Introduction

Expression levels of proteins are optimized to maximize the fitness of organisms, and the evolutionary principle that determines those levels is considered to be demand and constraint (Bruggeman et al., 2020) (Figure 1A). Demand is the evolutionary pressure that works to increase the expression of a protein to meet the requirements of its function, and constraint is the evolutionary pressure that works to decrease the expression of a protein to avoid the negative effects produced by excess protein. Constraints appear as disturbances in cellular function when the protein is overexpressed (Figure 1A, the red arrow). The mechanisms of constraining expression levels, or defects due to overexpression, can be roughly classified into four major categories (Moriya, 2015) (Figure 1B). “Resource overload” (also called “burden”) (Kastberg et al., 2022) is a disorder in which intracellular resources used for protein processing, such as synthesis, localization, degradation, and modification of proteins, are depleted or monopolized by the processing of that protein, resulting in a disruption of processing of other proteins (Dong et al., 1995; Eguchi et al., 2018; Kintaka et al., 2020, 2016); (Stoebel et al., 2008). This phenomenon occurs in proteins with high expression levels. “Stoichiometry imbalance” is a disorder triggered by an imbalance of amounts between constituents within complexes, leading to untimely activation or inactivation (pathway modulation), degradation (resource overload), aggregation (promiscuous interactions), etc. (Kaizu et al., 2010; Makanae et al., 2013; Papp et al., 2003). This occurs due to the overexpression of proteins that form complexes. “Pathway modulation” is a disorder in which pathways are untimely activated or inactivated. This occurs especially with overexpression of proteins involved in pathway regulation, such as metabolic enzymes, kinases and phosphatases, and transcription factors (Prelich, 2012; Youn et al., 2017). “Promiscuous interaction” is a disorder in which weak interactions are enhanced by mass action to form complexes or aggregates that are not normally formed. This damages cellular function by inactivating essential proteins or overloading degradation resources, and is caused by overexpression of intrinsically disordered proteins or liquid-liquid phase-separating proteins (Bolognesi et al., 2016; Ma et al., 2010; Vavouri et al., 2009).

Figure 1.

As the expression level of a protein increases, it eventually encounters one of the barriers created by either these specified constraints or others that are unknown, all of which are related to the function and physicochemical properties of the protein, and damage cellular function. The expression level at which this occurs is referred to as the “expression limit” in this paper (see Figure 1A). If the constraints could be removed from the protein, the expression limit would increase. For example, if components of a complex are simultaneously expressed to resolve a stoichiometry imbalance, the expression limits of those proteins will be increased (Kaizu et al., 2010; Makanae et al., 2013). In this manner, when (one) constraint is removed from a protein, the expression limit would increase until the next constraint barrier is hit (Figure 1A, dotted lines). The amount of expression limit created by a constraint barrier depends on the type of constraint. Stoichiometry imbalances, pathway modulations, and promiscuous interactions would create different expression limits for each protein. This is because they are determined by the activity, regulatory mechanism, and physicochemical properties of each protein. On the other hand, resource overload is presumed to be determined to some extent by the capacities of the processes in the cell (Kintaka et al., 2020) (Figure 1C). Essentially, the synthesis, in which all proteins are processed, should have the largest capacity, while processes such as transport, folding, and degradation, in which only a fraction of proteins are processed, should have smaller capacities. In fact, adding transport or degradation signals to a cytosolic protein lowers the expression limit (Kintaka et al., 2016). In that light, the final constraint barrier that emerges when all possible constraints are removed is the overloading of the synthesis process.

Overloading of the synthesis process is specifically referred to as “protein burden” or “protein cost” (Eguchi et al., 2018; Kafri et al., 2016). It has long been the subject of study, mainly as an effect on growth produced by excess (gratuitous) proteins that are not necessary for cellular function (Bruggeman et al., 2020; Scott et al., 2010). This idea was first proposed in studies of the cost of unnecessary induction of the lac operon in Escherichia coli (Koch, 1983). In recent years, much attention has been paid to the actual mechanisms by which massive expression of excess protein inhibits growth and the response of cells (Farkas et al., 2018; Kafri et al., 2016; Metzl-Raz et al., 2020). During protein burden, it is presumed that depletion of the materials, machinery, and factors necessary for transcription and translation occurs, along with competition from these monopolies. In particular, amino acids and aminoacyl-tRNAs are major candidates for depletion (Kastberg et al., 2022). In addition, ribosomes, which are crucial for protein synthesis, themselves require significant resources for their synthesis and have therefore been considered major candidates for depletion during protein burden, as they compete for these limited resources (Metzl-Raz et al., 2017). Mechanistic analysis of protein burden in budding yeast, a model eukaryotic cell, has been intensively conducted by Barkai and colleagues, who reported that transcription is rate-limiting in phosphate starvation and translation in nitrogen source starvation (Kafri et al., 2016), and that basic transcription factors are limiting resources (Metzl-Raz et al., 2020). In addition, a cellular response in the form of increased cell size and enhanced protein synthesis capacity has been observed (Kafri et al., 2016). Thus, the previously expected decrease in ribosomes has not been detected.

As mentioned above, to investigate protein burden, we need to overexpress proteins with minimal constraints. This is because overexpressing proteins with additional constraints cannot overload the synthetic process, as the additional constraints reduce the expression limits (Figure 1A). Also, the cellular damage and cellular response generated by the constraint mask the protein burden response. However, it remains unclear which proteins are not subject to constraints other than protein burden. Proteins that would induce protein burden have been determined by assuming that they would be minimally constrained based on their properties, and this assumption has been confirmed experimentally. Model proteins selected in this manner in yeast include fluorescent proteins (FPs) and certain glycolytic enzymes (Eguchi et al., 2018; Farkas et al., 2018; Kafri et al., 2016; Kintaka et al., 2020, 2016). FPs do not have a specific function or interaction partner in yeast cells and are not localized. On the other hand, glycolytic enzymes are naturally expressed at very high levels and are thus considered to be less constrained. Notably, these proteins exhibit an expression limit of approximately 15% of total protein in yeast cells (Eguchi et al., 2018; Kintaka et al., 2016). However, it is not clear whether these proteins are free from constraints other than protein burden. This is because there may be unknown constraints, and there is no evidence that the measured expression limit is the “true limit” at which protein burden occurs.

In addition to the protein (amino acid sequence) itself, there is another aspect to be considered when selecting a model for protein burden. That is the nucleotide sequence of the gene used for expression, i.e., transcriptional, and translational regulatory sequences, codon usage, and so on (Figure 1D). For example, if a protein is to be expressed to the limit by a less codon-optimized and thus less translationally efficient mRNA, a large amount of mRNA would be required, which would induce a greater burden on transcription. Conversely, a similar attempt by an mRNA with high translation efficiency would induce a greater burden on translation. These are the constraints inherent in protein synthesis itself. Although there is some evidence that protein synthesis in nature is economically designed to optimally load transcription and translation (Frumkin et al., 2017; Hausser et al., 2019; Mahima and Sharma, 2023), it still needs to be determined what the genetic design should be to minimize the burden.

Considering the above circumstance, a realistic approach to protein burden requires more than just drawing straightforward conclusions from the overexpression of a single model protein (or gene). Rather, it requires experimentally identifying proteins with the highest possible expression limits, investigating the negative impacts of their overexpression, and, if any constraints are revealed, examining whether their removal can increase expression limits. Through successive iterations, this process gradually deepens our understanding of protein burden (Figure 1E). In each iteration, the least constrained protein (and its gene) will be generated and utilized as a benchmark for further investigation into proteins with high (or low) expression limits and the adverse effects of their overexpression. This will (unintentionally) elucidate a new mechanism of constraint. Indeed, the enhanced green fluorescent protein (EGFP), which we have used as a model for protein burden (Kintaka et al., 2020, 2016; Makanae et al., 2013), was found to form a chaperone-entrapped aggregate upon overexpression and trigger the heat shock response due to its low folding ability (Namba et al., 2022). Its overexpression also causes abnormal yeast cell morphology through proteasomal stress as it contains cysteines. The moxGFP (mox), which improved folding properties and lacks cysteine residues (Costantini et al., 2015), showed a higher expression limit and no longer induced abnormal morphologies. However, even when overexpressing mox, the anticipated depletion and competition for factors, or cellular responses to them have not been observed (Namba et al., 2022).

In this study, we found that a mutation causing loss of fluorescence in mox, and other fluorescent proteins increased the expression limit up to 3-fold, to more than 40% of total protein. Introducing a neutrality index (NI) to assess protein constraint, we found that non-fluorescent mox (mox-YG) and enzymatically inactive Gpm1 (Gpm1-CCmut) are currently the least constrained. Upon overexpression of mox-YG, depletion of amino acid or nitrogen sources, decreased ribosomal expression, and a metabolic shift from glycolysis to oxygen respiration were observed. Thus, physiological states, as anticipated with such protein burdens, are only generated by massive overexpression of proteins with significantly lower constraints. Some of the responses found appear to be due to inactivation of the TORC1 pathway, suggesting that eukaryotic cells use this regulatory pathway to buffer protein burden. In the burdened cells, the nucleolus was dysplastic, and mutations in the nuclear exosome partially restored this. Thus, the abnormal formation of liquid-liquid phase-separated organelles due to dilution may be one of the defects caused by the strong protein burden.

Results

Fluorescent property is a constraint on the expression of fluorescent proteins

To evaluate the expression limits of FPs, we overexpressed FPs to levels that induced growth inhibition. Gene constructs, in which the TDH3 promoter controls the target protein genes, were incorporated into plasmids designed for the genetic-tug-of-war (gTOW) method (Figures 2A, 2_S1A) (Moriya et al., 2006). Subsequently, yeast cells were transformed and cultured in a medium depleted of leucine. The maximum growth rate (MGR) of cells was calculated from serial measurements of optical density at 660 nm (OD₆₆₀). Protein levels were determined through the analysis of proteins from cells in the logarithmic growth phase (Figure 2_S1B). For each protein, we also analyzed a mutant in which the Tyr residue forming the fluorophore was replaced with Gly (YG mutation), resulting in a loss of fluorescence. This is because we unexpectedly found that this mutation substantially increased the expression limits of FPs (see below).

Figure 2.

Overexpression of the FPs and mutants all caused growth inhibition (Figure 2B, C). EGFP and mox were expressed at about 15% of the total protein as previously reported (Kintaka et al., 2016), but mCherry (mChe) and YG mutants were expressed at higher levels. Mox-YG expressed 44% of the total protein (Figures 2D, E). Our previous work had evaluated only expression limits (Eguchi et al., 2018), but we realized that we could not properly assess the constraints on each protein if the growth rates at the expression limit were different. For example, the mChe’s expression limit (20%) is higher than that of sfGFP (17%), but the relative MGRs are 20% and 80%, respectively, which is much lower for mChe. The relationship between %MGR and %protein level when the vector is set at 100% is shown in the scatter plot in Figure 2F. The lower left indicates more constraint as lower expression levels lead to growth inhibition, and the upper right indicates less constraint as higher expression levels do not cause growth inhibition. Here, we describe “unconstrained” as “neutral” and calculate the neutrality index (NI) as the product of %MGR and %protein level.

Note that the regime of NI in Figure 2F shows a theoretical neutrality distribution but does not show the relationship between growth rate and the expression level when the expression level of a protein with a particular neutrality is increased or decreased. For example, 20% of the growth rate would be maintained when the expression of mox-YG (NI = 2,400) is increased to 100%, but this is not realistic.

The NI of each protein indicates that introducing a non-fluorescent mutation in FPs uniformly increases neutrality (Figure 2G), suggesting that “having fluorescence” is a constraint on the expression of FPs. The most neutral proteins were sfGFP-YG and mox-YG, with NIs of about 2,400. Notably, despite significant differences in the degree of growth inhibition and expression limits between the two proteins, their NIs are nearly identical (Figures 2F, G). They differ only in two amino acid residues (Cys - Ser) (Figure 2_S3), which seem to generate differences in expression efficiency. However, this does not appear to affect their NIs.

We also investigated the impact of expression efficiency on NI with another FP, mChe. The mChe used in this study is codon optimized (the codon adaptation index (CAI) = 0.584) (Keppler-Ross et al., 2008), while the mChe used by Kafri et al. for protein cost analysis (here referred to as mChe-Kafri) is not codon optimized (CAI = 0.236) (Kafri et al., 2016). When overexpressed, mChe-Kafri (and mChe-Kafri-YG) had lower expression and growth inhibition than mChe (and mChe-YG) (Figures 2_S4A-D). This may be due to the lower translation rate of mChe-Kafri mRNA. However, their NIs were not significantly different, especially for the YG mutation (Figures 2_S4E, F). Thus, the constraints indicated by NI are apparently not affected by translation efficiency. In the study using this mCherry-Kafri, it was also reported that the more mChe-Kafri was expressed, the more the cells enlarged (Kafri et al., 2016). However, no cell enlargement was observed with mox-YG, which can be expressed more than mCherry. Interestingly, the mChe-YG cells did not show enlargement (Figures 2_S5A, B), indicating that cell enlargement is a phenomenon that occurs specifically when fluorescent mChe is overexpressed.

It should be noted that the growth inhibition caused by the overexpression of mox-YG in this study is a burden on translation rather than on transcription. This is because overexpression of a frameshift mutant, in which four nucleotides were inserted directly after the ATG of the mox-YG gene, abolished the notable growth delay (Figures 2_S5C, D).

Constraint on fluorescent protein expression arises from fluorophore properties

Next, we further analyzed the possibility that fluorescence constrains the expression of FPs. When yeast was grown in the dark, the degree of growth inhibition of mox was the same (Figure 2_S6A) and the expression limit of mox was rather increased (Figure 2_S6B). Phototoxicity does indeed exist when expressing mox with fluorescence. This is because, when exposed to strong light, cells overexpressing mox show a reduced growth rate (Figure 2_S6C-E) and abnormalities in their internal structures (Figure 2_S6F). However, the expression limits of mox and mox-YG were almost unchanged by the light (Figure S2_S6G, H). Thus, the constraint was not due to phototoxicity itself.

The number of Tyr residues was also irrelevant because NI was like that of mox-YG when a single Tyr was added to the C-terminus of mox-YG (Figure 2_S7C-E). Reactive oxygen species (H₂O₂) are generated during fluorophore formation. When H₂O₂ was added to the medium, stronger growth inhibition occurred only in mox (Figure 2_S7F-H). This suggests that the reactive oxygen species generated during fluorophore formation are responsible for the constraint. However, this possibility is not supported by the omics analysis described later (see Discussion). Mox-YG, like mox, was present throughout the cytoplasm and did not cause any characteristic localization or aggregation (Figure 2_S8).

During this study, we examined a moxGFP (mox-T203I) in which Thr203 was accidentally replaced by Ile (Figure 2_S9). The fluorescence of this mutant is reduced to 78% of mox (Figures 2_S9A-C). The growth rate of the cells expressing this mox-T203I was like that of mox and mox-YG (Figure 2_S9E). Still, the expression limit was intermediate between mox and mox-YG (Figure 2_S9F), resulting in the NI intermediate between mox and mox-YG (Figure 2_S9G). This result further supports that fluorescence is related to constraints. However, as mentioned above, fluorescence emission itself should not be toxic (Figure 2_S6). We next focused on the formation process of the fluorophore: the tyrosine of the GFP fluorophore undergoes phenol and phenolate state changes, with Thr203 stabilizing the phenol form and substitution to Ile reducing the phenolate form (Kummer et al., 2000). Furthermore, in enhanced GFPs such as mox, substitution of Ser65 before Tyr66 for Thr anchors the phenolate form. To investigate the possibility that this is a factor determining the limit of mox, we produced mox with Thr65 replaced with Ser and examined its expression limit. The mox-T65S mutant was expressed 12% more than mox (Figures 2_S9H, I), suggesting that the phenol-type fluorophore can be a limiting factor for the expression level of the fluorescent protein, while the concrete mechanism is still unclear.

The activities of the glycolytic enzymes Gpm1 and Tdh3 may constrain their expression

We previously investigated the expression limits of a group of glycolytic enzymes in S. cerevisiae, with Gpm1 and its catalytic center (CC-) mutant having the highest expression limits (Eguchi et al., 2018). Here, we evaluated the neutrality of the wild types and their CC-mutants of Tdh3 and Gpm1 (Figure 3). The growth rates at the expression limits were higher in the CC-mutant for both Tdh3 and Gpm1, and in Gpm1, the CC-mutant grew two-fold faster (Figures 3A, B, 3_S1). The protein levels of wild types and mutants of both proteins were not significantly different (Figures 3C, D). The calculated NIs were highest for Gpm1 CC-mutant, which was equivalent to that of mox-YG (Figures 3E, F). Thus, mox-YG and Gpm1 CC-mutant are the least constrained proteins to this point. We previously concluded that the activities of glycolytic enzymes generally do not affect their expression limits (Eguchi 2018), but their neutralities suggest that their activities can become constraints in some way. This may be due to a disturbance in metabolism or a property of the active enzyme, such as an effect of reactive amino acids (His or Cys).

Figure 3.

Overexpression of mox-YG causes an amino acid starvation response

To understand the cell physiology under protein burden, we next performed transcriptome (RNA-seq) analysis of the cells overexpressing mox-YG, the least constrained protein. Although the expression limit of mox-YG is about 3-fold higher than that of mox (Figure 2E), the growth rates at the expression limits of both are not largely different (Figure 2C). Thus, by comparing transcriptomes upon their overexpression, we may be able to distinguish and extract the general transcriptional response due to reduced growth rates from the (protein burden) response associated with the massive expression of a less constrained protein.

The expression changes under mox and mox-YG overexpression were calculated as fold change (FC) over the vector control (Figures 4_S1A, B). The overall trend of expression changes was similar (r = 0.64, Figure 4A). This similar variation may include responses associated with the expression system and reduced growth rate. On the other hand, there were also different transcriptional responses between the two. Gene categories that showed significant expression changes in either mox or mox-YG overexpression are shown in Figure 4B. In addition to the enhanced ribosome synthesis and decreased glycolysis observed in previous mox overexpression (Namba et al., 2022), enhanced oxidative phosphorylation was detected. Interestingly, “Ribosome biogenesis”, which was elevated in mox overexpression, showed a rather decreasing trend in mox-YG overexpression (Figure 4B). Indeed, IFH1, which is involved in ribosomal gene transcription (Martin et al., 2004), was upregulated in mox, but downregulated in mox-YG (Figure 4A). Different trends were also observed for amino acid metabolism (Figure 4B). The amino acid transporter genes AGP1 and GAP1 were upregulated only in mox-YG overexpression (Figure 4A). Expressions of specific amino acid and nitrogen source transporter genes were strongly induced in mox-YG (Figure 4C). The expression of glucose transporter genes was also very different (Figure 4C). The transcriptional responses in mox-YG overexpression obtained by RNA-seq were confirmed by transcriptional reporter analysis (Figure 4D, 4_S2). These results suggest that mox and mox-YG induce similar but distinct transcriptional responses, particularly that mox-YG overexpression may result in depletion of amino acid and nitrogen sources and decreased ribosome synthesis.

Figure 4.

Overexpression of mox-YG causes a proteomic response that partially overlaps with TORC1 inactivation

Protein burden is an overload on synthetic resources, and thus, when this overload occurs strongly, it is expected that there would be a reduction in the amount of proteins other than those being overexpressed due to competition for synthetic resources. Indeed, in cells overexpressing mox-YG, the amounts of proteins other than mox-YG were reduced (Figure 5A). Interestingly, the total amount of protein per cell remained almost constant, with the amount of other proteins decreasing as mox-YG was overexpressed. This trend was generally true for the overexpression of other FPs and glycolytic enzymes (Figure 5_S1). This suggests that there is a mechanism (or constraint) that maintains the overall cellular protein levels. To examine changes in protein expression, proteomic analysis of mox-YG overexpressing cells was performed. The distribution of intensities obtained from the analysis is shown in the Proteomaps in Figure 5B. In mox-YG overexpressing cells, mox-YG accounted for 40% of the total protein expression. Figure 5A suggests that in mox-YG overexpressing cells, proteins other than mox-YG are overall reduced to about 60% compared to the vector. Therefore, we considered it necessary to normalize the data overall to accurately represent changes in the per-cell proteome. However, in this study, we used data without such correction to focus on how the composition of the proteome, other than mox-YG, changed (see Materials and Methods and Figures 5_S2 for more details).

Figure 5.

Figure 5C illustrates the categories that showed significant changes upon mox-YG overexpression (all categories are shown in Figure 5_S3). Elevated mitochondrial functions, such as oxidative phosphorylation and TCA cycle, and decreased ribosomal protein and ribosomal synthesis were observed. As mentioned above, RNA-seq results suggested nitrogen source starvation in mox-YG overexpression. Therefore, we investigated whether this response is influenced by the regulation through the TORC1 pathway, a primary mechanism responsive to nitrogen sources. By comparing expression changes due to rapamycin (a TORC1 inhibitor) treatment (Gowans et al., 2018) and mox-YG overexpression, we found that about 40% of the proteins with altered expression were congruent with the group of proteins regulated by TORC1 (Figures 5D, E). Thus, the proteome expression changes due to mox-YG overexpression may include responses due to inactivation of the TORC1 pathway as well as other responses. Figure 5C shows which categories had significant expression changes when separated into rapamycin-responsive (Rap) and rapamycin-non-responsive (Non-Rap) gene populations. Carbohydrate metabolism, elevated TCA cycle, and decreased ribosome synthesis were suggested to be responses due to TORC1 inactivation, while endocytosis, cytoskeleton, protein transport and lipid synthesis, and glycolysis were suggested to be due to responses other than TORC1 inactivation (Figure 5_S4). Oxidative phosphorylation included both Rap responses (12/32) and non-responses (20/32) (Figure 5_S4C). We confirmed an increase in Rip1, one of the proteins in the oxidative phosphorylation category, using fluorescence microscopy observations (Figure 5_S5). Thus, the massive expression of mox-YG resulted in a decrease in intracellular proteins and changes in proteome composition. The changes included decreased ribosomes and an increase in mitochondrial proteins the primary cause of which seemed to be the inactivation of the TORC1 pathway due to nitrogen source depletion.

Overexpression of mox-YG causes a dysplastic nucleolus, which is alleviated by mutations in the nuclear exosome

Ribosomes are a crucial factor for protein burden because they are themselves protein synthesizers while being synthesized in the highest amounts in the cell. Therefore, we further focused on the behavior of ribosomes in mox-YG cells and expression changes were investigated in detail for each ribosomal protein (Figure 6_S1). While the majority of the 84 ribosomal proteins detected (45 of them significantly) decreased, six (Rps12, Rps31, Ppp0, Rpp2a, Rpl5, and Rpl12) showed significant increases. Interestingly, all decreased ribosomal proteins are assembled into ribosomes in the nucleolus, while all increased ribosomal proteins are assembled into ribosomes in the cytoplasm (Lafontaine, 2015). They constitute the “beak” of the 40S subunit and the “P-stalk” and the “central protuberance (pc)” of the 60S subunit (Figure 6A). Electron micrographs showed an abnormal nucleolar formation of mox-YG overexpressing cells (Figures 6B, 6_S3), in addition to reduced ribosome density in the cytoplasm (Figure 6_S2A). This abnormal nucleolar formation was also confirmed by fluorescence microscopy images of Nsr1 (Nucleolin), the major protein of the nucleolus (Figure 6_S2B-D).

Figure 6.

Based on these results, we hypothesized that overexpression of mox-YG creates an imbalance between the amount of ribosomal protein synthesized in the cytoplasm and the amount of rRNA synthesized in the nucleolus (Figure 6C, D). The imbalance may lead to the creation of misassembled ribosomes, which are eventually degraded. We speculated that the primary enzyme degrading the misassembled ribosome should be the nuclear exosome. This is because our previous genome-wide mutation analysis identified exosome mutants (dis3-1, mtr3-ts, mtr4-1, rpn42-ph, rpn45-ph, and ski6-ph) as primary mitigators of protein burden (Kintaka et al., 2020). We confirmed that the mtr4-1 mutant does not show growth retardation upon mox-YG overexpression (Figure 6E). In fact, in mtr4-1 cells, there was no difference in growth rate between the vector and mox-YG overexpression (Figure 6_S5B), but rather a shorter lag time for the mox-YG overexpression than the vector control (Figure 6E). We next quantified protein upon mox-YG overexpression in mtr4-1 cells (Figure 6F). Compared to the wild-type (Vector), the total protein content in mtr4-1 (Vector) cells was reduced to 78%, which was likely related to the decreased growth. In the mtr4-1 mox-YG overexpressing cells, the amount of other proteins decreased (to 54%), overwhelmed by the 40% expression of mox-YG. However, the overall protein content increased compared to the mtr4-1 vector control, with a composition like the wild-type mox-YG overexpressing cells. Therefore, in mtr4-1 cells, even though mox-YG overexpression reduced the amount of other proteins (to 54%) compared to the vector (78%), the growth rate was not reduced (Figure 6-S5B). This suggests that mox-YG overexpression restores the effect of growth inhibition caused by mtr4-1. Next, we observed the composition of the nucleolus in mtr4-1 (Figures 6G, 6_S6B, E). The failure of nucleolus formation observed upon mox-YG overexpression was partially rescued in the mtr4-1 strain. This supports our hypothesis that misassembled ribosomes are degraded by the exosome (Figure 6D), and with low exosome activity, nucleolus formation (and efficient ribosome assembly) would be possible even with overexpression of mox-YG (Figure 6_S4D).

While experimenting with mox-YG overexpressing cells, we found that these cells stop growth earlier (lower final density) at higher temperatures (38°C) than the vectors (Figure 6_S6A). Cells at this time did not show cell death such as cell bursting (Figure 6_S6). Recently, it was reported that intracellular osmotic pressure increases at high temperatures, resulting in less formation of liquid-liquid phase-separated structures such as nucleoli, which are recovered when cells are placed in hyperosmotic conditions (Watson et al., 2023). We hypothesized that the growth inhibition of mox-YG overexpressing cells at high temperatures may be due to the enhanced failure of nucleolus formation, and thus cultured mox-YG overexpressing cells under hyperosmotic pressure. At 30°C, hyperosmolarity caused the same decrease in growth for the vector control and mox-YG overexpression (Figures 6I, 6_S6C). On the other hand, at 38°C, hyperosmolarity caused a decrease in growth only in the vector control and no further decrease in mox-YG overexpression cells (Figures 6J, 6_S6D). The same trend was observed regarding the size of the nucleosome (Figures 6_S6E, F). Under the conditions with added sorbitol and under mox-YG overexpression, the size of the nucleoli was smaller than the control at 30°C, whereas, at 38°C, it was rather larger than the control. This suggests that mox-YG overexpression and hyperosmolarity may mitigate each other’s growth-inhibitory effects, and one explanation may be that hyperosmolarity suppresses the abnormal organization of the nucleolus caused by mox-YG overexpression.

Discussion

In this study, we found that the fluorescent properties of FPs constrain their expression limits. This fluorescence-induced cytotoxicity appears as different phenotypes depending on the FPs; an excess of fluorescent mCherry causes cell enlargement, but such a phenotype was not seen with green fluorescent proteins. On the other hand, the cause of the cytotoxicity is not clear currently. Fluorescence itself does not seem to constrain the expression limit, because the mox expression limit remained the same when cultured in the dark or under strong light (Figure 2_S6). We also believe that H₂O₂ generated during fluorophore maturation (Craggs, 2009) is unlikely. In the presence of H₂O₂, mox overexpressing cells grew slightly slower than mox-YG overexpressing cells (Figure 2_S6B). However, no transcriptional response to oxidative stress, including the catalase gene (CTT1), was seen with mox overexpression (FDR <0.187, over the vector, Data S1). Because CTT1 expression was elevated in mox-YG overexpression (FDR <0.006, over the vector, Data S1), presumably due to enhanced respiration, mox-YG overexpression might rather increase oxidative stress tolerance. Conversely, “inactive” proteins might accumulate more without interfering with intracellular functions because both inactive fluorescent proteins and inactive glycolytic enzymes showed higher neutrality (Figures 2 and 3). However, our current data did not clarify the mechanism, and investigating more proteins and their variants would be required.

The currently least cytotoxic proteins, non-fluorescent mox-YG and inactive Gpm1 express more than 40% of the total protein, whereas yeast cells can maintain growth under these conditions (Figures 2 and 3). This is higher than the prediction in E. coli, which stops growing when 37% of the excess protein is expressed (Bruggeman et al., 2020). This may be due to the very low cytotoxicity of the proteins expressed in this study, or it may be because eukaryotic cells have greater capacity than prokaryotic cells. In the mox-YG overexpressing cells, we observed a previously expected but unobserved cellular physiology of the protein burden: depletion of protein synthesis resources (i.e., nitrogen source; Figure 4) and a decrease in other proteins (glycolytic enzymes and ribosomes; Figure 5), possibly due to competition for synthesis. In addition, metabolic conversion to mitochondria and failure of nucleolus formation were observed (Figures 5 and 6). As a matter of fact, the nitrogen source depletion response (Figure 4) itself is not self-evident. Because the total protein content in the mox-YG overexpressing cells is the same as that in the vector control (Figure 5A), the intracellular nitrogen source used for protein synthesis should not be depleted. Then, why does the nitrogen source depletion response occur? Excess mox-YG may promote wasteful consumption of amino acids and nitrogen, or there may be pathways that sense the amount of “useful proteins” and their ability to be synthesized, or there may be depletion of specific amino acids in mox-YG excess. Another possibility is that there may be some mechanism that maintains the amount of total protein in the cell even when the nitrogen source is depleted. Depletion of nitrogen sources probably results in the inactivation of the TORC1 pathway (Figure 5C-E), metabolic conversion from fermentation to respiration, and ribosome depression to an economic mode. In E. coli, the ppGpp pathway plays this role (Shachrai et al., 2010). While the above is thought to be a physiological response to protein burden, a defect in ribosome synthesis in the nucleolus was also suggested (Figure 6B). This could be due to a conflict between nuclear rRNA synthesis and cytoplasmic protein synthesis or to the negative effects of the inability to create a liquid-liquid phase-separated structure due to dilution effects (see below). The partial restoration of nucleolus formation by a mutation of the exosome (Figure 6G) supports this possibility.

When one type of protein is expressed as much as 40% of the total, as in the present study, the cytoplasmic crowding, viscosity, pH, etc. may change due to the nature of the protein and the effect of the reduced protein by competition. These may alter the movement of substances. It is known that when ribosomes are reduced, the fluidity of large proteins changes (Delarue et al., 2018). Water activity-osmotic pressure may change (Watson et al., 2023), and pH may also be affected. These may affect the formation of liquid-liquid phase-separated structures (Delarue et al., 2018; Watson et al., 2023). The failure of nucleolus formation observed in mox-YG overexpression (Figure 6B) may be due to such a condition change. In fact, the growth of mox-YG overexpressing cells is significantly impaired at high temperatures (Figure 6H), where liquid-liquid phase separation is less likely to occur (Watson et al., 2023), while the effect appears to be reduced at high osmolarity (Figure 6J), where it is more likely to occur (Watson et al., 2023). A mutation of the nuclear exosome restored nucleolus formation (Figure 6G), which might be due to the reformation of the liquid-liquid phase-separated structure (i.e., nucleolus) by increasing the concentration of rRNA. These have not been considered phenomena that occur with protein overexpression. The engineering of physicochemical properties of the cytoplasm thus may be possible by mass expression of proteins with specific properties.

In this study, in addition to mox-YG, inactive Gpm1 was also expressed in large amounts as well (Figure 3). Both are globular proteins with relatively small molecular weight, but it is not yet clear what characteristics are necessary for them not to be cytotoxic. Further investigation of the neutrality of various proteins will clarify these features. In this study, expression limits were investigated, but no experiments were performed to gradually increase expression levels. For example, it would be meaningful to investigate at what expression level of mox-YG a shift to the economy mode or failure of nucleolus formation occurs. Also, due to the limitations of the experimental system, it may not be possible to express mox-YG to its true limit. The expression of one specific protein cannot be 100% of the total protein, and there is always a limit. As the number of ribosomes continues to decrease due to competition for synthesis, the excess proteins cannot be synthesized either. Thus, there must be a final equilibrium point. A more powerful and controllable expression system would be needed to investigate this. Although the translation system appears to be the primary target of burden in mox-YG overexpression (Figure 2_S5C, D), the bottleneck in the expression system may also differ depending on gene design (Hausser et al., 2019).

Materials and Methods

Strains, growth conditions, and yeast transformation

BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) (Brachmann et al., 1998) was used as the host strain for the experiments. Yeast culture and transformation were performed as previously described (Amberg et al., 2005). A synthetic complete (SC) medium without uracil (Ura) or leucine (Leu), as indicated, was used for yeast culture. The strains used in the reporter assay are listed in the Key Resource Table.

Plasmid construction

The plasmids were constructed by homologous recombination activity of yeast cells (Oldenburg 1997), and their sequences were verified by DNA sequencing. Plasmids used in this study are listed in the Key Resource Table.

Genetic tug-of-war

To overexpress a target protein to a level that causes growth inhibition (in this study this is referred to as “expression limit”), we used gTOW (Figure 2_S1A) (Moriya et al., 2012, 2006). The gene of the target protein is incorporated into the gTOW plasmid (here, pTOW40836) and introduced into yeast cells lacking ura3 and leu2 genes. The selection marker in this case is uracil (–Ura conditions). Since this plasmid has the replication origin of 2 µm plasmid (2 µ ORI), this plasmid becomes multicopy in the cell (about 30 in the case of the vector). When yeast cells carrying the plasmid are transferred to a medium without leucine and uracil (–Leu/Ura conditions), the copy number of the plasmid in the cells increases (to about 120 in the case of the vector). This is because leu2-89 on the plasmid is a LEU2 allele with a large deletion in its promoter, so cells with higher plasmid copy numbers will grow faster in –Leu/Ura conditions. In other words, leu2-89 acts as a bias to raise the plasmid copy number in – Leu/Ura conditions. If overexpression of the target protein inhibits growth, the plasmid copy number will be lower than the copy number that gives rise to the critical expression of that protein. In other words, the target gene acts as a bias to lower the plasmid copy number. The resulting tug-of-war between the selection biases created by the two genes leads to a target protein being expressed at close to the critical expression level in –Leu/Ura conditions (the green-colored area in Figure 2_S1A).

Measuring growth rate

Cellular growth was measured by monitoring OD₆₆₀, every 10 min using a compact rocking incubator (TVS062CA, ADVANTEC). The max growth rate (MGR) was calculated as described previously (Moriya et al. 2006). Average values, SD, and P-values of Welch’s t-test were calculated from at least three biological replicates. The growth data of Figures 2_S9A, 2_S9B, and 4_S2 was measured by monitoring OD₅₉₅, respectively, every 10 min using a microplate reader (Infinite F200, TECAN).

Protein analysis

BY4741 cells overexpressing a target protein were cultivated in SC–Leu/Ura medium. The total protein was extracted from log-phase cells (OD₆₆₀ = 1.0) with a NuPAGE LDS sample buffer (Thermo Fisher Scientific) after 0.2 mol/l NaOH treatment (Kushnirov 2000). For each analysis, the total protein extracted from 1 optical density (OD) unit of cells with OD₆₆₀ was used. For total protein visualization, the extracted total protein was labeled with Ezlabel FluoroNeo (ATTO), as described in the manufacturer’s protocol, and separated by 4–12% sodium-dodecyl sulfate acrylamide gel electrophoresis (SDS-PAGE). Proteins were detected and measured using the LAS-4000 image analyzer (GE Healthcare) in SYBR–green fluorescence detection mode and Image Quant TL software (GE Healthcare). Protein quantification was performed as shown in Figure 2_S1B. Average values, SD, and P-values of Welch’s t-test were calculated from three biological replicates.

Measuring GFP fluorescence

GFP fluorescence (Ex485nm/Em535nm) was measured every 10 min using a microplate reader (Infinite F200, TECAN). Average values, SD, and p-values of Welch’s t-test were calculated from eight biological replicates.

RNA-seq analysis

BY4741 cells overexpressing a target protein were cultured in SC–Leu/Ura medium and harvested at the logarithmic growth phase. RNA extraction was performed according to (Köhrer and Domdey, 1991). The purified RNA was quality checked by BioAnalyzer (Agilent) or MultiNA (Shimazu), and the concentration was measured by Qubit (Thermo Fisher Scientific). Purified RNA was stored at −80°C until subsequent experiments. cDNA libraries were prepared using the TruSeq Stranded Total RNA kit (Illumina), using half the protocol of the TruSeq RNA library prep kit. Three biological replicates were analyzed for all strains. The sequences were checked for read quality by FastP (Chen et al., 2018) and then aligned using Hisat2 (Kim et al., 2019). The aligned data were formatted into bam files by Samtools (Li et al., 2009) and quantified by StringTie (Pertea et al., 2015). Finally, expression level variation analysis was performed by EdgeR (Robinson et al., 2010). The raw data were deposited into DDBJ (accession number: PRJDB18064). The processed data (Data_S1_RNAseq.csv) included the average expression levels (log10CPM) of 6685 genes in the vector control, the changes in expression (log2FC) upon overexpression of mox or mox-YG compared to the vector control, and the significance values of expression change (mox_FDR, mox-YG_FDR). Figures 4_S1A, B is the visualization of this data using volcano plots. This data is used in subsequent data analyses below.

Reporter assay

The strains used for the reporter assay were created by introducing the promoter region of each gene and the sequence of mScarlet-I, a type of RFP, into the FCY1 locus of the genome as shown in Figure 4_S2A. Correct integration was checked by histidine prototrophy and 5-FC resistance of the strain. The cell density (OD₅₉₅) and RFP fluorescence (Ex535nm/Em590nm) of the created strains were measured every 10 min using a microplate reader (Infinite F200PRO, TECAN). Average and SD of max RFP fluorescence, and p-values of Welch’s t-test were calculated from eight biological replicates.

Proteome analysis

The vector control and mox-YG overexpressing cells (three biological replicates) were cultured in 5 ml of SC–LeuUra medium with shaking and cells were collected at OD₆₆₀ = 1. Cells were washed twice with PBS, frozen on dry ice, and transported to Kazusa DNA Research Institute. The samples were prepared as described in the previous study (Hughes et al., 2019; Kawashima et al., 2022). The cells were dissolved in 100 mM Tris-HCl (pH 8.0) containing 4% SDS and 20 mM NaCl using BIORUPTOR BR-II (SONIC BIO Co., Kanagawa, Japan). The 20 μg of extracted proteins were quantified using Pierce™ BCA Protein Assay Kit (Thermo Fisher Scientific, WA, USA) at 500 ng/μL. The protein extracts were reduced with 20 mM tris (2-carboxyethyl) phosphine for 10 min at 80 °C followed by alkylation with 30 mM iodoacetamide for 30 min at room temperature in the dark. Protein purification and digestion were performed using the sample preparation (SP3) method (Kawashima et al., 2022). The tryptic digestion was performed using 500 ng/μL Trypsin/Lys-C Mix (Promega, Madison, WI, USA) for overnight at 37 °C. The digests were purified using GL-Tip SDB (GL Sciences, Tokyo, Japan) according to the manufacturer’s protocol. The peptides were dissolved again in 2% ACN containing 0.1% TFA and quantified using BCA assay at 150 ng/μL. The digested peptides were loaded directly using a 75 μm × 12 cm nanoLC nano-capillary column (Nikkyo Technos Co., Ltd., Tokyo, Japan) at 40°C and then separated with a 30-minute gradient (mobile phase A = 0.1% FA in water, B = 0.1% FA in 80% ACN) consisting of 0 min 8% B, 30 min 70% B at a flow rate of 200 nL/min using an UltiMate 3000 RSLCnano LC system (Thermo Fisher Scientific). The eluted peptides were detected using a quadrupole Orbitrap Exploris 480 hybrid mass spectrometer (Thermo Fisher Scientific) with normal window DIA. The MS1 scan range was set as a full scan with m/z 490-745 at a mass resolution of 15,000 to set an Auto Gain Control (AGC) target for MS1 as 3 × 10⁶ and a maximum injection time of 23 msec. The MS2 were collected at more than m/z 200 at 30,000 resolutions to set an AGC target of 3 × 10⁶, maximum injection time of “auto”, and fixed normalized collision energy of 28%. The isolation width for MS2 was set to 4 m/z, and for the 500-740 m/z window pattern, an optimized window arrangement was used in Scaffold DIA (Proteome Software, Inc., Portland, OR, USA). The raw data were searched against an in silico predicted spectral library using DIA-NN (Demichev et al., 2020). (version:1.8.1, https://github.com/vdemichev/DiaNN) First, an in-silico predicted spectral library was generated from the human protein sequence database (UniProt id UP000005640, reviewed, canonical,

20381 entries) using DIA-NN. The DIA-NN search parameters were as follows: protease, trypsin; missed cleavages, 1; peptide length range, 7-45; precursor charge range, 2-4; precursor mass range, 495-745; fragment ion m/z range, 200-1800; mass accuracy, 10 ppm; static modification, cysteine carbamidomethylation; enabled “Heuristic protein interferences,” “Use isotopologues,” “MBR,” and “No shared spectra.” Additional commands were set as follows: “mass acc cal 10,” “peak translation,” and “matrix spec q”. The protein identification threshold was set at < 1% for both peptide and protein false discovery rates (FDRs). The signal intensity data for 4241 proteins (Data_S2_Proteome.csv) was used for further data analysis below.

Data analysis

The analysis and visualization of RNAseq data and proteome data were conducted using custom Python code, and coding and execution were carried out in the Collaboratory. In the RNAseq data, 6,685 protein-coding genes were detected (Data_S1_RNAseq.csv). Among these, 4,291 genes assigned to various categories of KEGG Orthology level 3 (Kanehisa et al., 2023) were placed into 77 categories. Analysis was performed on 47 categories that contained more than 10 gene elements. We investigated whether the expression of genes in each category, as a group, was significantly higher or lower upon overexpression of mox and mox-YG compared to the vector. Specifically, mox/vector (log2FC) and mox-YG/vector (log2FC) were calculated along with their FDR values, and each gene was assigned to a KEGG orthology level 3 category. Next, whether the gene groups within each category showed significantly higher or lower expression changes compared to gene groups outside of their category was tested using the Mann-Whitney U test. Significant differences were indicated in the graphs with a red border and red stripes. The graphs showed the overall distribution of genes within each category using violin plots, and only genes showing significant differences in mox/vector (log2FC) and mox-YG/vector (log2FC) were highlighted in swarm plots. An asterisk indicated significant differences on the graphs when the expression changes (log2FC) between mox/vector and mox-YG/vector were significant by the Mann-Whitney U test. The significance threshold was set at FDR < 0.05 for all. Only categories where either mox or mox-YG overexpression showed significant expression changes were extracted and presented as Figure 4B, and all 47 categories were shown in Figure 4_S1.

The proteome analysis was conducted using the signal intensity data (Data_S2_proteome.csv). This data includes proteins that were detected only in some samples under certain conditions or replicates, especially those with low signals or large expression variations. Additionally, the total protein signal sum excluding mox-YG was almost the same between the vector control and mox-YG overexpression (Figure 5_S2A). This result contradicts the SDS−PAGE findings, which indicate that the total protein mass per cell, including mox-YG, is the same between the vector control and mox-YG overexpression (Figure 5A). Therefore, it is considered necessary to adjust the total protein signal sum, including mox-YG, to match that of the vector control to understand the proteome change per cell. In this study, however, we used the most conservative approach by removing proteins with zero values and not applying any overall corrections (Figure 5_S2B), which allowed the analysis of 3,588 proteins. Although this may not correctly analyze “per cell” proteome change, it enables us to capture the change in proteome composition (excluding mox-YG), thus providing information about cellular responses. If the total correction is applied, only 138 proteins were increased (Figure 5_S2B), most of which are mitochondrial proteins.

For the detected 3,588 proteins above, we compared the increase or decrease of protein groups within each category of KEGG Orthology Level 3 with those outside the category. Analysis was performed on 45 categories, each containing more than 10 proteins, to investigate whether the expression of protein groups in each category was significantly higher or lower under mox-YG overexpression compared to the vector control. Specifically, the average log2FC and its FDR values were calculated using data from three replicates for mox-YG/vector, and each protein was assigned to a KEGG orthology level 3 category. Next, whether the protein groups within each category showed significantly higher or lower expression changes compared to protein groups outside of their category was tested using the Mann-Whitney U test. Significant differences were indicated in the graphs with a red border and red stripes. The graphs showed the overall distribution of proteins within each category using violin plots, and only proteins showing significant differences in mox-YG/vector (log2FC) were highlighted in swarm plots. The significance threshold was set at FDR < 0.05 for all. Only categories where mox-YG overexpression showed significant expression changes were extracted and presented in Figure 5C and all 47 categories were shown in Figure 5_S3. Published data (Gowans et al., 2018) was used to separate protein groups into rapamycin-responsive and non-responsive, and similar analysis and visualization were conducted (Figures 5_S4A, B).

Fluorescence microscopic observation

Cells overexpressing target proteins were cultured in SC-LeuUra medium and observed under an inverted microscope (DMI 6000 B, Leica) at log phase (OD660 = 1.0). Images were acquired with Leica Application Suite and processed with Leica Application Suite X software (Leica Microsystems). Cells in brightfield images were separated from the images using yeast_segmentation-master_v3 (Lu et al., 2019). Cell morphological traits were assessed using parameters obtained when analysed using CellProfiler (Ver. 4.2.6) (Carpenter et al., 2006). “AreaShape_Area” was used to analyse FP-overexpressing cell size in Figure 2_S5B, Nsr1-GFP in Figure 6_S2C and Nsr1-mScarlet-I in Figure 6_S5E, S6F. For analysis of the fluorescence intensity of Rip1-mScarlet-I in Figure 5_S5B, “Intensity_IntegratedIntensity” was used. Intensity_MeanIntensity” was used to analyze the fluorescence intensity of Nsr1-GFP in Figure 6_S2D. Ten images per experiment were used for analysis.

Electron microscopic observation

Cells grown to log phase (OD₆₀₀ = 1.0) at 30°C in SC–LeuUra medium were collected and transported to Tokai Electron Microscope Co. Transported samples were sandwiched between copper plates, flash-frozen with liquefied propane, and then dehydrated with anhydrous ethanol. cells were fixed with a 5:5 mixture of propylene oxide and resin (Quetol-651, Nissin EM Co.). Fixed cells were cut with an ultramicrotome (Ultracut UCT, Leica) to prepare 80 nm thick sections, stained with 2% uranyl acetate and lead stain solution (Sigma-Aldrich Co.), and observed with a transmission electron microscope (JEM-1400 Plus, JEOL Ltd.).

Acknowledgements

We thank the members of the Moriya laboratory (Okayama University) for their helpful discussions. This work was partly supported by JSPS KAKENHI Grant Numbers 18H04824 20H03242, and 20H04870.