Heterologous DNA is maintained in host cells on plasmids or as genome-integrated expression cassettes. Although the energetic cost of replicating extra plasmid DNA is deemed negligible in haploid S. cerevisiae (Krogh et al., 2008; Rugbjerg and Sommer, 2019), researchers studying the instability of plasmid copy number in S. cerevisiae in both haploid and diploid strains producing heterologous aprotinin found that diploid cells adapted to burden through plasmid loss (Krogh et al., 2008). In contrast, despite an observed decrease in aprotinin production, haploid strains maintained constant plasmid copy number while adapting to extended growth in minimal medium (Krogh et al., 2008). We expect this to be similar in K. phaffii but to our knowledge, no research is available in this organism on this topic. However, genomic integration is a more common approach in K. phaffii, which we will get back to. Maintaining plasmids, often requires selection marker expression, such as genes encoding for antibiotic resistance or metabolic enzymes, the expression of such drains even more resources from cells and adds to burden (Karim et al., 2013; He et al., 2020; Shen et al., 2021). Therefore, chromosomal integration of heterologous genes without the need for selection markers can help mitigate the fitness cost associated with carrying high-copy number plasmids (Harrison et al., 2012; Vieira Gomes et al., 2018). Alternatively, a weaker promoter driving expression of the selection marker can be utilized, which can help relieve burden as recently reported in K. phaffii (Shen et al., 2021). In this case, a 3.3-fold higher plasmid copy number and significantly higher expression levels of insulin precursor were achieved when employing the 300 bp promoter region upstream 2-deoxyglucose-6-phosphate phosphatase gene (P_Dog2p300), compared to the popular TEF1 promoter (Shen et al., 2021).

In S. cerevisiae production strains, plasmid-based expression and chromosomal integration strategies are both widely employed for heterologous protein production. For K. phaffii, chromosomal integration of expression cassettes is preferred since plasmids are much less stable in this species. However, high-copy integration strains can be genetically unstable, likely as a consequence of increased burden from heterologous protein production (Zhu et al., 2009; Yu et al., 2020). In rare instances, loss of gene copies through recombination events in genetically unstable cells can occur, providing a growth advantage over burdened cells with higher heterologous protein expressions levels (Zhu et al., 2009). Thus, chromosomal integrations provide many advantages for heterologous protein expression in both S. cerevisiae and K. phaffii, but they also present potential challenges if copies of identical sequences or similar genetic elements are used.

The primary biosynthetic cost from high-copy plasmids or chromosomal integrations encoding heterologous genes is protein overexpression rather than the cost of replication (Kim et al., 2001; Harrison et al., 2012; Eguchi et al., 2018; Huang et al., 2018; He et al., 2020; Yu et al., 2020). Consistent with this, a non-linear correlation between protein production and gene dosage is often observed in yeast production strains (Mellitzer et al., 2014; Cámara et al., 2016; Fang et al., 2017; Huang et al., 2018; He et al., 2020). The optimal threshold for gene dosage varies between studies and depends on the protein of interest (Mellitzer et al., 2012). For example, a two-copy strain of K. phaffii producing Rhizopus oryzae lipase performed better (12.73UA-10¹²/cell) than strains with more than two copies (8.19-12.09UA-10¹²/cell) (Cámara et al., 2016). In another study, a 10-copy strain of K. phaffii producing 1.2 g/L Thermomyces lanuginosus xylanase A performed better than 6- and 18-copy strains producing ~0.85 and 0.25 g/L, respectively (Mellitzer et al., 2012). In a third study, K. phaffii production of the porcine insulin precursor reached a maximum of 0.9 g/L at 6 copies, showing approximately the same production titers for 6, 12, and 18 copies (Chen et al., 2017). These findings indicate that cells possess mechanisms for downregulating heterologous genes when reaching a certain gene dosage plateau, a Goldilocks range of maximum expression, above which burden is triggered (Mellitzer et al., 2014).

Strategies for delivering and maintaining heterologous DNA in production strains, as discussed above, correspond to engineering target 1 in Figure 1. Overall, plasmids or integrated genes in high copy number present potential sources of strain instability. Heterologous DNA maintenance also clearly requires cellular resources from the host. However, the biosynthetic cost of protein production from such foreign DNA has a significantly larger impact on triggering a burden state in the production host. This can be divided into the different steps linked to metabolic pathways, which we will discuss in the following sections.

The biosynthetic process of transcription has been reported to be a limiting process for growing S. cerevisiae and K. phaffii under certain conditions, including in phosphate limiting conditions or when grown in standard medium with increasing gene-copy number (Kafri et al., 2016; Cámara et al., 2017; Farkas et al., 2018). Engineering target 1 in Figure 1 also highlights the potential strategies for controlling optimal heterologous gene expression. It is conceivable that burden could be induced when employing strong promoters due to competition from the heterologous gene promoter for essential transcription factors against native promoters, such as the K. phaffii transcription factor Mxr1p responsible for stimulating expression from the AOX1 promoter (Görgens et al., 2001; Cámara et al., 2017). It is equally possible that high expression levels of the heterologous gene drain other transcriptional resources, such as the transcriptional machinery or ribonucleotides, thereby reducing growth rates and biomass yield (Görgens et al., 2001; Cámara et al., 2017; Rugbjerg and Sommer, 2019). Having said that, eukaryotic genomes are widely transcribed, including many rapidly degraded non-coding RNAs from intergenic and anti-sense regions (Jensen et al., 2013), implying additional transcription from a heterologous gene present in one or more copies should have a negligible impact (cellular protein content is 5–10-fold higher than nucleic acid content). Instead, burden from heterologous genes and their strong transcription is more likely indirect since the mRNAs produced might overload costly downstream processes involved in protein synthesis (Saeki et al., 2020). Nevertheless, increased transcriptional activity of recombinant genes likely contributes, however modestly, to burden (Farkas et al., 2018).

Protein translation is a costly process consuming energy, nutrients, and other resources. Overall, this step is reported to be an important limiting process for protein production in yeast (Kafri et al., 2016; Cámara et al., 2017). Although different resources are critical during translation, the most obvious components for this process include ribosomes and amino acyl tRNAs. Intuitively, high expression levels from heterologous genes leaves fewer ribosomes available to translate native proteins, eventually leading to growth defects (Saeki et al., 2020). However, multiple studies in yeast suggest that the major limiting resources during translation are in fact metabolic products like energy and metabolic precursors, with the occupation of the translational machinery representing a minor source of burden (Heyland et al., 2011b; Kafri et al., 2016). Thus, identifying and enhancing the availability of limiting resources for protein synthesis is expected to reduce burden and enhance overall protein production. This critical step corresponds to engineering target 2 in Figure 1.

The availability of certain tRNAs can also be limited during translation. Aminoacyl-tRNAs formation consumes considerable amounts of cellular energy through hydrolyzation of ATP molecules, while GTP molecules are also subsequently required for amino acid polymerization (Schimmel, 1993). We will discuss causative factors of energy limitations later in this review. The copy number for genes encoding different tRNAs controls to some extent the availability of tRNAs for translation. tRNA gene copy numbers encoding specific anti-codons varies almost 20-fold in S. cerevisiae (Chan and Lowe, 2016) and 10-fold in K. phaffii (De Schutter et al., 2009). Thus, codon optimizing a heterologous gene is expected to be important to avoid the depletion of specific tRNAs, especially rare ones. Based on differing tRNA gene copy numbers in S. cerevisiae and K. phaffii, it is however plausible that cognate tRNA availability is a more important consideration for burden mitigation in S. cerevisiae production strains. Indeed, tRNA availability and recharging has been suggested to be a non-limiting factor in burdened K. phaffii (Mellitzer et al., 2014). Mellitzer et al. applied different codon optimization strategies and observed that certain genes resulted in higher titers independent of the promoter used to drive transcription and carbon source used, conditions under which the aminacyl-tRNA pool was otherwise expected to change and thus have an influence on what optimized genes would result in the highest protein titer or burden the translation machinery. Changes in tRNA concentrations can impact cellular mRNA expression profiles by altering mRNA stability (Presnyak et al., 2015). With this parameter also comes translational speed, which can indirectly contribute to burden since fast translation kinetics can cause incorrect protein folding triggering cellular stress responses (Drummond and Wilke, 2008; Yang et al., 2014). For example, slowing translation through the use of less “optimal” codons can help correct protein folding for some difficult to express heterologous proteins (Mellitzer et al., 2014; Yang et al., 2014). On the other hand, Mellitzer et al. observed a 23-fold difference in expression levels for cellobiohydrolase 2 with different synonymous codon changes in K. phaffii (Mellitzer et al., 2014).

Notably, if codon optimality is around the average tRNA adaptation index score [tAI score developed by dos Reis et al. (2004)] of a gene in the specific host organism−0.37 in S. cerevisiae (Eguchi et al., 2018); 0.47 in K. phaffii (Xu et al., 2021)—the translation elongation rate is often not high enough for an overexpressed protein to reach the burden-limit triggering growth defects. Authors using S. cerevisiae as a model organism estimate this to be around 15% of total cellular protein (Eguchi et al., 2018). Similar estimations for K. phaffii have not yet been experimentally determined or modeled. Whether tRNA abundance is a factor that should be considered in burdened yeast cell factories and the extent to which codon optimization is a useful strategy to significantly mitigate burden remains unanswered in most cases.

As highlighted above, heterologous gene expression triggers an increased demand for nucleotides, tRNAs, and amino acids, removing resources from the pool of free intracellular building blocks required for mRNA and protein synthesis (Gonzalez et al., 2003; de Ruijter et al., 2018). By comparing S. cerevisiae metabolite profiles or flux distribution in production strains to wild type, an increased rate of consumed amino acids has been observed (Gonzalez et al., 2003; de Ruijter et al., 2018). This leaves fewer metabolic precursors for downstream pathways like the tricarboxylic acid cycle (TCA) cycle and pentose phosphate pathway (PPP), leading to decreased flux into energy and reducing power formation. This explains decreased growth rates observed and why supplementing cultivation medium with certain amino acids improves heterologous protein production (Görgens et al., 2005; Heyland et al., 2011b; Van Rensburg et al., 2012; Liu et al., 2016; Huang et al., 2017).

Glutamate and glutamine are important building blocks for nascent polypeptides and amino donors for the biosynthesis of other amino acids (Huang et al., 2017). By supplementing these two amino acids in growth medium, their increased availability helps relieve metabolic limitations by contributing also as additional carbon sources (Heyland et al., 2011b; Nie et al., 2014). Addition of glutamine alone to standard defined medium was reported to cause an increase in β-aminopeptidase 3-2W4 BapA from 0.37 to 0.42 mg/g (Heyland et al., 2011b), whereas in another experiment addition of glutamate caused an increase in β-galactosidase (in this case measured as activity) from 14,570 to 20,460 U/g_DCW (Nie et al., 2014). In line with this, RNA-seq revealed that genes involved in de novo biosynthesis of glutamate and glutamine were upregulated despite supplementation, as well as amino group transfer genes to help convert them into other amino acids and genes encoding membrane transporters (Huang et al., 2017). Thus, glutamine and glutamate especially might play a central role in adapting to the changing environment, but also amino acid composition of the heterologous protein produced might change the need and limitations in the burdened host cells. Huang and coworkers also found a significant upregulation of YCT1 and ERC1 gene transcription encoding cysteine transporters. In their study they produced a-amylase which amino acid composition has a 9.3-fold higher requirement for cysteine compared to that of the average yeast cell protein (Huang et al., 2017) indicating amino acid composition can affect limitations related to amino acids. On the other hand, the authors of another study, producing insulin precursor from S. cerevisiae, found that heterologous protein production had an impact on the cellular free amino acid pool, but that this had no correlation with the relative amino acid abundance in native yeast protein and insulin precursor protein (Kazemi Seresht et al., 2013). Thus, cells appear to adapt to changing demands from heterologous protein production by enhancing both amino acid uptake via membrane transporters as well as de novo synthesis.

Glutamate addition changes flux distributions of central carbon metabolism and increases the pool of intermediates in the TCA cycle to improve energy generation in K. phaffii (Nie et al., 2014; Liu et al., 2016). Notably, TCA cycle activity appears to react differently in K. phaffii and S. cerevisiae. For K. phaffii, an upper limit of absolute TCA cycle activity to catabolize carbon was reported (Heyland et al., 2011b). Thus, the capacity of the TCA cycle is not always robust enough to catabolize sufficient carbon to meet the increased demands for energy, which will limit growth rates and protein production in K. phaffi (Heyland et al., 2011b). On the contrary, the relative TCA cycle flux for S. cerevisiae correlates with glucose uptake (Blank and Sauer, 2004). Therefore, S. cerevisiae might not have such an upper limit as has been reported for K. phaffii, highlighting a potentially important metabolic difference to consider when engineering more resilient strains of each organism.

In summary, amino acids play an essential role in protein production as both building blocks as well as additional carbon sources for central carbon metabolism. Evidence supports supplementation of certain amino acids in growth medium to alleviate burden in yeast cell factories, as indicated for engineering strategy 5 in Figure 1. Particularly, glutamate and glutamine enhance protein production by relieving metabolic limitation as an added nitrogen and carbon source. However, due to the economic cost of using such supplementations in large-scale bioreactors (Heyland et al., 2011b), this process engineering strategy can be suboptimal at industry-level. Instead, genetic engineering strategies to overcome or bypass such metabolic limitations likely offer more cost-effective alternatives to mitigate burden at this stage in the protein production process.

High demand for amino acids can also contribute to establishing a burden state in production cells due to limitations in the redox cofactor nicotinamide adenine dinucleotide phosphate (NADPH). NADPH is required in many anabolic processes, including de novo synthesis of amino acids, and demand for this redox cofactor predictably rises in production strains (Zahrl et al., 2019).

NADPH can be generated through two main pathways: the oxidative phase of PPP or the acetate formation pathway (Grabowska and Chelstowska, 2003; Kwolek-Mirek et al., 2019). Primary generation of NADPH via PPP was reported for burdened K. phaffii production strains where acetate did not accumulate despite increased demands for NADPH (Nie et al., 2014). Therefore, upregulating genes in the oxidative phase of PPP might be expected to help cells meet the growing demand for NADPH. Such approaches are highlighted as engineering strategy 6 in Figure 1 and have been confirmed in K. phaffii strains producing human superoxide dismutase, whereby enhancing the first two steps in the oxidative phase of PPP proved to increase production by 3.8-fold (Nocon et al., 2016). However, it did not alleviate the observed decreases in specific growth rate and biomass yield. This potential solution might not work in burdened S. cerevisiae strains since RNA-seq and proteomic analyses show that genes involved in PPP are not upregulated, but rather downregulated (Huang et al., 2017; Wright et al., 2020). Instead, increased acetate formation has been reported from an insulin-producing S. cerevisiae strain, indicating preferential NADPH production via the acetate formation pathway to meet increased demand resulting from augmented amino acids needs (Wright et al., 2020). Others however argue that S. cerevisiae production strains meet NAPDH demands by simply reducing biomass yield in general (Huang et al., 2017).

In contrast to S. cerevisiae, researchers studying protein production in K. phaffii observed an increase in biomass yield (from 0.39 to 0.49 g_CDWgglucose-1), and a decrease in growth rate (from 0.3 to 0.18h⁻¹) and glucose uptake rate (from 4.2 to 2.2 mmol g⁻¹h⁻¹) for BapA production strains compared to the reference strain (Heyland et al., 2011b). The PPP flux correlated with biomass yield, whereas correlations between PPP and protein production were low. This indicates a higher demand for NADPH due to an increase in biomass synthesis rather than protein production. In this context, it is also important to mention that flux through PPP also depends on substrates used for cultivation. In general, methylotrophic yeasts like K. phaffii have retained high PPP activity during evolution (Riley et al., 2016). The assimilation of methanol requires a higher carbon flux through PPP (Jordà et al., 2012) since the ketose sugar PPP intermediate, xylulose 5-phosphate, is needed to drive this reaction. Thus, increased PPP flux does not only reflect higher NADPH demand in cases where methanol is used during cultivation (Nie et al., 2014). Meanwhile, S. cerevisiae is unable to grow on methanol, using simple sugars such as glucose as carbon sources.

In summary, these two organisms rely on different pathways for NADPH production in burdened production strains: the acetate formation pathway is favored in S. cerevisiae, while the oxidative phase of PPP in K. phaffii.

Another important redox cofactor in cells is NAD+/NADH. During oxidative phosphorylation in mitochondria, NADH is required for ATP energy formation through the respiratory chain reaction. ATP is needed for almost all growth-related activities, cellular maintenance, and heterologous protein production. Hence, the burden from heterologous protein production increases the demand for carbon allocated to the TCA cycle and oxidative phosphorylation for ATP production (Heyland et al., 2010). For this reason, ATP is likely to decrease in high producing strains as a response to burdensome protein production, compensated by slower growth and carbon uptake (Tyo et al., 2012; Nie et al., 2014). This further highlights the central role of carbon metabolism on burden.

High NADH demand in production strains requires greater redox balance. Under aerobic conditions, S. cerevisiae and K. phaffii can use glycerol as an electron acceptor to oxidize NADH. This was used as an explanation for glycerol production in K. phaffii to help maintain redox balance and avoid cytosolic NAD+ depletion (Nie et al., 2014). In the same study, the authors also found that higher producing strains reduced by-product formation, such as glycerol production, and concluded this is a strategy for the cell to compensate for a higher demand of redox-cofactors and energy (Nie et al., 2014). In line with this, an increase in glycerol production can also lead to lower biomass yield as carbon is removed from central carbon metabolism, contributing to burden (Krogh et al., 2008). Of note for K. phaffii, problems associated with NADH limitations will likely be worse in glucose cultivation as opposed to growth on methanol in co-fed cultures because of the energetic regulation of enzymes involved in methanol oxidation and assimilation pathway (Jordà et al., 2012). While redox requirements for translation are higher in production strains, post-translational processes also require redox factors and other resources. The impact on burden from post-translational processes is discussed next.

Secretion of post-translationally modified proteins by yeast cell factories is advantageous as it facilitates simple downstream purification and processing steps (Nielsen, 2013). The downside is that it increases demand for heterologous protein translocation into the endoplasmic reticulum (ER), folding in the ER, and export via the Golgi apparatus. Together this can introduce bottlenecks that place additive burden on yeast cell factories and trigger cellular stress responses (Jordà et al., 2012; Tyo et al., 2012; Huang et al., 2017; de Ruijter et al., 2018). Indeed, the impact of post-translational processes on the burden state is thought to be a major limitation in yeast, especially considering relatively lower expression levels for these organisms compared to bacteria (Mattanovich et al., 2004).

Proteins are modified, folded, and transported in the ER. Yeast cells must compensate for the burden conferred by the energy-consuming secretory pathway. This is achieved by allocating more resources to cellular maintenance for energy production and away from growth (Jordà et al., 2012; Chen et al., 2020). Even when directing expressed proteins to mitochondria, in contrast to cellular secretion, resources for transportation are a limiting factor causing plasmid expression repression and growth defects (Kintaka et al., 2016; Eguchi et al., 2018). Similarly, anchoring heterologous β-glucosidase to the yeast cell membrane induces burden (Ding et al., 2018). Thus, in order to fully understand burden in yeast, we must better elucidate the burdensome steps during protein transportation and secretion.