signal peptide, a pro-peptide domain and a mature protease domain. The mature protease is formed after removal of the signal peptide and the pro-peptide domain [47]. The pro-peptide region is a critical element for both secretion and correct folding of most [20, 40, 45].

As a serine protease characterized by a broad pH activity range and gastric acid tolerance, PrtA demonstrates significant potential for industrial applications such as in animal feed [48]. However, its industrial utilization has been severely constrained by low production yield and high costs, compounded by the fact that previous attempts to express this enzyme in yeast systems have proven unsuccessful—highlighting the inherent challenges associated with its heterologous expression [20]. Therefore, systematically studying and optimizing the expression and secretion of PrtA to overcome its yield bottleneck is crucial for advancing this protease with excellent properties toward practical applications.

In this study, the heterologous expression of PrtA in K.phaffii was initially very low. The enzyme activity in the fermentation supernatant after 48 h was only 5.08 U/mL. Therefore, various strategies required to enhance its expression level. Directed evolution is a commonly used protein engineering approach that can effectively enhance the catalytic activity and stability of enzymes. However, reports on improving the expression levels of enzymes through directed evolution are relatively rare. By combining random and site-directed mutagenesis, the expression of Bacillus licheniformis CotA-laccase in E. coli was improved [49]. Luo et al. [50] improved the functional expression of CotA-laccase through site-directed mutagenesis.

Optimizing enzymes functional expression remains critical for enabling industrial-scale applications [51]. And the amino acid sequence is a critical determinant of expression efficiency [52]. Recent advances in site-directed mutagenesis have demonstrated improved functional expression of recombinant enzymes in Bacillus licheniformis [49] and E. coli [50]. However, despite clear evidence that sequence impacts expression, a universal mechanistic framework to predict these effects across diverse proteins remains elusive. Emerging deep learning tools now offer a promising approach for systematically decoding the sequence determinants that govern heterologous expression [43]. Guided by this potential, we employed a predictive deep learning model to systematically improve PrtA expression. The predictive model employed in this study effectively enhanced the solubility of heterologous proteins. It demonstrated strong performance not only in prediction accuracy but also in preserving protein function and maintaining cross-species applicability [43]. Using this model, key sites that contribute to the high-efficiency expression of PrtA were successfully identified. I342 was confirmed as a critical residue for enhancing PrtA expression. Further saturation mutagenesis at I342 revealed that the I342K mutant also conferred a pronounced increase in band intensity and a 17.5% boost in activity. The efficacy of substituting I342 with charged residues like D, E, N, and K aligns with previous reports that small, charged amino acids (e.g., G, A, E, D, K) often correlate with elevated heterologous expression levels. Importantly, all characterized I342 mutants retained functional properties comparable to the WT enzyme. The expression-prediction model MPEPE enabled screening of mutants enhancing protein expression in E. coli, elevating specific activities of laccase 13B22 and glucose dehydrogenase FAD-AtGDH by 3.49-fold and 7.86-fold, respectively [43]. Although expression improvement in K. phaffii was less pronounced than in E. coli, the model successfully identified mutants capable of significantly enhancing heterologous expression levels.

Having improved the core protein sequence, we next explored whether the secretion signal could be further optimized for the enhanced PrtA variant. Signal peptide Albumin, α-MF pre-peptide, and its native signal peptides led to higher expression levels of Q245K/I342D-PrtA, with increases of 1.75-fold, 2.27-fold, and 2.25-fold, respectively. This demonstrates that for Q245K/I342D-PrtA, the albumin, α-factor, and its native signal peptides more conducive to higher expression levels. It was demonstrated that the α-MF pro-peptide was dispensable for Q245K/I342D-PrtA secretion, with the 19-AA pre-peptide sequences alone competent to direct co-translational translocation [53]. The alkaline protease gene from A. oryzae was successfully expressed in the heterologous K. phaffii with native signal peptide or a-factor secretion signal peptide. The yield of the recombinant alkaline protease with native signal peptide was about 1.5-fold higher than that with a-factor secretion signal peptide [40]. It has been reported that increasing the length or hydrophobicity of the core hydrophobic region (H-region) of signal peptides positively correlates with improved protein production. Consequently, this study analyzed the peptide hydrophobicity revealed that while increased hydrophobicity generally correlates with higher expression (cf. invertase, lysozyme, killer toxin signals vs. albumin/α-factor/PrtA-native signals), extreme hydrophobicity (e.g., glucose oxidase, inulinase, PHO1 signals) or hydrophilicity detrimental. Optimal secretion requires a balance of hydrophobic and hydrophilic regions within the signal peptide.

Collectively, these results indicate that while enhanced hydrophobicity in the signal peptide contributes to higher protein expression, the relationship is not strictly linear. Therefore, optimal expression is achieved not by maximized hydrophobicity, but by a finely tuned equilibrium between the hydrophobic and hydrophilic domains. For PrtA, which possesses an intrinsic propeptide, the truncated α-MF signal peptide (without pro-leader) outperformed the full-length version. The suboptimal expression with the full-length α-MF is attributed to a functional redundancy and potential incompatibility between the two pro-regions, whereas the native PrtA propeptide is more proficient for its secretion.

Our findings on how signal sequences affect PrtA secretion in K. phaffii. show that optimization is protein-specific. Successful signal sequence optimization increased the intracellular transport of the proenzyme. Consequently, this led to a heavy accumulation of mature PrtA, which caused cellular damage and ultimately reduced the final protease yield. Collectively, targeted single-amino acid substitutions profoundly modulated expression performance. By integrating directed evolution with MPEPE-guided mutagenesis and signal peptide screening, we achieved about 6-fold cumulative enhancement in PrtA production. This work establishes a multidimensional optimization framework for industrial-scale serine protease expression.

Conclusions

By combining directed evolution and deep learning models, we obtained protease PrtA mutants Q245K, I342D, and their double mutant Q245K/I342D, which showed 1.35-fold, 1.48‐fold, and 1.84‐fold higher expression levels than the wild type, respectively, with no significant changes in specific activity or enzymatic properties. We further demonstrated that using its native signal peptide doubled the protease activity per OD600 in shake-flask cultures compared to the commonly used α-MF signal peptide. Validation via 15 L high-cell-density fermentation confirmed that the final expression level of the double mutant with the native signal peptide was increased by 4.98-fold relative to the original engineered strain. Importantly, these mutations and signal peptide replacement significantly enhanced secretion in K. phaffii without compromising the catalytic performance or biochemical properties of the protease, laying a solid foundation for the industrial-scale application of PrtA. These strategies have demonstrated significant efficacy in enhancing PrtA expression, laying the foundation for the industrial-scale application of the protease PrtA.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Baozhen Zhao and Tao Dong wrote the original draft. Huiying Luo and Tao Dong revised the manuscript. Baozhen Zhao, Qiao Zhou, Panpan Wei and Tao Dong conducted the experiments and analyzed the data. Honghai Zhang and Xing Qin contributed to validation and investigation. Tao Tu and Huoqing Huang provided resources. Bin Yao and Huiying Luo supervised the research and acquired funding. All authors reviewed and approved the final manuscript.

Funding

This work was supported by National Natural Science Foundation of China (32130101, U24A6011), the Agricultural Science and Technology Innovation Program (CAAS-ZDRW202304) and the China Agriculture Research System of MOF and MARA (CARS-41).

Data availability

The datasets supporting the conclusions of this article are included within the article (and its additional file).

Declarations

Competing interests

The authors declare no competing interests.