Systematic optimization of Komagataella phaffii (Pichia pastoris) for high-efficiency recombinant protein production: Advancements in strain engineering, fermentation optimization, and AI-driven technologies

Abstract

Komagataella phaffii (formerly Pichia pastoris) has emerged as an indispensable eukaryotic expression host for the production of industrial enzymes, food proteins, and therapeutic proteins owing to its Generally Recognized as Safe (GRAS) status, efficient protein secretion capacity, eukaryotic post-translational modification system, and high cell-density fermentation capabilities. In recent years, advances in genetic engineering tools, improved understanding of protein synthesis and secretion pathways in K. phaffii, and the development of systematic strategies for enhancing recombinant protein expression have further established this yeast as an effective eukaryotic expression system. This review summarizes the advances in the design and screening of expression cassette components, including promoters, signal peptides, and terminators, in recombinant strain engineering of K. phaffii. It also provides a comprehensive overview of the key molecular processes affecting protein synthesis and secretion, including transcription, translation, translocation, folding, post-translational modification, endoplasmic reticulum homeostasis, and vesicular transport, together with the corresponding engineering strategies to enhance expression performance. Based on this, strategies for optimizing high-cell-density fermentation are discussed. This review highlights the recent progress in emerging artificial intelligence (AI)-driven technologies applied to the optimization of K. phaffii expression systems, including AI-based protein sequence design and expression prediction, intelligent control of fermentation processes, and genome-scale metabolic modeling-guided strain engineering. Finally, current challenges and future perspectives of the K. phaffii expression system are discussed. Overall, this review systematically summarizes the optimization strategies and underlying mechanisms for recombinant protein expression in K. phaffii, aiming to provide a comprehensive reference for researchers and accelerate the deployment of efficient yeast cell factories across sustainable manufacturing and advanced bioprocess industries.

Graphical abstract

1. Introduction

Driven by expanding industrial, medical, food manufacturing, and agricultural applications, the global demand for diverse, high-performance bioactive proteins, such as industrial enzymes, antibodies, food proteins, and antimicrobial peptides, has been continuously increasing. The development of versatile and cost-effective recombinant protein production platforms is of paramount importance for sustainable and green manufacturing (Fig. 1).

Fig. 1.

Applications and strategies of the K. phaffii expression system. (a) Recombinant protein types successfully expressed in K. phaffii. (b)Different optimization strategies for the K. phaffii protein expression system.

Komagataella phaffii (formerly Pichia pastoris) has been demonstrated to be an efficient, sustainable, and economically advantageous host for heterologous protein expression. Over the past three decades, numerous universities and research institutions have adopted K. phaffii as a platform for recombinant protein production. Many heterologous proteins have been successfully produced in K. phaffii, overcoming various expression bottlenecks. The molecular processes and key factors underlying heterologous protein production in K. phaffii have been progressively elucidated, making it one of the most powerful protein expression systems currently available [[1], [2], [3]].

Compared with engineered hosts, such as Escherichia coli and Saccharomyces cerevisiae, K. phaffii exhibits several notable advantages [4]. It harbors the strong methanol-inducible endogenous promoter P_AOX1 and can utilize methanol as its sole carbon source, thereby providing a robust foundation for high protein expression [5]. The genetic engineering system for recombinant strain construction in K. phaffii is also well-established, relatively simple to operate, and compatible with high-cell-density fermentation [6]. As a eukaryotic host, it can perform post-translational modifications on heterologously expressed proteins [7]. Additionally, its secretory capability facilitates downstream purification [8].

K. phaffii has also been designated by the U.S. Food and Drug Administration (FDA) as a Generally Recognized as Safe (GRAS) microorganism, indicating its safety for use in food-related applications and highlighting its strong potential in the food and pharmaceutical industries [9]. The K. phaffii expression system has enabled successful production of a wide range of proteins, including industrial enzymes [[10], [11], [12]], food proteins [13], antibodies [14], antimicrobial peptides [15], and collagen [16]. >6000 recombinant proteins have been successfully expressed in K. phaffii, and some proteins have reached titers exceeding 20 g/L [17]. These results strongly demonstrate that K. phaffii is an ideal platform for recombinant protein expression.

2. Advances in strain construction and expression cassette design

After decades of research, methods for constructing recombinant protein-expressing strains of K. phaffii have become well-established. Relevant strains and plasmids have been developed into commercial products that are now widely used. Among these, products from Invitrogen (USA) are the most well-known, and the K. phaffii expression manual is widely regarded as a fundamental reference for research (www.thermofisher.com). During strain construction, the expression cassette components within plasmids—promoters, signal peptides, and terminators—form the basis for the secretory expression of recombinant proteins. These elements exhibit distinct characteristics, and different combinations can lead to substantial variations in expression performance.

2.1. Plasmids and strains

Commonly used plasmids for K. phaffii strain construction include pPIC3.5 K and pPIC9K, which contain a kanamycin resistance marker and a histidine auxotrophic complementation marker, as well as pPICZαA and pPICZαB, which carry a bleomycin resistance gene. Zhao et al. designed a modified plasmid, pPICHKA, via fragment replacement and the addition of cloning sites. This design reduces the plasmid size while introducing more cloning sites, thereby facilitating molecular cloning [18]. Li et al. developed a recyclable resistance plasmid based on the Cre/loxP system. After plasmid integration into K. phaffii, the selection marker can be excised upon induction, overcoming the limitation of the one-time use of antibiotic markers. This system enables the construction of multi-copy strains and co-introduction of multiple genes [19]. To improve transformation efficiency, the above-mentioned plasmids generally require linearization prior to transformation. Only linearized plasmids can be integrated into the K. phaffii genome via homologous recombination, ensuring stable inheritance of the target gene [7]. Nakamura et al. demonstrated an alternative non-integrative approach that engineered an episomal plasmid using the K. phaffii autonomous replication sequence Cen2. Due to its ability to replicate independently in host cells, this vector bypasses the requirement of pre-transformation linearization by classical integrative vectors, thereby preserving genomic integrity [20]. In addition, compared with earlier autonomous replication elements, such as PARS1 [21] and panARS [22], Cen2 exhibits higher stability and reduced plasmid loss. Notably, the instability of autonomously replicating plasmids is not necessarily disadvantageous, because transient gene expression followed by inducible plasmid loss can enable scarless gene removal in certain applications.

Currently, most parental strains used for recombinant protein expression in K. phaffii are derived from the original strain. One of the most commonly used strains is X-33, a wild-type strain that is compatible with plasmids carrying a bleomycin resistance marker. GS115 is a histidine auxotrophic strain compatible with HIS4 complementation vectors. Another commonly used strain is SMD1168, in which endogenous proteases are knocked out; this strain is more suitable for expressing proteins that are susceptible to degradation compared to other strains. KM71, another commonly used strain, is a slow methanol-utilizing strain. Detailed descriptions and comparative characteristics of these strains have been summarized in previous studies [2,23].

2.2. Promoters

Promoters are essential to initiate gene expression. The endogenous alcohol oxidase 1 promoter of K. phaffii, P_AOX1, exhibits substantially higher transcriptional activity than many other native promoters, and, therefore, has been widely used [24,25]. P_AOX1 is induced by methanol and repressed by other carbon sources, such as glucose and glycerol [26]. Its activity is regulated by multiple transcription factors. Activators, such as Mit1p, Prm1p, and Mxr1p, enhance P_AOX1 activity, whereas repressors, such as Nrg1p, Mig1p, and Mig2p, suppress its transcriptional output [27,28].

Promoter engineering is an effective strategy to improve gene expression. Hartner et al. constructed a deletion mutant library of P_AOX1 and identified key regions that significantly influenced promoter activity [29]. Based on these findings, Li et al. generated a P_AOX1 variant by deleting the −777 to −712 region and duplicating the −203 to −190 fragment. This engineered promoter drives significantly higher phytase expression than the wild-type P_AOX1 [30]. Huang et al. [31] engineered another endogenous promoter, P_CAT1, by reintroducing predicted transcription factor-binding motifs at their original positions. The resulting P_CAT1-P10 variants showed markedly improved secretion of the lipase CALB and green fluorescent protein compared with both wild-type P_CAT1 and P_AOX1 (Table 1).

Table 1.

List of strong inducible and constitutive promoters for recombinant protein production in Komagataella phaffii.

Promoter	Description	Control	Relative expression level (fold)	Reporter protein	Ref.
AOX1	K. phaffii endogenous alcohol oxidase gene promoter	-	-	-
AOXm	AOX1 promoter partial deletion and duplication variants	AOX1	1.4	phytase	[30]
CAT1	Strong methanol-inducible promoter	AOX1	1.6	D-LDH	[32]
CAT1-P4	CAT promoter RAP1.05-binding-site repeat variants	AOX1	1.3	CALB	[31]
CAT1- P10	CAT promoter RAP1and Hap1.01-binding-site repeat variants	AOX1	1.5	patatin	[31]
GAP	K. phaffii endogenous constitutive promoter	-	-	-
GAP-G1	GAP promoter variant	GAP	19.6	EGFP	[33]
GCW14	K. phaffii endogenous constitutive promoter	GAP	10	EGFP	[34]
GCW14-M + 20	GCW14 promoter variant	GCW14	1.3	CALB	[35]

The constitutive promoter P_GAP is another commonly used native promoter in K. phaffii, although its activity is generally lower than that of P_AOX1 [36,37]. P_GAP is mainly used in fermentation using non-methanol carbon sources, such as glycerol and glucose. Recent studies have also focused on the engineering and determining constitutive promoters to improve recombinant protein production under non-methanolic conditions. A mutant library of P_GAP generated via error-prone PCR identified a variant that increased green fluorescent protein expression by 19.6-fold compared to the wild-type P_GAP [33]. Liang et al. identified another promoter in the K. phaffii genome, P_GCW14, that exhibited a significantly higher capacity for recombinant protein expression than P_GAP, providing additional options for expression optimization [34]. Building on these results, Zhang et al. further characterized the key regions affecting P_GCW14 activity using multiple strategies and constructed a mutant variant, P_GCW14−M + 20. This engineered promoter showed a 33% increase in the expression of Candida antarctica lipase B compared with that of the wild-type P_GCW14 promoter [35]. These constitutive promoters are methanol-independent and typically utilize safe carbon sources, such as glycerol and glucose, during fermentation, making them suitable for the production of food-grade and therapeutic proteins (Table 1).

2.3. Signal peptides

The ability of K. phaffii to secrete recombinant proteins into the extracellular medium greatly simplifies the downstream purification needed for industrial applications. Secreted proteins require a signal peptide for guidance, and the most widely used and well-established signal peptide in the K. phaffii expression system is the S. cerevisiae α-factor signal peptide [38]. The α-factor signal peptide corresponds to the N-terminal prepro-leader sequence of the mating pheromone secreted by S. cerevisiae α-cells, mating factor 1 (MF1). It consists of 86 amino acids, with residues 1–19 forming the pre-sequence and residues 20–86 forming the pro-sequence [39]. The secretion of recombinant proteins directed by the S. cerevisiae α-factor leader undergoes standard post-translational translocation, endoplasmic reticulum (ER)-to-Golgi vesicular trafficking, and sequential proteolytic processing by the Kex2p and Ste13p proteases [40,41].

The commonly used α-factor signal peptide typically contains two mutation sites (L42S and D83E), originating from commercial plasmids developed by Invitrogen (USA). Compared to the wild-type sequence, this engineered version shows enhanced secretion efficiency and contains a XhoI restriction site. During gene cloning, restriction enzymes, such as EcoRI and NotI, are often used for vector linearization, which may introduce extra residues between the signal peptide and target protein. Even after complete signal peptide processing, two additional amino acids may remain at the N-terminus of the recombinant protein. In contrast, the XhoI site is located near residues 82–83 of the α-factor signal peptide. By using XhoI-based cloning, only a PCR-introduced XhoI site and an AAAAGA sequence (which restores the Kex2 recognition site KR) need to be added at the 5′ end of the target gene, enabling the production of proteins without extra N-terminal residues [42,43].

Aza et al. introduced the mutations A9D and A20T into the α-factor signal peptide (including L42S and D83E) and observed significantly enhanced secretion of multiple enzymes in yeast [44]. Ito et al. reported that a V50A variant of the α-factor signal peptide (also containing L42S and D83E) improved the secretion of an anti-lysozyme scFv antibody in K. phaffii [45]. Based on Jpred secondary structure prediction and three-dimensional modeling, Chahal et al. designed deletion and duplication variants within the prepro-sequence of the α-factor signal peptide; structural modifications were found to enhance the secretion of horseradish peroxidase and lipase [39]. Ahn et al. enhanced recombinant protein expression through synonymous codon optimization of the α-factor signal peptide without altering its amino acid sequence. This strategy specifically tailors individual codon usage and codon context parameters to the host’s translational machinery [46]. Subsequent studies further confirmed that the Ahn-designed α-factor signal peptide significantly enhances recombinant protein secretion [47], highlighting its strong performance.

To facilitate the secretion of structurally complex proteins, Barrero et al. generated a hybrid leader sequence by replacing the pre-region of the α-factor signal peptide with an Ost1 signal sequence from S. cerevisiae. This engineered construct demonstrated remarkable performance, yielding a 10-fold increase in the expression level of the lipase BTL2 compared to its native counterpart [48]. Duan et al. identified several endogenous signal peptides in K. phaffii, showing that some native signal peptides, such as Mbs2 and Dan4, can outperform the α-factor signal peptide for multiple recombinant proteins [49]. Through multi-omics mining encompassing both the secretome and transcriptome of K. phaffii, Shen et al. identified a novel endogenous signal peptide derived from the gene PAS-chr3-0030. For recombinant enzyme production, this newly discovered element exhibited superior secretion efficiency that outpaced the conventional α-factor signal peptide [50] (Table 2).

Table 2.

List of efficient signal peptides for recombinant protein production in Komagataella phaffii.

Signal peptide	Description	Control	Relative expression level (fold)	Reporter protein	Ref.
α-factor SP	α-mating factor secretion signal from Saccharomyces cerevisiae	-	-	-
α-opt	α-factor SP mutant containing A9D, A20T, L42S, and D83E	α-factor SP	20	PeL laccase	[44]
α-mut	α-factor SP mutant containing L42S, V50A and D83E	α-factor SP	8	scFv	[45]
α-factor Δ(57–70)	α factor SP with deletion of amino acids 57–70	α-factor SP	1.7	horseradish peroxidase	[39]
αCC	α factor SP with codon bias optimized for K. phaffii	α-factor SP	4	CALB	[46]
pre-Ost1-pro-α-factor	Chimeric secretion signal with Ost1 pre-region replacing the α-factor SP pre-region	α-factor SP	20	E2-Crimson	[48]
Dan4	K. phaffii endogenous signal peptides	α-factor SP	3.8	SECA	[49]
Gas1	K. phaffii endogenous signal peptides	α-factor SP	16.8	Gal	[49]
Msb2	K. phaffii endogenous signal peptides	α-factor SP	1.9	SECA	[49]
Fre2	K. phaffii endogenous signal peptides	α-factor SP	1.4	yEGFP	[49]
SP0030	K. phaffii endogenous signal peptides	α-factor SP	1.9	mCherry	[50]

2.4. Terminators

Transcriptional termination is highly complex and involves the coordinated action of multiple factors. Terminators are located at the 3′ end of the target gene and function to terminate transcription; they are also referred to as termination sequences. In brief, when RNA polymerase reaches the terminator region, associated proteins are recruited to release the nascent mRNA from the transcription complex and catalyze polyadenylation at the 3′ end of the transcript [51,52].

The AOX1 terminator originates from the endogenous AOX1 gene of K. phaffii and is widely used in protein expression systems; however, studies on its engineering or alternative superior terminators remain limited. Although research on K. phaffii terminators is relatively scarce, multiple studies have demonstrated that terminator sequences regulate the transcription termination efficiency, thereby influencing protein expression. For instance, higher termination efficiency promotes the earlier release of RNA polymerase, facilitating subsequent transcription cycles [53]. The 3′-untranslated region (UTR) defined by the terminator also plays a major role in determining the mRNA half-life, and increased mRNA stability can enhance overall expression levels [54,55].

In a large-scale screening of termination regulatory elements, 72 terminators originating from K. phaffii and S. cerevisiae were evaluated by Ito et al. to assess their effects on EGFP expression. While the majority of S. cerevisiae-derived terminators exhibit functional compatibility within the K. phaffii host, the endogenous AOX1 terminator consistently maintains its position as a top-performing candidate [56]. To identify superior alternatives to the conventional AOX1 terminators, Ramakrishnan et al. investigated an array of endogenous and heterologous terminators for CALB lipase production. Interestingly, their findings revealed that the S. cerevisiae DHAS (Dihydroxyacetone synthase) terminator outperformed its conventional AOX1 counterpart, offering a promising alternative module for driving high-efficiency protein expression systems [57]. Vogl et al. unexpectedly identified a mutated AOX1 terminator generated during PCR amplification, which contained a NotI site; this terminator increased reporter gene expression by 37% compared to the wild-type AOX1 terminator. This improvement may be attributed to changes in GC content and mRNA secondary structure [58]. Moreover, Xu et al. [59] replaced the AOX1 terminator with terminators from the co-translational translocation proteins PMP1p and PMP2p during the expression of recombinant human serum albumin (HSA) and fungal α-amylase in K. phaffii. Together with the replacement of the signal peptide with Ost1-proα, this strategy significantly enhanced protein expression levels. A similar strategy was previously validated in S. cerevisiae [60]. Curran et al. redesigned a yeast terminator by modifying the efficiency element length, adjusting the GC content, and introducing palindromic sequences, resulting in a series of highly efficient synthetic terminators and providing new strategies for terminator engineering in yeast [61]. In addition, given their close evolutionary relationships, insights from terminator studies in S. cerevisiae are highly informative for K. phaffii systems [62,63] (Table 3).

Table 3.

List of efficient terminators for recombinant protein production in Komagataella phaffii.

Terminator	Description	Control	Relative expression level (fold)	Reporter protein	Ref.
AOX1 TT	Transcription terminator of the aox1 gene from K. phaffii	-	-	-
DHAS TT	Transcription terminator derived from the dihydroxyacetone synthase gene of Saccharomyces cerevisiae	AOX1 TT	3	CALB	[57]
AOX1 TT+NotⅠ	Variant of the K. phaffii AOX1 terminator	AOX1 TT	1.4	EGFP	[58]
PMP1 t	3′-UTR from K. phaffii plasma membrane proteolipid (PMP1) gene, for transcription termination and mRNA stability	AOX1 TT	2.3	β-glucanase	[59]
PMP2 t	3′-UTR from K. phaffii plasma membrane proteolipid (PMP2) gene, for transcription termination and mRNA stability	AOX1 TT	1.8	β-glucanase	[59]

Overall, compared to promoters and signal peptides, there have been relatively fewer reports on the development of efficient terminator tools and the elucidation of their regulatory mechanisms in K. phaffii. Transcription termination is a critical step in the regulation of gene expression. In future studies, the design of highly efficient terminators could be pursued by enhancing transcription termination efficiency and improving the stability of the 3′-UTR, thereby providing more powerful tools for expression cassette engineering.

2.5. Increasing gene dosage

After identifying an optimal expression cassette for a target protein, the gene dosage can be increased by constructing multi-copy strains, thereby enhancing the transcriptional output and providing more mRNA templates for translation.

Common strategies for multi-copy construction include the in vitro assembly method [64], antibiotic gradient screening [65], 2A sequencing [66], and CRISPR and Cre/loxP methods. Furthermore, integration into the ribosomal DNA non-transcribed intergenic spacer (NTS) represents a novel strategy for constructing multi-copy strains [67]. The rDNA arrays in yeast genomes contain multiple repeated NTS units. By flanking the expression cassette with rDNA-NTS homologous sequences, multiple integration events can occur simultaneously at rDNA loci, greatly improving the efficiency of multi-copy strain generation. This NTS-based integration approach confers greater stability to the resulting multi-copy strains than the aforementioned in vitro assembly and antibiotic gradient screening methods. In both the construction of multi-expression cassette plasmids and the integration of multiple gene copies at the same genomic locus, the close proximity of identical DNA sequences increases the risk of gene loss during propagation through homologous recombination, thereby compromising stable inheritance [68]. However, it should be noted that the NTS integration method has not yet been widely adopted, and its effectiveness across different reporter proteins remains to be validated.

3. Recombinant protein synthesis and secretion pathways

Recombinant protein synthesis and secretion in K. phaffii involves multiple interconnected processes, including transcription, translation, folding, post-translational modification, ER homeostasis, and vesicular transport. These processes engage nearly 1000 genes and represent a highly complex metabolic network with extensive feedback regulation. This section describes the molecular mechanisms underlying each step of the secretion pathway and summarizes engineering strategies that can enhance recombinant protein expression at each step.

3.1. Transcription

Transcription is the first step in gene expression and generates mRNA from the target gene. The alcohol oxidase genes AOX1 and AOX2 are essential for methanol utilization in K. phaffii, with AOX1 exhibiting substantially higher expression than AOX2. Under non-growth-limiting conditions, AOX1 and AOX2 are strongly repressed by carbon sources, such as glucose, glycerol, and ethanol [69]. However, under methanol-induced conditions, AOX1 mRNA accounts for approximately 5% of the total cellular mRNA. Although it is only one of >5000 genes in the K. phaffii genome, the AOX1 gene encodes a protein that accounts for up to 30% of the total cellular protein content [70]. This exceptionally strong promoter forms the basis for most recombinant protein expression systems in this organism.

The transcriptional activation of P_AOX1 is regulated primarily by Mit1, Prm1, and Mxr1. The methanol signal is first transmitted to Prm1p, which responds through self-activation and rapidly increases its expression. Prm1p subsequently activates the strong expression of Mit1, leading to its accumulation. Mxr1 expression is also induced by methanol. Ultimately, Mit1, Prm1, and Mxr1 bind directly to distinct sites within P_AOX1 and collectively activate its transcription [27,71] (Fig. 2a). In the presence of repressing carbon sources such as glucose or glycerol, transcriptional repressors, including Mig1, Mig2, and Nrg1, inhibit P_AOX1 activity. Nrg1p directly binds to P_AOX1 and competes with Prm1p and Mxr1p for binding sites. In contrast, Mig1p and Mig2p do not bind directly to P_AOX1; instead, they target the promoter of MIT1 (P_Mit1), thereby repressing Mit1 expression and indirectly suppressing P_AOX1 activity [28].

Fig. 2.

The mechanisms underlying transcriptional and translational initiation for recombinant protein expression in K. phaffii. (a) Mechanism of transcriptional initiation of the cargo protein gene driven by the P_AOX1 promoter under different carbon sources. When methanol is used as the carbon source, the transcription factors Mxr1p, Prm1p, and Mit1p bind to the AOX1 promoter, thereby activating transcriptional initiation. The presence of repressing carbon sources, such as glucose or glycerol, leads to the recruitment of transcriptional repressors Mig1p, Mig2p, or Nrg1p to repress the P_AOX1. (b) Mechanism of translational initiation. During translation initiation, eIF2, a eukaryotic initiation factor that binds methionyl-tRNA, assembles with eIF1, eIF1A, eIF3, eIF5, and the 40S ribosomal subunit to form the 43S pre-initiation complex (43S PIC). Subsequently, the 43S PIC is recruited to mRNA with the assistance of eIF4F, eIF4B, and the poly(A)-binding protein Pab1. The eIF4F complex consists of eIF4E, eIF4G, and eIF4A. eIF4E recognizes and binds the 5′ cap structure of mRNA, while eIF4A functions as an RNA helicase that, together with eIF4B, resolves secondary structures near the 5′ untranslated region. eIF4G serves as a scaffold protein and interacts with Pab1. These initiation factors promote the formation of a closed-loop mRNA conformation, in which the 5′ and 3′ ends are brought into proximity. This structure stabilizes mRNA and significantly enhances translation efficiency. Subsequently, the 43S PIC scans the mRNA from the 5′ to 3′ direction with the assistance of eIF1, eIF1A, and eIF5. Upon recognition of the start codon (AUG), conformational changes occur in the complex, leading to dissociation of eIF1, eIF2, eIF3, and eIF5. The 60S ribosomal subunit then joins to form the functional 80S ribosome, initiating translation elongation according to codon sequence until a stop codon is reached.

Based on the regulatory mechanism of P_AOX1, engineered strains with enhanced transcriptional output were developed. Poodeh et al. [72] overexpressed Mit1 and observed a significant increase in eGFP expression in K. phaffii. Bachleitner et al. [73] co-overexpressed Mit1 and Mxr1, leading to a substantial increase in heterologous lactate dehydrogenase expression and, consequently, enhanced lactate accumulation. To decouple recombinant protein expression from methanol induction, a combined genetic engineering approach was executed by Shi et al., involving the simultaneous knockout of transcriptional repressors (Mig1p, Mig2p, and Nrg1p) and overexpression of the activator Mit1p. This genetic rewiring successfully yielded a methanol-independent phenotype capable of producing detectable levels of the AOX1 protein, even when utilizing alternative carbon sources such as glycerol or glucose [74]. Zhu et al. developed an elegant transcriptional signal amplification device to further boost the driving force of this expression system. By fusing a bacterial DNA-binding protein to a transcriptional activation domain and integrating the corresponding binding sequences upstream of the AOX1 core promoter, this artificial platform not only permitted glycerol to activate P_AOX1 but also achieved an approximately five-fold increase in recombinant protein yield relative to the conventional methanol-inducible system [75].

During experimental investigations, the transcript levels of recombinant strains can be used to determine whether the transcription step constitutes a potential bottleneck in protein expression. For example, quantitative PCR (qPCR) analysis can be used to assess the transcription levels of strains harboring different gene copy numbers. If no corresponding increase in transcript levels is observed with increasing gene copy number, one possible explanation is that the transcription of the target gene is limited. In such cases, overexpression of the aforementioned transcription factors may be considered or, alternatively, the codon optimization status of the target gene should be re-evaluated.

3.2. Translation

Translation initiation serves as a pivotal rate-limiting step in heterologous protein synthesis and is heavily dependent on the assembly of the 43S pre-initiation complex (43S PIC) and the formation of a closed-loop mRNA conformation mediated by eIF4F (comprising eIF4E, eIF4G, and eIF4A) and Pab1 (Fig. 2b) [[76], [77], [78], [79]]. To exploit this mechanism for strain optimization, Staudacher et al. overexpressed components of the 43S PIC, including eIF2 and eIF3, in K. phaffii. Neither individual nor combined overexpression of these factors increased the production of the bivalent nanobody vHH. In contrast, the overexpression of closed-loop-associated factors (eIF4E, eIF4A, and eIF4G) and Pab1p significantly enhanced vHH expression: the combined overexpression of these factors resulted in a 2.5-fold increase in vHH yield. The same strategy applied to HSA also led to an approximately two-fold increase in expression, supporting the general applicability of this approach. Interestingly, overexpression of these closed-loop components and Pab1 not only increased protein yield but also elevated vHH mRNA levels. This effect was proposed to result from increased mRNA stability rather than enhanced transcription [80]. Based on these findings, Huang et al. [31] demonstrated that the overexpression of either eIF4E or eIF4G increased patatin production in K. phaffii. Liu et al. [81] further showed that the overexpression of translation factors, including eIF4A, eIF4E, eIF4G, and Pab1p, could individually enhance xylanase production in K. phaffii.

3.3. Nascent polypeptide translocation

During secretory expression, nascent polypeptide chains are directed to the ER by signal peptides. In eukaryotes, two distinct translocation pathways exist depending on the type of signal peptide. One is post-translational translocation, in which the polypeptide is fully synthesized in the cytosol and subsequently translocated to the ER. The second is co-translational translocation, in which the signal peptide emerging from the ribosome is recognized by the signal recognition particle, causing translation to be redirected to the ER membrane, where translocation to the ER occurs concurrently with protein synthesis [82].

In the K. phaffii expression system, the α-factor signal peptide is predominantly used. This signal peptide mediates post-translational translocation. The polypeptide is first completely synthesized in the cytosol and maintained in an unfolded state through interactions with cytosolic molecular chaperones, primarily members of the Ssa or Ssb families. The Sec62/63 complex in the ER membrane serves as a specific receptor for post-translational translocation, recognizing the signal peptide and delivering the substrate to the Sec61 channel (composed of Sec61p, Sbh1p, and Sss1p) while the cytosolic chaperones are released. After entry into the Sec61 channel, the polypeptide translocates to the ER lumen via Brownian motion. Within the lumen, the Hsp70 chaperone, BiP (also known as Kar2 in yeast), is activated by the J-domain of Sec63. BiP hydrolyzes ATP and binds tightly to newly emerging polypeptide segments, functioning as a ratchet that prevents the backward sliding of the polypeptide chain into the cytosol. As additional segments enter the lumen, new BiP molecules continue to bind to and progressively pull the entire polypeptide into the ER. BiP can subsequently bind ATP, release the substrate, and reenter the next cycle of the ratchet mechanism [83] (Fig. 3).

Fig. 3.

Translocation of α-signal peptide-guided nascent polypeptides into the endoplasmic reticulum (ER) via post-translational transport. After the mRNA of the target gene is translated in the cytoplasm, the resulting nascent polypeptide remains unfolded under the protection of molecular chaperones. The Sec62/63 complex on the ER membrane specifically recognizes the α-signal peptide and directs the cargo protein to the Sec61 channel, accompanied by the release of cytosolic chaperones. Upon entry into the Sec61 channel, the activated form of Kar2 (Kar2-ADP) binds to the cargo protein and pulls it into the ER lumen, thereby completing the translocation of the cargo protein from the cytosol to the ER.

To synergistically enhance protein translocation efficiency from the cytosol to the ER, a conceptual “push-pull” model was engineered by Richard et al. through the coordinated co-overexpression of cytosolic chaperones (SSA1, YDJ1, SNL1) and ER lumen chaperones (BiP, LHS1). This clever configuration applied a simultaneous “pushing” force within the cytosol and a “pulling” force inside the ER lumen, effectively mitigating translocation bottlenecks. Consequently, this strategy orchestrated a dramatic five-fold increase in the secretion of antibody fragments (Fabs and scFvs), driving final titers beyond 1.3 g/L in bioreactor fermentation [14].

Recent studies have suggested that the post-translational translocation mode mediated by the α-factor signal peptide may limit translocation efficiency, particularly for structurally complex proteins. Such proteins may partially fold in the cytosol prior to translocation, reducing their ability to enter the ER, thereby limiting their expression. Signal peptide replacement is a common strategy for shifting from post-translational to co-translational translocation. The Ost1 signal peptide is a typical co-translational targeting signal. Substituting the pre-region of the α-factor signal peptide with the Ost1 signal sequence generates a hybrid signal peptide (Ost1(pre)-α-factor(pro)), which has been shown to significantly enhance recombinant protein secretion. Besada et al. [60] reported that this hybrid signal peptide increased the production of a bacterial endoglucanase (CelA) by nearly two-fold and improved lipase BTL2 expression by ten-fold compared with the native α-factor signal peptide. Xu et al. [59] also demonstrated that the Ost1(pre)-α-factor(pro) hybrid signal peptide increased the expression levels of fungal amylase (AmyA), bacterial β-glucanase (BgA), and HSA by 1.3–3.4-fold in yeast.

3.4. Protein folding and modification in the ER

After entering the ER, nascent polypeptides must undergo correct folding and post-translational modifications before proceeding to the downstream processing steps (Fig. 4), as improperly folded proteins tend to aggregate in the ER. The Hsp70 family is one of the most abundant classes of molecular chaperones; they participate in the regulation of polypeptide translocation, folding, and degradation. Hsp40 proteins stimulate Hsp70 activity through their J-domain, promoting ATP hydrolysis and enabling substrate binding. Nucleotide exchange factors subsequently promote ATP binding, leading to substrate release and allowing Hsp70 to reenter its functional cycle. The most well-characterized Hsp70 protein in K. phaffii is Kar2, which binds to the hydrophobic regions of nascent polypeptides to prevent aggregation and facilitate folding [84]. Under the protection of Kar2, the protein disulfide isomerase PDI1 catalyzes the correct disulfide bond formation between adjacent cysteine residues. During this process, PDI1 gets reduced and loses its activity. ER oxidoreductin 1 (ERO1) re-oxidizes PDI1 by accepting electrons, thereby restoring its catalytic activity. ERO1 ultimately transfers electrons to molecular oxygen, generating hydrogen peroxide, and completing the redox cycle [85] (Fig. 4). PDI1 also contributes to the folding of proteins lacking disulfide bonds in a substrate-dependent manner [86]. Numerous studies have shown that the individual or combined overexpression of Kar2, PDI1, and ERO1 can increase recombinant protein secretion in K. phaffii by 2- to 4-fold [[87], [88], [89]].

Fig. 4.

Overview of recombinant protein synthesis and secretion pathways. The target gene is first transcribed in the nucleus to produce mRNA, which then initiates translation under the action of eIFs. The resulting nascent polypeptide enters the ER via post-translational translocation. Initial folding of the nascent polypeptide occurs in the ER, where, under the protection and guidance of Kar2, PDI facilitates the formation of stable disulfide bonds in the cargo protein. The reduced PDI then transfers electrons to ERO1, becoming reoxidized to its active form, with molecular oxygen serving as the final electron acceptor to generate hydrogen peroxide, thereby completing the redox cycle. If the cargo protein is correctly folded and undergoes initial glycosylation, it is subsequently transported to the Golgi apparatus via COPII vesicles. If the cargo protein fails to fold correctly, it gradually accumulates in the ER, and the accumulation of large amounts of unfolded proteins leads to aggregation. Under these conditions, Ire1 on the ER membrane senses the aggregated unfolded proteins and transitions into an active dimeric form, which splices the primary mRNA of the UPR pathway transcriptional activator Hac1 by removing the intron to produce translationally active Hac1 mRNA. Hac1p then activates the expression of various UPR pathway-related genes, such as those encoding molecular chaperones and folding helpers (e.g. Kar2, PDI1), to assist in the refolding of unfolded proteins. If the ER stress caused by the accumulation of unfolded proteins cannot be alleviated, the ERAD pathway is upregulated to target misfolded cargo proteins to the proteasome for degradation. Cargo proteins delivered to the Golgi apparatus undergo final modifications and are subsequently secreted extracellularly via vesicle transport and exocytosis.

For proteins containing N-glycosylation sites, nascent polypeptides are modified in the ER by the oligosaccharyltransferase complex, which transfers a preassembled oligosaccharide (Glc₃Man₉GlcNAc₂) to asparagine residues at specific motifs. This glycan is subsequently trimmed by glucosidase I, glucosidase II, and α−1,2-mannosidase, removing three glucose residues and one mannose residue from the terminal structure. Further glycan processing occurs in the Golgi apparatus [90].

O-linked glycosylation is also initiated in the ER and occurs on the hydroxyl groups of serine or threonine residues without a strict consensus sequence. The O-mannosyltransferase family catalyzes the transfer of mannose from dolichol phosphate to polypeptides. Subsequent modifications are then performed in the Golgi apparatus [91].

3.5. ER homeostasis

In addition to folding assistance and glycosylation, the ER plays a central role in protein quality control and cellular homeostasis. As the rate of polypeptide synthesis increases, more proteins enter the ER. When the load exceeds the capacity of the available molecular chaperones, misfolded proteins accumulate and aggregate. Even with sufficient chaperone availability, structurally complex proteins can be misfolded. The excessive accumulation of misfolded proteins triggers ER stress, which is cytotoxic [92]. To counteract this stress, the ER relies on an evolutionarily conserved signaling network that monitors the protein folding status and maintains proteostasis. These pathways are collectively referred to as the unfolded protein response (UPR). When large amounts of misfolded proteins are detected in the ER lumen, the UPR is activated. Here is what happens. First, the synthesis of the target protein is temporarily reduced to limit further accumulation [93]. Second, membrane biogenesis is promoted, leading to the expansion of the ER volume [94]. Third, the expression of molecular chaperones and folding enzymes is upregulated to enhance folding capacity [95,96]. Fourth, the ER-associated degradation (ERAD) pathway is induced to remove accumulated misfolded proteins (Fig. 4) [97]. Finally, when homeostasis cannot be restored, UPR triggers programmed cell death to eliminate damaged cells [98].

In yeast, Ire1 functions as the primary ER stress sensor that initiates the UPR. Ire1 is an ER transmembrane protein containing a luminal sensing domain and a cytosolic kinase/endoribonuclease domain. Under ER stress, the luminal domain directly recognizes and binds misfolded proteins, inducing conformational changes that promote Ire1 dimerization [99]. Dimerization brings the cytosolic kinase domains into proximity, enabling transautophosphorylation and the activation of the RNase domain. The primary substrate of RNase activity is the mRNA encoding the transcription factor Hac1 [100]. Under normal conditions, the Hac1 mRNA contains an intron that prevents translation. Activated Ire1 specifically removes this intron, enabling the production of the Hac1 protein [101]. Hac1 then translocates into the nucleus and activates the transcription of multiple UPR target genes (Fig. 4), including ER chaperones, such as KAR2 and PDI1, components of the ERAD system, and genes involved in phospholipid biosynthesis that expand ER membrane capacity [94,102]. In addition, active Ire1 can degrade ER-localized mRNAs via regulated Ire1-dependent decay (RIDD), thereby reducing the influx of newly synthesized proteins into the ER and alleviating folding stress [103].

Among the current strategies for improving recombinant protein production in K. phaffii, the overexpression of Hac1 is one of the most widely used approaches. >40 studies have reported increased recombinant protein yields using this strategy, including industrial enzymes [104] and nanobodies [105]. This is likely due to the pleiotropic nature of Hac1, which promotes the expression of multiple genes that are beneficial for protein folding, secretion, and quality control.

ERAD is a central component of ER homeostasis and is enhanced by the UPR. Its primary function is to recognize and eliminate misfolded or unassembled proteins in the ER, retro-translocate them into the cytosol, and degrade them via the ubiquitin-proteasome system. In yeast, three main ERAD pathways exist: ERAD-L, ERAD-M, and ERAD-C. Among these, ERAD-L mediates the degradation of luminal proteins and is a major pathway for secretory protein degradation in yeast [106]. Therefore, this subsection focuses on the mechanism of the ERAD-L pathway. In this pathway, the glycosidases Mns1 and Htm1 first trim the glycan chains of misfolded proteins, exposing α−1,6-mannose residues. These modified proteins are recognized by Yos9, which delivers them to the ER membrane protein Hrd3. Hrd3 then transfers the substrates to the retrotranslocation channel formed by the Hrd1 complex. Der1 is also a component of this complex; it provides an additional substrate recognition route parallel to Hrd3 [107]. The misfolded proteins are then exported from the ER through the Hrd1 channel, where they are ubiquitinated by Hrd1 and subsequently delivered to the proteasome for degradation [108].

Although ERAD is essential for maintaining ER homeostasis, excessive ERAD activity may continuously degrade recombinant target proteins, leading to product loss and reduced expression efficiency. Therefore, a moderate suppression of ERAD could enhance product accumulation. Xu et al. [59] showed that the deletion of Der1 weakened ERAD activity and increased β-glucosidase (BgA production by 40% without affecting cell growth. Wang et al. [109] reported that supplementing the culture medium with nitrogen sources can reduce the ERAD-mediated degradation of target proteins. Under optimized conditions, maintaining ammonium concentrations at 300 mM and casein hydrolysate amino acids at 1.0% (w/v) in bioreactor cultures significantly reduced product loss and increased the production of prolyl endopeptidase (PEP) by 1.48-fold.

The efficiency of protein folding in the ER and maintenance of ER homeostasis are critical determinants of recombinant protein expression, particularly for proteins with complex structures that impose a substantial burden on the ER. Therefore, strategies aimed at enhancing the ER folding capacity, raising the threshold for ER stress tolerance, and promoting ER expansion are beneficial for improving the expression of structurally complex proteins.

3.6. Vesicular transport

After correct folding in the ER, secretory proteins must be transported to the Golgi apparatus for final folding and modification into mature proteins. COPII-coated vesicles mediate protein transport from the ER to the Golgi apparatus. First, the ER membrane protein Sec16 recruits the guanine nucleotide exchange factor Sec12. Sec12 activates Sar1 by converting it into its GTP-bound form. Activated Sar1 then inserts its N-terminal amphipathic helix into the inner leaflet of the ER exit site, increasing membrane curvature and initiating COPII assembly [110]. Subsequently, Sar1-GTP recruits Sec23 and Sec24 to form a heterodimeric pre-budding complex. Sec24 is responsible for cargo recognition and binding [110], either directly or indirectly through Erv29 [111]. The Sar1-GTP/Sec23/Sec24 subcomplex then recruits Sec13 and Sec31. The Sec13/Sec31 coat interacts with the inner layer complex to form the outer cage of COPII vesicles [112]. Sec31 is a key structural component of the polyhedral cage, whereas Sec13 facilitates membrane bending by stabilizing Sec31 assemblies [113]. Upon completion of vesicle assembly, Sec31 stimulates the GTPase activity of Sec23, which converts Sar1-GTP to Sar1-GDP. The loss of GTP-bound Sar1 reduces membrane affinity, leading to membrane dissociation and vesicle scission. COPII vesicles then bud from the ER and are transported to the Golgi apparatus [114].

During transport, the TRAPPI complex binds COPII vesicles through interaction between its Bet3 subunit and Sec23, while Ypt1 is incorporated into the vesicle membrane. TRAPPI catalyzes the exchange of GDP for GTP on Ypt1, after which activated Ypt1-GTP recruits Uso1. Through its flexible structure, Uso1 tethers vesicles to the Golgi membrane [115]. Upon arrival at the Golgi, the COPII coat disassembles, exposing v-SNARE proteins. The v-SNARE proteins on the vesicles pair with t-SNARE proteins on the target membrane with assistance from the SM protein Sly1, forming a trans-SNARE complex that brings the membranes in close proximity, overcomes hydration repulsion forces, and drives lipid bilayer fusion. Finally, the cargo proteins are delivered to the Golgi lumen [116].

After maturation in the Golgi apparatus, cargo proteins are transported to the extracellular space via vesicular trafficking mechanisms that resemble, but are not identical to, COPII transport. After uncoating, the vesicles undergo Rab GTPase conversion. Ypt31/32 (homologs of mammalian Rab11) is first activated and then recruits the guanine nucleotide exchange factor Sec2. Sec2 then activates the central Rab protein, Sec4 (a homolog of Rab8), converting it into its active GTP-bound form (Sec4-GTP). Sec4-GTP then directly recruits the type V myosin motor protein Myo2p. Myo2p transports secretory vesicles along actin cables nucleated by formin proteins (Bni1p/Bnr1p) to the plasma membrane, typically at the budding site or regions of active secretion, driving ATP-dependent directional transport toward the cell surface [117].

At the plasma membrane, vesicles are tethered to the exocyst complexes. The exocyst is an octameric complex comprising Sec3, Sec5, Sec6, Sec8, Sec10, Sec15, Exo70, and Exo84. Sec15 directly interacts with Sec4-GTP, while Exo70 and Sec3 bind to activated Rho GTPases such as Cdc42 and Rho3 at the plasma membrane [118]. Tethering brings the vesicle and plasma membranes into close proximity, but membrane fusion requires additional energy to overcome the hydration repulsion. This energy is provided by the SNARE machinery.

Vesicles carry v-SNARE proteins (Snc1p/Snc2p), whereas the plasma membrane contains t-SNAREs composed of the transmembrane proteins Sso1p/Sso2p and the peripheral membrane protein Sec9p. The SM family protein Sec1 facilitates the assembly of v-SNARE and t-SNARE complexes. Their α-helical domains intertwine in a zipper-like manner to form a highly stable trans-SNARE complex. Progressive N-to-C terminal “zippering” releases substantial free energy, which pulls the two membranes together, drives fusion, and ultimately forms a fusion pore. The vesicle lumen then becomes continuous with the extracellular environment, enabling the secretion of cargo proteins [119].

Shi et al. [120] reported that the overexpression of Sec12 in K. phaffii increases glucose oxidase production by nearly 30%. Wu et al. [121] showed that the overexpression of the COPII components Sec23, Sar1, and Erv29 increases the secretion of leghemoglobin by 48%, 28%, and 39%, respectively. Guan et al. [87] demonstrated that the overexpression of Sec1 or Sly1 increased the secretion of IL2-HSA fusion proteins by 2.5-fold and 1.9-fold, respectively. Liu et al. [122] found that overexpression of the Golgi vesicle core component Sec4 increased glucoamylase production by 2.5-fold. These findings indicate that enhancing the vesicular trafficking capacity can effectively improve recombinant protein secretion.

Collectively, this evidence shows that the secretory expression of recombinant proteins relies on sustained vesicular trafficking within cells. Efficient vesicular transport not only facilitates the secretion of the target protein but also helps alleviate the burden on the ER.

3.7. Golgi-mediated modification of secretory proteins

After arrival at the Golgi apparatus, cargo proteins generally undergo basic three-dimensional folding. The primary role of the Golgi apparatus is to fine-tune post-translational modifications (mainly glycosylation) and quality sorting to ensure that only properly processed proteins are directed to their final destination.

Secretory proteins, particularly those expressed using the α-factor signal peptide system in K. phaffii, require proteolytic processing to become mature and active. Kex2p is a Golgi-localized membrane-associated protease that recognizes and cleaves dibasic motifs (e.g., Lys-Arg), removing the N-terminal propeptide from cargo proteins [123]. Overexpression of Kex2 was investigated by Sun et al. to optimize the processing efficiency of N-terminal signal sequences within the Golgi. This targeted genetic modification significantly enhances leader peptide cleavage, subsequently boosting the production of bovine lactoferrin, a high-value functional food protein [124]. A similar strategy was reported by Huang et al. [125], who demonstrated the improved expression of the proprotein convertase furin in K. phaffii. Aggarwal et al. [126] found that the substitution of Glu at the P1’ position of the Kex2 cleavage site on the α-factor signal peptide with Val or Ala increased cleavage efficiency, thereby enhancing the expression of human granulocyte colony-stimulating factor in K. phaffii.

In the ER, N-glycosylation produces a core oligosaccharide structure of Man₈GlcNAc₂. In the Golgi apparatus, following a series of glycosylation events, a highly glycosylated structure containing >100 mannose residues is ultimately generated [127]. This extensive glycosylation explains why secreted proteins expressed in K. phaffii often display apparent molecular weights that are significantly higher than their theoretical values on SDS-PAGE. Similar to N-glycosylation, O-glycosylation also occurs in the Golgi apparatus and plays important roles, including protection against proteolysis [128]. Unlike N-glycans, the O-glycan chains in K. phaffii are relatively short, typically consisting of only a few mannose residues, and are even shorter than those observed in humans [129].

Recombinant proteins produced by K. phaffii often exhibit extensive glycosylation, which differs substantially from that observed in human and mammalian cells. These differences can induce strong immunogenic responses when recombinant therapeutics or food proteins are used. Consequently, humanized glycosylation engineering in K. phaffii has attracted increasing attention. Current strategies include the deletion of endogenous glycosylation enzymes and introduction of human glycosylation pathway genes [130,131].

To facilitate reader reference, the genetic engineering strategies involved in different stages of protein expression in K. phaffii are summarized in Table 4.

Table 4.

Summary of factors involved in the regulation of protein synthesis and secretion pathways for enhanced recombinant protein production in Komagataella phaffii.

Protein synthesis-related factors	Process	Description	Relative expression level (fold)	Reporter protein	Ref.
Mit1	transcription	AOX1 promoter transcription activator	3.2	EGFP	[72]
Mxr1		AOX1 promoter transcription activator	1.8	lactate dehydrogenase	[73]
ΔMig1		Mit1 promoter transcription inhibitor	The combined deletion of Mig1, Mig2, and Nrg1, together with the overexpression of Mit1, relieves the glucose- and glycerol-mediated repression of the AOX1 promoter		[74]
ΔMig2		Mit 1 promoter transcription inhibitor
ΔNrg1		AOX1 promoter transcription inhibitor
elF4E	translation	mRNA 5′ cap-binding protein	1.2	nanobody	[80]
elF4A		ATP-dependent DEAD-box RNA helicase that resolves secondary structures in the 5′-untranslated region (UTR) as a core subunit of the eIF4F complex	1.6	nanobody	[80]
elF4G		Scaffold protein that bridges eIF4E, eIF4A, and Pab1 to promote mRNA recruitment and circularization	2	nanobody	[80]
Pab1		Poly(A)-binding protein that binds the mRNA poly(A) tail and interacts with eIF4G to stabilize the closed-loop mRNA complex for efficient translation	2	nanobody	[80]
Ssa1*	folding	Cytoplasmic molecular chaperone	5	scFvs. (Co-overexpression of Ssa1, YDJ1, SNL1, Bip, and LHS1)	[14]
YDJ1*		Cytoplasmic molecular chaperone
SNL1*		Cytoplasmic molecular chaperone
Kar2		Endoplasmic reticulum (ER)-resident Hsp70 chaperone that facilitates protein folding and translocation into the ER	1.9	human albumin	[87]
PDI1		Multifunctional oxidoreductase that catalyzes the formation and rearrangement of disulfide bonds in the ER	2.2	human albumin	[87]
ERO1		ER membrane-anchored sulfhydryl oxidase that re-oxidizes PDI1, providing the oxidative power for disulfide bond formation	2.3	human albumin	[87]
HAC1	protein homeostasis	The unfolded protein response (UPR)-activated transcription factor	15	Nanobody molecule A	[105]
ΔDer1		A subunit of the Hrd1 complex; Hrd1 serves as the retrotranslocation channel for unfolded proteins in the ERAD pathway	1.4	β-glucosidase	[59]
ΔPep4		Endogenous protease of K. phaffii	0 to 3 g/L	human type II collagen	[132]
Sec12	vesicle transport	Guanine nucleotide exchange factor (GEF) that activates Sar1 to initiate COPII vesicle formation at the ER membrane	1.3	glucose oxidase	[120]
Sec23		COPII coat component that helps package proteins into transport vesicles	1.5	leghemoglobin	[121]
Sar1		Small GTPase that initiates COPII vesicle formation at the ER membrane	1.3	leghemoglobin	[121]
Erv29		Cargo receptor that sorts specific proteins into COPII vesicles for ER export	1.4	leghemoglobin	[121]
Sec1		A vesicle trafficking protein that regulates SNARE complex assembly to promote membrane fusion	2.5	human albumin	[87]
Sly1		An SM protein that binds to syntaxin family SNAREs to regulate ER-Golgi vesicle fusion	1.9	human albumin	[87]
Sec4		A Rab-family GTPase that regulates post-Golgi vesicle transport to the plasma membrane	2.5	glucoamylase	[122]
Kex2	Golgi processing	A Golgi-localized membrane-bound protease that recognizes and cleaves dibasic amino acid sites	2.9	bovine lactoferrin	[124]
CAT1	oxidative stress	Catalase	1.3	lipase	[133]
SOD1		Superoxide dismutase, Involved in antioxidant activity	1.4	lipase	[133]
GLR1		Glutathione reductase, Involved in antioxidant activity	1.5	lipase	[133]
GSH2		Glutathione synthetase, involved in antioxidant activity	1.7	lipase	[133]
Mpr1		N-acetyltransferase, involved in antioxidant activity	1.2	alpha-glucosidase	[134]

The factors listed in this table cover the entire process from transcription to secretion. Factors without the "Δ" prefix indicate overexpression, whereas those with the "Δ" prefix indicate gene deletion. Factors marked with an asterisk (*) are derived from heterologous hosts, whereas those without an asterisk are endogenous to K. phaffii.

4. Other optimization strategies in strain construction

In addition to the regulatory strategies targeting the protein synthesis pathway described above, multiple strain engineering approaches have been developed to achieve high-level recombinant protein expression in K. phaffii. These include the use of fusion tags and deletion of proteases.

The E. coli maltose-binding protein (MBP, ∼370 amino acids) has been widely validated as a solubility-enhancing fusion tag. When fused to a target protein (MBP at the N-terminus and the target protein at the C-terminus), it increases secretion levels by approximately 0.5–2-fold. This effect is mainly attributed to the strong hydrophilicity of MBP, which helps prevent the aggregation of partially folded cargo proteins and facilitates their correct folding into higher-order structures. In addition, MBP may promote higher expression efficiency owing to efficient transcription and translation initiation, thereby indirectly enhancing the production of cargo proteins [135,136].

Similar to MBP, the small ubiquitin-related modifier (SUMO) is another commonly used solubility-enhancing fusion tag. SUMO is a small ubiquitin-like protein that functions as a highly soluble chaperone-like protein. The fusion of SUMO to target proteins increases protein stability and solubility and is widely used for recombinant protein expression in K. phaffii [137,138].

Insulin-like growth factor 1 (IGF-1) is difficult to express in K. phaffii due to its small size (70 amino acids), which makes it prone to proteolytic degradation. To circumvent this, Lu et al. fused IGF-1 with a highly expressed protein, the xylanase XynCDBFV, using a “high-expression carrier driving low-expression cargo” strategy. This approach increased IGF-1 production to 1 g/L. In this strategy, a Kex2 cleavage site was introduced between the fusion partners, enabling intracellular self-cleavage so that the secreted IGF-1 is released without a residual xylanase sequence [139].

For very small peptides, such as antimicrobial peptides (typically 20–40 amino acids), efficient secretion in yeast is often difficult owing to their instability and degradation. Fusion with highly expressed partner proteins significantly enhances protein production. This strategy has been shown to be effective for peptides shorter than 40 amino acids [140].

K. phaffii contains multiple endogenous proteases that can degrade recombinant proteins, resulting in reduced production efficiency. This is particularly relevant for the vacuolar proteases Pep4 and Prb1. The mis-sorting of cargo proteins into vacuoles or the release of vacuolar enzymes during cell lysis can lead to the degradation of secreted products. Wang et al. observed the complete degradation of human type II collagen (Col II) after 72 h of induction in a 5 L bioreactor; the deletion of Pep4 reversed this phenotype and increased production to 3 g/L [132]. Ruan et al. used RNA to suppress Prb1 expression, resulting in a 33% increase in EGFP expression [141]. Lukas et al. further showed that the disruption of CORVET or HOPS complexes reduces vesicular trafficking from the Golgi to the vacuole, thereby preventing mis-sorting and degradation of cargo proteins [142,143].

During methanol metabolism, K. phaffii generates hydrogen peroxide, a reactive oxygen species (ROS). The excessive accumulation of ROS causes cellular damage and limits cell growth. In addition, extensive protein folding in the ER generates ROS via the PDI-ERO1 electron transfer cycle (see Section 3.4). To counteract oxidative stress, cells activate a series of responses to maintain redox homeostasis, collectively referred to as the oxidative stress response. Yap1 is a key transcription factor that responds to oxidative stress. When intracellular ROS levels increase, Yap1 transiently translocates to the nucleus and activates genes involved in the glutathione redox system, thereby reducing ROS accumulation [144].

Lin et al. individually expressed several antioxidant-related genes in K. phaffii including catalase (CAT1), superoxide dismutase (SOD1), glutathione reductase (GLR1), and glutathione synthetase (GSH2). These factors significantly reduced intracellular ROS levels during methanol induction and improved lipase expression [133]. Similarly, Wu et al. showed that Mpr1 overexpression enhanced cellular antioxidant capacity, reduced ROS accumulation, and increased α-glucosidase production [134].

5. Fermentation optimization

After strain construction is completed, cultivation scale-up in bioreactors is required to determine the optimal fermentation strategies and simulate industrial production conditions, representing the final step toward industrial applications. This section focuses on bioreactor-based fermentation processes and the related optimization strategies. Specifically, major bioreactor optimization strategies include medium optimization, temperature, and pH control, mixed carbon source feeding, and process control.

K. phaffii is typically cultivated in bioreactors using a basal medium composed of inorganic salts. This method is simple and cost-effective. Although buffered methanol-complex medium (BMMY) is nutrient-rich, its high cost makes it unsuitable for large-scale bioreactors.

Fermentation generally comprises two phases. The first phase is the glycerol growth phase, during which glycerol serves as a carbon source and cells grow rapidly, accumulating biomass for subsequent induction. The second phase is the methanol induction phase, in which methanol activates the AOX1 promoter and drives the high-level expression of the target protein.

Based on transcriptomic analysis, Matthews et al. [145] investigated the physiological and metabolic responses of K. phaffii during high-density fermentation and designed a novel medium (RDM) based on cellular nutrient and metabolite preferences. This medium reduces salt content and includes additional components, such as glutamine and lipids. The cost of this medium was lower than that of the conventional BSM. Across multiple comparisons, RDM resulted in higher recombinant protein yields than both BSM and BMMY media.

Methanol undergoes a high degree of reduction and generates a substantial amount of heat during catabolism. When used as the sole carbon source, a large fraction of methanol must be metabolized via dissimilatory pathways to meet the energy demands, producing formaldehyde and hydrogen peroxide while consuming large amounts of oxygen. This imposes a significant metabolic burden on cells. Therefore, mixed carbon source feeding is an effective strategy. Sorbitol and glycerol can be co-fed during methanol induction [146]. In the presence of these carbon sources, cells preferentially utilize them as energy and carbon sources, thereby reducing their reliance on methanol dissimilation. This shifts methanol metabolism toward assimilation pathways, improving protein expression while reducing heat generation and oxygen demand. Numerous studies have confirmed that mixed feeding is a widely applicable and effective strategy [147,148].

Some recombinant proteins are sensitive to proteolytic degradation during the late fermentation stages, particularly small peptides, such as antimicrobial peptides. Sang et al. [149] supplemented 2% casamino acids into the culture medium, which acted as a competitive substrate for proteases and effectively prevented the degradation of antimicrobial peptides during fermentation; the addition of surfactants, such as Tween-80, improved oxygen transfer in the medium and increased membrane fluidity, thereby enhancing protein secretion [150,151]. However, it should be noted that additives, such as casamino acids or Tween-80, may impose additional burdens on downstream purification processes. Therefore, the selection of such strategies should be carefully weighed against actual industrial costs.

Process control is critical for fermentation performance. Key parameters include temperature, pH, the oxygen uptake rate, respiratory quotient, feeding rate, aeration/oxygen flow, and bioreactor pressure; heat generation rate has received less attention compared with these commonly monitored variables. Katla et al. [152] proposed that methanol metabolism and biomass synthesis in K. phaffii generate substantial heat, which serves as an important process signal that reflects cellular physiological states in real-time. Based on this concept, they developed a heat-monitoring system to regulate the carbon feed rates and improve the product yield.

Balancing growth and production is essential. Gautam et al. [153] showed that different growth rates at different fermentation stages can maximize product synthesis. Within a certain range, reducing and maintaining moderate growth rates improves recombinant protein production efficiency and increases the carbon source conversion efficiency.

As described earlier, ROS accumulation occurs throughout the fermentation process and can cause irreversible cellular damage and reduce productivity. Jia et al. [154] applied a cyclic pulse feeding strategy to the glycerol phase. Each cycle consisted of maintaining a low glycerol concentration with 20% dissolved oxygen, followed by a short high-glycerol phase with dissolved oxygen reduced to 0%. After five cycles, intracellular ROS levels were significantly reduced, biomass accumulation increased, and human lysozyme production doubled compared with conventional feeding strategies.

The balance between the specific growth rate during methanol induction, intracellular ROS accumulation, and final protein yield was demonstrated by Hu et al. Specifically, although a high specific growth rate (0.065 h⁻¹) accelerated early stage biomass and product formation, it concomitantly triggered excessive ROS accumulation, ultimately suppressing protein production in later phases. In contrast, maintaining a moderate growth rate (0.05 h⁻¹) effectively mitigated ROS toxicity, thereby achieving the maximal recombinant protein yield [155].

6. Artificial intelligence (AI)-enabled strain engineering and fermentation optimization

Traditional optimization of the K. phaffii expression system has largely relied on extensive trial-and-error strain engineering, including random combinations of expression cassette elements or empirical overexpression of folding-related factors to enhance the secretion levels of target proteins, where a large number of fermentation batches are often required to identify optimal cultivation conditions. However, with a deeper understanding of the K. phaffii metabolic network and the rapid development of AI technologies, machine learning and genome-scale metabolic modeling have enabled the treatment of cellular physiological processes as predictable and modelable complex systems. This shift overcomes the limitations of labor-intensive, time-consuming, and costly experimental workflows, marking a transition from experience-driven to intelligence-driven bioprocess optimization.

6.1. Machine learning-guided protein design and sequence prediction

Several studies have reported the application of machine learning in enzyme engineering to improve catalytic activity, thermal stability, and substrate specificity. This section focuses on approaches that predict protein expression levels from amino acid or DNA sequences as well as strategies that enhance expression through sequence optimization.

High-quality datasets are fundamental for building accurate machine learning models; however, large-scale datasets are often difficult to obtain in biological experiments. Nikolados et al. [156] systematically evaluated the impact of dataset size and sequence diversity on model performance, addressing the reliance of deep learning approaches on large training sets. Their results showed that with properly designed sequence diversity, approximately 1000–2000 training samples were sufficient to build highly accurate protein expression prediction models for both bacteria and yeast.

To address the persistent challenge of predicting the efficiency of recombinant protein secretion in K. phaffii, an advanced machine learning framework named Presep was developed by Tian et al. Utilizing a random forest algorithm, this model enabled researchers to evaluate the secretion propensity prior to wet-lab experimentation. Remarkably, the predictions demonstrated a striking correlation with experimentally validated secretion levels (R² = 0.967), thereby establishing a robust predictive tool for host suitability assessments [157].

By integrating pretrained protein language models with transfer learning, Liu et al. engineered a cross-species computational suite comprising MPB-EXP for expression prediction and MPB-MUT for mutational design to guide protein optimization across 88 species. The practical efficacy of this system was exemplified using the baking-related enzyme xylanase Pm10868. While the wild-type protein completely failed to express in E. coli, three mutants designed via MPB-MUT achieved successful soluble expression, with the top-performing variant exhibiting a notable enzymatic activity of 4554.59 U/L [158].

Ding et al. constructed a mutation predictor named MPEPE by harnessing deep learning architectures trained on an extensive dataset of 6438 heterologous proteins. Big data analysis revealed that highly expressed proteins were significantly enriched in charged amino acids and fast-translating codons. When applied to optimize industrially important food enzymes—specifically the laccase 13B22 and glucose dehydrogenase FAD-AtGDH—this strategy successfully guided the design of seven-site combinatorial mutants, which achieved dramatic 12.4-fold and 7.9-fold increases in enzymatic activity relative to their respective wild types.

Tunney et al. [159] used a feedforward neural network to model ribosome profiling data in S. cerevisiae and extracted sequence features associated with translational elongation efficiency. Optimized synonymous codon substitutions enable the prediction of DNA sequences with improved translation efficiency and provide a new strategy for codon optimization.

Protein aggregation is a major disadvantage of low-solubility proteins, because they tend to form intracellular aggregates. Iglesias et al. [160] developed AggreProt, a deep neural network-based server for predicting aggregation-prone regions (APRs). Experimental validation showed a strong overlap between solubility-enhancing mutations and predicted APRs. This study further emphasizes the importance of solvent accessibility, showing that buried and surface-exposed APRs require distinct solubilization strategies that are relevant for rational protein design in high-expression systems.

6.2. Machine learning-guided fermentation optimization

In recombinant protein production, fermentation process optimization and control directly determine product yield, quality, and production cost. Conventional approaches rely on empirical knowledge or statistical designs, such as response surface methodology. These methods are time-consuming and often fail to capture nonlinear interactions and dynamic behaviors in fermentation systems. Deep learning methods have enabled soft sensing, real-time state prediction, feed strategy optimization, and model predictive control, thereby promoting a shift from experience-driven to data-driven fermentation optimization.

Protein titer during fermentation is typically measured offline using ELISA or SDS-PAGE, which is time-consuming and unsuitable for real-time monitoring. To address this, Hosseini et al. [161] developed a feedforward neural network-based soft sensor using data from 38 industrial batches. Three indirect features—specific growth rate, biomass mass, and dry biomass weight—were used as inputs, and the final product titer was used as the output. The model achieved a high prediction accuracy for hepatitis B surface antigen production in K. phaffii (R² = 0.9693), enabling real-time quality monitoring in industrial bioprocesses.

Allampalli et al. [162] addressed the difficulties in controlling the specific growth rate during methanol induction by developing a metabolic heat rate-based soft sensor using a fermentation calorimeter. This enabled real-time estimation and adaptive PID control of growth rate, increasing interferon-α2b production by 2.2-fold compared to that of feedforward control. This shows that metabolic heat signals are effective indicators of key process variables.

To overcome the poor generalization of soft-sensing models caused by batch-to-batch variability, Wang et al. proposed a domain-adaptive ensemble long short-term memory (LSTM) method. In this framework, LSTM autoencoders are utilized to extract unlabeled dynamic features, while ensemble learning successfully incorporates cross-batch information into model loss functions. When applied to monitor the fermentation of the industrial food enzyme inulinase, this intelligent approach achieved an R² > 0.98 for predicting both enzyme production and cell concentration, significantly outperforming conventional LSTM models [163].

Based on the intrinsic correlation between ammonia consumption and biomass formation, an online estimation method for specific growth rate was developed by Tavasoli et al. By deploying a neural network-optimized robust controller, the specific growth rate was precisely maintained at 0.03 h⁻¹, ultimately yielding 324 mg/L of α1-antitrypsin—a 1.5-fold increase over conventional oxygen limitation strategies [164].

Advancing beyond traditional response surface methodologies for culture medium optimization, an artificial neural network-genetic algorithm hybrid model was introduced by Das et al. for the global optimization of tocilizumab production. This evolutionary machine learning approach vastly outperformed conventional statistical methods in terms of prediction accuracy (R² > 0.97) and optimization capability, successfully orchestrating 3.8-fold and 2.9-fold increases in full-length antibodies and Fab titers, respectively [165].

6.3. Simulation of cellular physiology via genome-scale metabolic modeling

Genome-scale metabolic models (GEMs) are computational frameworks used to simulate and predict intracellular metabolic flux. They are constructed based on genome annotation and gene-protein–reaction relationships, converting metabolic networks into stoichiometric matrices. Under steady-state assumptions and environmental constraints, flux balance analysis and linear programming can be used to solve flux distributions, enabling the in silico simulation of cellular metabolism under different conditions. As a key bridge between the genotype and phenotype, GEMs have been widely applied in strain engineering, target identification, media optimization, drug target discovery, and multi-omics integration.

The first genome-scale metabolic model of K. phaffii, iPP668, was introduced by Chung et al. [166] in 2010. It included 668 genes, 1361 reactions, and 1177 metabolites distributed across eight compartments. The model successfully predicted the growth phenotypes under different carbon sources and suggested sorbitol as a co-substrate during methanol induction. In the same year, Sohn et al. [167] developed the PpaMBEL1254 model, comprising 1254 reactions, 1147 metabolites, and 540 genes across eight compartments. Unlike iPP668, this model is the first to explicitly integrate heterologous protein production into a GEM framework. Simulations suggested that moderate oxygen limitation could enhance recombinant protein production, providing theoretical guidance for process optimization.

The architectural evolution of K. phaffii GEMs matured after the introduction of iLC915 by Caspeta et al. (2012). Comprising 1423 reactions, 899 metabolites, and 915 genes partitioned across 7 compartments, this landmark reconstruction markedly improved gene coverage and metabolic completeness, marking the first implementation of the comprehensive methanol utilization pathway and the peroxisomal compartment in yeast GEMs [168].

To reconcile the discrepancies among earlier platforms, a high-fidelity consensus model designated as iMT1026 was subsequently established by Tomas–Gamisans et al. by integrating iPP668, PpaMBEL1254, and iLC915. Encompassing 2035 reactions, 1689 metabolites, and 1026 genes, this unified model successfully incorporated core methanol metabolism and complex glycosylation pathways while correcting oxidative phosphorylation representations, thereby restricting prediction deviations to <11% across diverse carbon sources [169].

This predictive framework was further augmented through the creation of iAUKM by Bose et al., who merged iMT1026 with a RAVEN-based reconstruction to yield an expanded network of 1033 metabolic genes, 1750 metabolites, and 2309 reactions. Notably, this model was exploited to identify genetic targets for bolstering the synthesis of S-adenosylmethionine (SAM)—a high-value nutritional and functional food ingredient; subsequent wet-lab validation confirmed that engineering the inorganic pyrophosphatase pathway enhanced SAM production by 14–16% [170].

Kocabas et al. [171] further developed an enzyme-constrained model, ecPichia, using the GECKO 3.0 framework. It contains 1054 genes, 2378 reactions, and 1776 metabolites across 10 compartments, representing the most comprehensive K. phaffii GEM to date. By integrating enzyme kinetic parameters predicted via deep learning, flux variability was reduced by approximately 500-fold compared with conventional models, significantly improving the predictive accuracy for metabolic engineering applications.

7. Discussion and conclusions

After decades of research and development, the K. phaffii expression system has accumulated a wide range of genetic tools, including diverse host strains, expression plasmid vectors, various expression cassette elements, and CRISPR-based genome editing systems, such as CRISPR-Cas9 [172], CRISPR-Cpf1 [173], and CRISPR-based marker-less integration [174]. The molecular mechanisms underlying protein synthesis and secretion were also systematically characterized, providing a solid theoretical foundation for strain engineering and further optimization. The integration of AI has further transformed recombinant protein optimization from an empirical, trial-and-error-driven process into a predictive, intelligence-driven paradigm, significantly shortening development cycles. Overall, the K. phaffii expression system has evolved into a mature and highly efficient protein production platform—and remains one of the most powerful systems for heterologous protein expression today.

This review systematically summarized the strategies for optimizing protein expression in K. phaffii from multiple perspectives. These strategies can be divided into several sequential steps according to the workflow of expression system engineering. The first step involved sequence optimization of the target gene. Codon optimization can be performed using existing computational tools. In addition, machine learning models or web prediction platforms can be used to identify high-expression variants, enabling the design of amino acid or DNA sequences with improved expression potential while maintaining protein function. The second step involved optimization of expression cassette components. Highly efficient elements including promoters, signal peptides, and transcription terminators can be systematically combined to identify the most suitable expression cassette configuration for a given target gene. Third, the gene dosage was increased. After determining the optimal expression cassette, a multi-copy strain construction can be used to enhance the transcriptional output of the target gene. The fourth step involved the metabolic engineering of hosts. Once multi-copy strains are obtained, further optimization can be achieved by regulating the pathways involved in protein synthesis and secretion, including transcription, translation, folding, and transport. This stage must consider the practical realities of protein expression and requires careful experimental design. For example, when low expression is observed, the transcriptional levels, oxidative stress status, and subcellular localization of the target protein should first be evaluated. If transcription is limited, overexpression of relevant transcription factors may be considered. If elevated oxidative stress is detected, antioxidant pathways can be activated to improve cellular robustness. Subcellular localization analysis can help identify bottlenecks. Predominant cytosolic accumulation suggests limitations in the translocation of nascent peptides, where the overexpression of membrane transport components or molecular chaperones may be beneficial. Predominant protein accumulation in the ER indicates insufficient folding and processing capacity, where the overexpression of chaperones, folding catalysts, and the UPR regulator Hac1 can improve ER processing capacity. In addition, enhancing vesicular transport components may facilitate protein trafficking from the ER to the Golgi apparatus and ultimately to the extracellular space, thereby reducing intracellular accumulation. If protein degradation is observed during fermentation, the disruption of ERAD components or endogenous proteases may be considered. Multi-omics data and genome-scale metabolic models support rational strain engineering. Finally, fermentation is optimized. This stage transitions from shake-flask cultivation to bioreactor-based processes, aiming to identify the optimal fermentation conditions and provide methodological support for industrial-scale production.

It should be noted that, during the application of various strategies to optimize protein expression, careful attention must be paid to whether the increase in yield is accompanied by maintenance of the activity of the target product. In addition, the metabolic differences between small-scale and high-density fermentation should not be overlooked. Strain engineering should not be validated solely in shake-flask cultures because certain expression bottlenecks often manifest only under high-density fermentation conditions. Several factors, such as oxygen mass transfer, shear stress, and downstream purification, also warrant serious consideration during the scale-up process.

Despite the substantial progress in optimizing K. phaffii expression systems, several challenges remain unresolved. The first is the contradiction between the generalizability of optimization strategies and target protein specificity. Successful strategies often show strong protein dependence: a method that is effective for one protein may be ineffective for another. For instance, the overexpression of the UPR regulator Hac1 is widely used for enhancing expression in K. phaffii, yet Liao et al. [175] reported that this strategy failed to improve expression of recombinant β-fructofuranosidase. At present, the optimization of expression cassettes and engineering of secretion pathways still rely heavily on extensive experimental screening to identify protein-specific factors. Future studies may benefit from the development of recommendation systems based on the physicochemical properties of target proteins, including their molecular weight, isoelectric point, hydrophobicity, and structural features, combined with AI to reduce the experimental burden and cost.

The second challenge is the limited availability of machine learning-ready datasets for K. phaffii. In recent years, AI-driven biological approaches have expanded rapidly; however, robust model construction requires large-scale high-quality datasets. For E. coli, numerous datasets have enabled the development of high-performance predictive models [176,177]. In contrast, publicly available datasets for K. phaffii are limited, restricting the integration of AI with K. phaffii expression systems. Future efforts should leverage high-throughput technologies, such as fluorescence-activated cell sorting and droplet microfluidics to build comprehensive datasets for K. phaffii, thereby facilitating AI-assisted strain engineering.

The third challenge is product humanization. The extensive glycosylation machinery in K. phaffii may introduce hyperglycosylation into therapeutic and food proteins, potentially triggering strong immune responses in humans. Therefore, the humanization of glycosylation pathways is of critical importance. Additionally, human collagen adopts a stable triple-helical structure in its native state, whereas K. phaffii lacks the prolyl hydroxylase activity required for collagen formation. Consequently, recombinant human collagen expressed in yeast fails to spontaneously form triple helices. The introduction of prolyl hydroxylase to convert proline residues into hydroxyproline promotes intramolecular hydrogen bonding and enables the formation of functional triple-helical collagen structures [178]. Future research should focus on humanlike PTMs to broaden the applicability of K. phaffii as a host for therapeutic proteins.

Finally, methanol safety is a critical concern. Although methanol utilization is a key advantage of the K. phaffii system, its flammability and toxicity pose significant risks to large-scale industrial processes. In addition, the use of methanol in food-grade protein production raises safety concerns. Recently, Cai et al. [5] developed an innovative transcription signal amplification device for K. phaffii that effectively alleviated methanol dependence. Using glycerol or glucose as the carbon source, this system achieved higher target protein production than the conventional methanol-driven P_AOX1 system. However, their applicability to diverse proteins requires further evaluation. Future studies should continue to develop efficient methanol-independent expression systems to ensure safe industrial production.

CRediT authorship contribution statement

Yue Huang: Writing – original draft, Visualization, Methodology, Investigation, Data curation. Jialiang Wang: Writing – review & editing, Investigation, Data curation. Aoxue Wang: Writing – review & editing, Investigation, Data curation. Liqi Zheng: Writing – review & editing, Investigation, Data curation. Ying Lin: Writing – review & editing, Supervision, Methodology, Conceptualization. Shuli Liang: Writing – review & editing, Supervision, Methodology, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability

• Data will be made available on request.