If a cell makes too much of a given protein, it can sometimes cause problems and impair the cell’s growth. Overproducing some proteins may deplete the cell’s limited resources, meaning it does not have enough to make other more essential proteins. This phenomenon is known as the protein burden effect. Theoretically, only harmless proteins can be overproduced up to a level where growth would be impaired in this way. Conversely, if an overproduced protein causes harm before it becomes a burden on resources, scientists must consider other mechanisms to explain the cell’s problems, namely that the protein itself is harmful.

Knowing the ultimate level of protein production that could cause the protein burden effect – the protein burden limit – would allow scientists to distinguish between harmful and non-harmful proteins. However, to date, this limit had not been defined for any cell.

Eguchi et al. have now tried to estimate the protein burden limit for budding yeast – one of the best-studied experimental organisms. The experiments first focused on enzymes involved in alcoholic fermentation because they were expected to be non-harmful. Some of these enzymes were overproduced to the level were the made up 15% of all the cell’s proteins before they started to cause growth defects. The same results were seen with versions of the enzymes that had been mutated to be less active, leading Eguchi et al. to conclude that this level is the protein burden limit.

In other experiments, harmful enzymes could only be overproduced to levels that were far less than this proposed protein burden limit. These enzymes caused problems for the yeast in several ways, including interfering with biochemical reactions and forming large aggregates in the cell. Lastly, Eguchi et al. looked at the yeast’s genetic code and saw that most of its genes seemed to have evolved to specifically limit the production of proteins to a level that would avoid the unwanted protein burden effect.

Together these findings establish a framework to clearly distinguish between harmful and non-harmful proteins. This framework will be useful to understand the different reasons why the overproduction of certain proteins, which is seen in neurodegenerative diseases and cancer cells, can cause problems for cells.

Protein overexpression is sometimes harmful to cellular growth (Makanae et al., 2013; Sopko et al., 2006), and a few mechanisms that could result in overexpression-triggered growth defects have been proposed (Moriya, 2015). Resource overload, stoichiometric imbalance, promiscuous interaction, and pathway modulation are triggered upon overexpression of, respectively, (i) a protein that has a high demand of cellular resources (Dong et al., 1995; Kintaka et al., 2016; Stoebel et al., 2008), (ii) a protein that is part of a protein complex (Kaizu et al., 2010; Makanae et al., 2013; Papp et al., 2003), (iii) a protein with a nonspecific interaction domain (Ma et al., 2010; Vavouri et al., 2009), and (iv) a protein that catalyzes a pathway (Prelich, 2012; Youn et al., 2017). The mechanism of protein overexpression-triggered growth defects depends on the protein’s structural and functional characteristics, which are not always fully understood yet. Therefore, it is still difficult to predict whether the overexpression of a particular protein will be harmful to cellular growth and which mechanisms cause the harmful effect.

The ultimate overexpression of a protein could be harmful for cellular growth, because it monopolizes and depletes limited resources that are involved in protein production, such as ribosomes and aminoacyl-tRNAs (Gong et al., 2006; Shachrai et al., 2010; Vind et al., 1993). This phenomenon is known as the protein burden/cost effect (Kafri et al., 2016; Snoep et al., 1995). Proteins that have no harmful effects on cellular functions can be overexpressed up to a level that causes protein-burden–triggered growth defects. Conversely, if a protein cannot be overexpressed up to that level because it adversely affects cellular functioning, then overexpression of that protein will cause growth defects at relatively low expression levels, and we should consider mechanisms causing the defects.

We previously developed a genetic tug-of-war (gTOW) method that can be used to estimate the expression limit of a target protein that triggers growth defects in the yeast Saccharomyces cerevisiae (Makanae et al., 2013; Moriya et al., 2006 , 2012 ). We estimated that the expression limit of a green fluorescent protein (GFP) was about 15% of the total cellular protein in S. cerevisiae (Kintaka et al., 2016). Because GFP is a highly structured cytoplasmic protein unrelated to the cellular functions of yeast and thus harmless, this level could be considered the expression limit for any protein that causes growth defects triggered by the protein-burden effect.

We predicted that the expression limits of some native highly expressed glycolytic proteins would be similar (>15% of the total cellular protein) (Moriya, 2015), suggesting that overexpression of these proteins would be harmless even though they have metabolic functions in yeast. The prediction was performed by the calculation of the proteins’ native expression levels (Kulak et al., 2014) and their gene copy number limits as determined by gTOW analysis, and has not yet been experimentally validated (Moriya, 2015). In this study, therefore, we tried to measure the expression limits of 29 glycolytic proteins to assess whether they are expressed up to levels that cause growth defects triggered by the protein-burden effect. There are five reasons why we chose glycolytic proteins: (1) because they are generally highly expressed and thus considered non-harmful upon high-level expression, they are excellent targets for examining whether they are expressed up to the protein-burden limit; (2) because they have been intensely studied, we have information that can allow us to manipulate their catalytic activities; (3) because the glycolytic pathway is one of the best-known metabolic pathways, we can predict and measure metabolic changes upon overexpression of these proteins; (4) they include a heteromer (Pfks), a mitochondrially localized protein (Adh3), and membrane proteins (Hxts), so we can assess how these characteristics affect expression limits; and (5) they include paralogs whose expressions are differently regulated, so that we can test how their differences affect their expression limits.

We found that the expression limits of most of the 29 proteins were comparable to that of GFP and were independently determined by their catalytic activities, as suggested by a kinetic model of yeast glycolysis, confirming that their overexpression was harmless. Also, some of the proteins had far lower expression limits than those that would create a protein burden (the expression limit of GFP), and their harmful effects were derived from their localization and metabolic perturbations. Owing to their codon optimality, native poorly expressed isozymes were not produced at levels sufficient to cause growth defects, even when they were expressed from the strong TDH3 promoter on the multicopy plasmids. Some glycolytic proteins formed S–S-bond-mediated aggregates when overexpressed, and this aggregation also seemed to restrict their expression limits.

According to the protein-burden concept (Dong et al., 1995; Kafri et al., 2016; Shah et al., 2013; Snoep et al., 1995; Stoebel et al., 2008), the ultimate overexpression of any protein could cause growth defects by overloading basic protein production resources. But only non-harmful proteins can be overexpressed up to the ultimate level, or the protein-burden limit, because the expression limit of harmful proteins should be restricted by their harmful effects. Knowing the protein-burden limit itself is thus essential when seeking to determine whether the overexpression of a protein is harmful to cellular functions. We previously estimated the protein-burden limit of S. cerevisiae cells by measuring the expression level of GFP that causes growth defects. This was 15% of the total cellular protein (Kintaka et al., 2016).

In this study, we first tried to measure the expression limits of yeast glycolytic proteins in order to confirm whether the protein-burden limit measured using GFP applies to endogenous proteins. Most of the glycolytic proteins studied here caused growth defects when they were expressed from a strong promoter on a multicopy plasmid (Figure 1). The expression levels of some glycolytic proteins in these conditions were, indeed, comparative or even higher than that of GFP (Figure 2). Also, their expression levels did not increase due to mutations in their catalytic centers (Figure 3A). These results strongly suggest that the protein-burden effect largely determines the expression limit, and that the limit is around 15% of the total cellular protein. Among the glycolytic proteins studied here, Pgk1, Gpm1, and Eno2 had the highest expression limits. Although Pgk1 (44.7 kDa) and Eno2 (46.9 kDa) are 1.5-fold larger than Gpm1 (27.6 kDa), their expression limits were similar to those of Gpm1 (Figure 2B). These results suggest that a protein's size does not affect its expression limit, at least for proteins in this molecular weight range. These data also suggest that the expression limits of proteins are not determined by the molar concentrations of those proteins but by the cost of the protein production.

Some other glycolytic proteins, such as Pfk1, Pfk2, Adh3, and Hxts, showed expression limits far below the protein burden limit of 15% (Figure 2), suggesting that overexpression of these proteins is harmful. Of the 18 glycolytic proteins studied, Pfk1 and Pfk2 were the only ones whose expression limits were significantly increased by mutations in their catalytic centers (Figure 3A–B), suggesting that their metabolic functions restrict their expression limits. We think, however, that the metabolic perturbations that trigged the overexpression only partially affect the expression limits because the expression limits of the mutant proteins were still far below those of other glycolytic proteins (Figure 3A–B). Pfk1 and Pfk2 form a hetero-octameric complex, and their stoichiometric imbalance leads to the formation of filamentous Pfk1 structures in the cytosol (Schwock et al., 2004). This stoichiometry-imbalance-triggered protein aggregate might cause growth defects upon overexpression of Pfk1 (and Pfk2), although we could not confirm this hypothesis because simultaneous overexpression of Pfk1 and Pfk2 did not increase the expression limits of these proteins (our unpublished observation).

The CC mutants of Fba1 and Pgk1 showed lower expression limits than their wild-type proteins (Figure 3A). We currently do not have any substantial and consistent explanation of why these CC mutants have lower expression limits. We can assume some general mechanisms: CC mutant proteins sequester the wild-type enzymes into inactive complexes; CC mutant proteins sequester the substrate molecules for the wild-type enzymes; or mutation in the catalytic center destabilizes the structure of the enzyme. For example, Fba1 is an essential homodimeric enzyme (UniProtKB: P14540). Overexpression of CC mutant Fba1 molecules might sequester active wild-type Fba1 molecules into inactive complexes. The limit of CC mutant Tdh3 was higher than that of the wild-type in low-copy conditions whereas it was lower in high-copy conditions (Figure 3A–B). This strange behavior might be related to its moonlighting function. The catalytic activity of Tdh3 did not seem to explain the difference in the expression limits of wild-type and CC mutant Tdh3 (Figure 5C). Beside its metabolic function, Tdh3 directly binds to Sir2 protein to promote transcriptional silencing, and a mutation in the catalytic center (C150G) reduces the silencing (Ringel et al., 2013). It is thus possible that the CC mutant Tdh3 (C150S) causes silencing in a dose-dependent manner by competing with wild-type Tdh3 for binding with Sir2.

We speculated that the localization of Adh3 to the mitochondria and of Hxts to the plasma membrane restricted their expression limits because localized proteins overload more-limited localization resources (Kintaka et al., 2016). This hypothesis was confirmed because the removal of the mitochondrial signal from Adh3 increased its expression limit (Figure 3C,D). We also speculated that the expression limits of membrane proteins such as Hxts should be restricted by their localization, although there is no experimental evidence to support this hypothesis yet.

The fact that the expression limits of most glycolytic proteins were not affected by mutations in their catalytic centers (Figure 3A) suggests that their overexpression does not cause metabolic perturbations. This finding was theoretically confirmed by simulations using a kinetic model of glycolytic metabolism (Figure 4). The reason why their overexpression does not cause metabolic perturbations is probably that they are bidirectional enzymes: the metabolic flux should be determined only by the availability of substrates when the concentrations of these enzymes are more than a certain level. To support this idea, the overexpression of 14 bidirectional enzymes showed minor metabolic changes, whereas the overexpression of 6 unidirectional enzymes (including Hxks, Pfks, Cdc19, and Pdc1) showed strong metabolic changes in the simulation (Figure 4). The expression limits of Hxks in the cells, however, were close to the protein burden limit (Figure 2B) and were not affected by mutations in the catalytic center (Figure 3A). These results suggest an additional mechanism that is not implemented into the model that allows cells to avoid the effects of big metabolic changes upon overexpression of Hxks: a mechanism that prevents these metabolic perturbations from occurring, or a mechanism that prevents these metabolic perturbations from causing growth defects.

Through the metabolic analysis, we realized that we currently do not have any systematic way to identify metabolic changes that are directly triggered by the overexpression of an enzyme, because metabolism is interconnected and the overexpression of a protein could cause non-specific perturbations that ultimately affect metabolism. Moreover, we know very little about how much change in which metabolite triggers a growth defect. Comparison of the metabolic changes in cells overexpressing wild-type and CC-mutant enzymes could be one solution for this. In fact, we observed a three-fold difference between cells expressing wild-type and CC mutant Pfk2 (Figure 5A). Nevertheless, once again, we cannot conclude from our current knowledge that this difference causes the difference in the expression limits of these two forms of Pfk2. By using a mathematical model, we tried to predict the potential metabolic changes that would be triggered by overexpression of an enzyme without considering unknown effects other than the enzyme's metabolic activity. In the simulations, overexpression of Pfks and Hxks triggered divergent and almost catastrophic metabolic changes (~1000-fold increase in some metabolites, Figure 4B,D), suggesting that their overexpression would cause growth defects due to these strong metabolic perturbations. We thus expected to obtain similar metabolic changes upon overexpression of Pfks, whose CC mutants had higher expression limits. We did not, however, observe such great changes (Figure 5A–B and Figure 5—figure supplement 1). To answer these issues precisely, we need a much deeper understanding of the connections between metabolite levels and cellular growth.

The translational rate of some glycolytic proteins, including Glk1, seemed low because of their lower codon optimality (Figure 6). Actually, the codon optimality of Glk1 (tAIg = 0.38) is close to the average for all the yeast genes (tAIg = 0.37), and the codon optimality of other glycolytic proteins studied here is exceptionally high (Figure 6—figure supplement 1). These observations suggest that the codon optimality of most yeast genes is not high enough to allow expression of their proteins up to the protein-burden limit, even if they are expressed from a strong promoter on a multicopy plasmid.

Overexpression of Eno2, Pgk1, and Tpi1 triggered S–S-bond-connected aggregation (Figures 7 and 8), and the aggregates that are formed contain other glycolytic proteins and translational factors (Figure 7C–D). We think that this aggregation is triggered by spontaneous non-specific S–S bond formation among proteins existing in high concentrations. Interestingly, we also detected the same proteins within the gel of the corresponding molecular weight in the vector control, although the amounts estimated by LC-MS/MS were lower and cannot be identified as visible protein bands (Figure 7—source data 1). Therefore, we speculated that the S–S-bond-mediated protein aggregation occurs even in normal physiological conditions, but it is accelerated by an increase in the concentration of cytoplasmic proteins upon overexpression of glycolytic proteins. This aggregation might affect the expression limits of cysteine-containing glycolytic proteins, because changing the cysteine residues of Tpi1 into serine residues increases the protein's expression limit (Figure 8B). As the amount of protein corresponding to the Tpi1 monomer was not changed by DTT treatment, the expression level of Tpi1 should not be reduced simply by aggregation but by the harmful effect of spontaneous S–S bond formation. This hypothesis is supported by the fact that the most highly expressed glycolytic protein Gpm1, which has a molecular weight similar to that of Tpi1, does not have a cysteine residue. The deleterious effect of this aggregation, however, seems protein-specific because the expression limits of Pgk1 and Eno1 were among highest measured (Figure 2A), and removal of their cysteine did not increase their expression limits (Figure 7—figure supplement 1).

As described above, we revealed mechanisms that restrict the expression limits of some glycolytic proteins. We do not think, however, that these mechanisms are the sole factors restricting the expression limits of these proteins. The expression limits of ΔMTS-Adh3 (0.45 AU, Figure 3D) and CoGlkl (1.07 AU, Figure 6A) are still lower than those of other high limit proteins such as Pgk1 and Gpm1 (2.26 AU and 2.63 AU, respectively, Figure 2B). It is thus likely that multiple mechanisms restrict the expression limits of these proteins.

Protein misfolding or misinteraction is considered to cause toxicity upon high-level expression of a protein with low translational robustness, low folding stability, or a high propensity for misinteraction (Drummond and Wilke, 2009; Zhang and Yang, 2015). In general, highly expressed proteins such as glycolytic proteins are thus evolved to avoid these characteristics (Zhang and Yang, 2015), and that should be a requirement for a protein to be expressed up to the protein-burden limit. Cdc19, one of the glycolytic proteins studied here, aggregates in a stress-induced and reversible manner through a region of low compositional complexity (Saad et al., 2017). This aggregation capacity of Cdc19 might explain why its expression limit (0.42 AU) is lower than the protein burden limit (>2.0 AU) (Figure 2B). Our finding in Figure 8 suggested that the high-level expression of a cysteine-containing protein could also cause a misinteraction-triggered toxic effect; hence unimportant cysteines should be avoided in highly expressed proteins. Concentration-dependent liquid phase separation is also considered to cause toxicity upon overexpression of structurally disordered and nucleic-acid-binding proteins (Bolognesi et al., 2016). We do not think that this mechanism caused growth defects upon overexpression of the glycolytic proteins studied here because they are less structurally disordered (Moriya, 2015) and not nucleic-acid-binding proteins.

We summarize our analysis in Supplementary file 1. In conclusion, we established the ultimate expression level that causes cellular growth defects due to the protein-burden effect as around 15% of the total cellular protein. The next interesting theme is to identify characteristics of proteins that can be overexpressed up to the protein-burden limit because such proteins are considered non-harmful to cellular functions. Those characteristics should conversely imply the properties of proteins that are harmful when they are overexpressed.

All data generated or analysed during this study are included in the manuscript and supporting files.

1. Book

   1. Amberg DC
   2. Burke D
   3. Strathern JN
   4. Cold Spring Harbor, L

   (2005)

   Methods in Yeast Genetics: A Cold Spring Harbor Laboratory Course Manual

   Cold Spring Harbor Laboratory Press.

   • Google Scholar
2. 1. Bolognesi B
   2. Lorenzo Gotor N
   3. Dhar R
   4. Cirillo D
   5. Baldrighi M
   6. Tartaglia GG
   7. Lehner B

   (2016) A Concentration-Dependent liquid phase separation can cause toxicity upon increased protein expression

   Cell Reports 16:222–231.

   https://doi.org/10.1016/j.celrep.2016.05.076

   • PubMed
   • Google Scholar
3. 1. Brachmann CB
   2. Davies A
   3. Cost GJ
   4. Caputo E
   5. Li J
   6. Hieter P
   7. Boeke JD

   (1998) Designer deletion strains derived from Saccharomyces cerevisiae S288C: a useful set of strains and plasmids for PCR-mediated gene disruption and other applications

   Yeast 14:115–132.

   https://doi.org/10.1002/(SICI)1097-0061(19980130)14:2<115::AID-YEA204>3.0.CO;2-2

   • PubMed
   • Google Scholar
4. 1. Dong H
   2. Nilsson L
   3. Kurland CG

   (1995) Gratuitous overexpression of genes in Escherichia coli leads to growth inhibition and ribosome destruction

   Journal of Bacteriology 177:1497–1504.

   https://doi.org/10.1128/jb.177.6.1497-1504.1995

   • PubMed
   • Google Scholar
5. 1. Drummond DA
   2. Wilke CO

   (2009) The evolutionary consequences of erroneous protein synthesis

   Nature Reviews Genetics 10:715–724.

   https://doi.org/10.1038/nrg2662

   • PubMed
   • Google Scholar
6. 1. Gong M
   2. Gong F
   3. Yanofsky C

   (2006) Overexpression of tnaC of Escherichia coli inhibits growth by depleting tRNA2Pro availability

   Journal of Bacteriology 188:1892–1898.

   https://doi.org/10.1128/JB.188.5.1892-1898.2006

   • PubMed
   • Google Scholar
7. 1. Jungbluth M
   2. Renicke C
   3. Taxis C

   (2010) Targeted protein depletion in Saccharomyces cerevisiae by activation of a bidirectional degron

   BMC Systems Biology 4:176.

   https://doi.org/10.1186/1752-0509-4-176

   • PubMed
   • Google Scholar
8. 1. Kafri M
   2. Metzl-Raz E
   3. Jona G
   4. Barkai N

   (2016) The cost of protein production

   Cell Reports 14:22–31.

   https://doi.org/10.1016/j.celrep.2015.12.015

   • PubMed
   • Google Scholar
9. 1. Kaizu K
   2. Moriya H
   3. Kitano H

   (2010) Fragilities caused by dosage imbalance in regulation of the budding yeast cell cycle

   PLoS Genetics 6:e1000919.

   https://doi.org/10.1371/journal.pgen.1000919

   • PubMed
   • Google Scholar
10. 1. Kato H
    2. Izumi Y
    3. Hasunuma T
    4. Matsuda F
    5. Kondo A

    (2012) Widely targeted metabolic profiling analysis of yeast central metabolites

    Journal of Bioscience and Bioengineering 113:665–673.

    https://doi.org/10.1016/j.jbiosc.2011.12.013

    • PubMed
    • Google Scholar
11. 1. Kintaka R
    2. Makanae K
    3. Moriya H

    (2016) Cellular growth defects triggered by an overload of protein localization processes

    Scientific Reports 6:31774.

    https://doi.org/10.1038/srep31774

    • PubMed
    • Google Scholar
12. 1. Kito K
    2. Ito H
    3. Nohara T
    4. Ohnishi M
    5. Ishibashi Y
    6. Takeda D

    (2016) Yeast interspecies comparative proteomics reveals divergence in expression profiles and provides insights into proteome resource allocation and evolutionary roles of gene duplication

    Molecular & Cellular Proteomics 15:218–235.

    https://doi.org/10.1074/mcp.M115.051854

    • PubMed
    • Google Scholar
13. 1. Kulak NA
    2. Pichler G
    3. Paron I
    4. Nagaraj N
    5. Mann M

    (2014) Minimal, encapsulated proteomic-sample processing applied to copy-number estimation in eukaryotic cells

    Nature Methods 11:319–324.

    https://doi.org/10.1038/nmeth.2834

    • PubMed
    • Google Scholar
14. 1. Kushnirov VV

    (2000) Rapid and reliable protein extraction from yeast

    Yeast 16:857–860.

    https://doi.org/10.1002/1097-0061(20000630)16:9<857::AID-YEA561>3.0.CO;2-B

    • PubMed
    • Google Scholar
15. 1. Le Novère N
    2. Hucka M
    3. Mi H
    4. Moodie S
    5. Schreiber F
    6. Sorokin A
    7. Demir E
    8. Wegner K
    9. Aladjem MI
    10. Wimalaratne SM
    11. Bergman FT
    12. Gauges R
    13. Ghazal P
    14. Kawaji H
    15. Li L
    16. Matsuoka Y
    17. Villéger A
    18. Boyd SE
    19. Calzone L
    20. Courtot M
    21. Dogrusoz U
    22. Freeman TC
    23. Funahashi A
    24. Ghosh S
    25. Jouraku A
    26. Kim S
    27. Kolpakov F
    28. Luna A
    29. Sahle S
    30. Schmidt E
    31. Watterson S
    32. Wu G
    33. Goryanin I
    34. Kell DB
    35. Sander C
    36. Sauro H
    37. Snoep JL
    38. Kohn K
    39. Kitano H

    (2009) The systems biology graphical notation

    Nature Biotechnology 27:735–741.

    https://doi.org/10.1038/nbt.1558

    • PubMed
    • Google Scholar
16. 1. Ma L
    2. Pang CN
    3. Li SS
    4. Wilkins MR

    (2010) Proteins deleterious on overexpression are associated with high intrinsic disorder, specific interaction domains, and low abundance

    Journal of Proteome Research 9:1218–1225.

    https://doi.org/10.1021/pr900693e

    • PubMed
    • Google Scholar
17. 1. Makanae K
    2. Kintaka R
    3. Makino T
    4. Kitano H
    5. Moriya H

    (2013) Identification of dosage-sensitive genes in Saccharomyces cerevisiae using the genetic tug-of-war method

    Genome Research 23:300–311.

    https://doi.org/10.1101/gr.146662.112

    • PubMed
    • Google Scholar
18. 1. Moriya H
    2. Makanae K
    3. Watanabe K
    4. Chino A
    5. Shimizu-Yoshida Y

    (2012) Robustness analysis of cellular systems using the genetic tug-of-war method

    Molecular BioSystems 8:2513–2522.

    https://doi.org/10.1039/c2mb25100k

    • PubMed
    • Google Scholar
19. 1. Moriya H
    2. Shimizu-Yoshida Y
    3. Kitano H

    (2006) In vivo robustness analysis of cell division cycle genes in Saccharomyces cerevisiae

    PLoS Genetics 2:e111.

    https://doi.org/10.1371/journal.pgen.0020111

    • PubMed
    • Google Scholar
20. 1. Moriya H

    (2015) Quantitative nature of overexpression experiments

    Molecular Biology of the Cell 26:3932–3939.

    https://doi.org/10.1091/mbc.e15-07-0512

    • PubMed
    • Google Scholar
21. 1. Oldenburg KR
    2. Vo KT
    3. Michaelis S
    4. Paddon C

    (1997) Recombination-mediated PCR-directed plasmid construction in vivo in yeast

    Nucleic Acids Research 25:451–452.

    https://doi.org/10.1093/nar/25.2.451

    • PubMed
    • Google Scholar
22. 1. Papp B
    2. Pál C
    3. Hurst LD

    (2003) Dosage sensitivity and the evolution of gene families in yeast

    Nature 424:194–197.

    https://doi.org/10.1038/nature01771

    • PubMed
    • Google Scholar
23. 1. Prelich G

    (2012) Gene overexpression: uses, mechanisms, and interpretation

    Genetics 190:841–854.

    https://doi.org/10.1534/genetics.111.136911

    • PubMed
    • Google Scholar
24. 1. Presnyak V
    2. Alhusaini N
    3. Chen YH
    4. Martin S
    5. Morris N
    6. Kline N
    7. Olson S
    8. Weinberg D
    9. Baker KE
    10. Graveley BR
    11. Coller J

    (2015) Codon optimality is a major determinant of mRNA stability

    Cell 160:1111–1124.

    https://doi.org/10.1016/j.cell.2015.02.029

    • PubMed
    • Google Scholar
25. 1. Ringel AE
    2. Ryznar R
    3. Picariello H
    4. Huang KL
    5. Lazarus AG
    6. Holmes SG

    (2013) Yeast Tdh3 (glyceraldehyde 3-phosphate dehydrogenase) is a Sir2-interacting factor that regulates transcriptional silencing and rDNA recombination

    PLoS Genetics 9:e1003871.

    https://doi.org/10.1371/journal.pgen.1003871

    • PubMed
    • Google Scholar
26. 1. Saad S
    2. Cereghetti G
    3. Feng Y
    4. Picotti P
    5. Peter M
    6. Dechant R

    (2017) Reversible protein aggregation is a protective mechanism to ensure cell cycle restart after stress

    Nature Cell Biology 19:1202–1213.

    https://doi.org/10.1038/ncb3600

    • PubMed
    • Google Scholar
27. 1. Schwock J
    2. Kirchberger J
    3. Edelmann A
    4. Kriegel TM
    5. Kopperschläger G

    (2004) Interaction of 6-phosphofructokinase with cytosolic proteins of Saccharomyces cerevisiae

    Yeast 21:483–494.

    https://doi.org/10.1002/yea.1114

    • PubMed
    • Google Scholar
28. 1. Shachrai I
    2. Zaslaver A
    3. Alon U
    4. Dekel E

    (2010) Cost of unneeded proteins in E. coli is reduced after several generations in exponential growth

    Molecular Cell 38:758–767.

    https://doi.org/10.1016/j.molcel.2010.04.015

    • PubMed
    • Google Scholar
29. 1. Shah P
    2. Ding Y
    3. Niemczyk M
    4. Kudla G
    5. Plotkin JB

    (2013) Rate-limiting steps in yeast protein translation

    Cell 153:1589–1601.

    https://doi.org/10.1016/j.cell.2013.05.049

    • PubMed
    • Google Scholar
30. 1. Smallbone K
    2. Messiha HL
    3. Carroll KM
    4. Winder CL
    5. Malys N
    6. Dunn WB
    7. Murabito E
    8. Swainston N
    9. Dada JO
    10. Khan F
    11. Pir P
    12. Simeonidis E
    13. Spasić I
    14. Wishart J
    15. Weichart D
    16. Hayes NW
    17. Jameson D
    18. Broomhead DS
    19. Oliver SG
    20. Gaskell SJ
    21. McCarthy JE
    22. Paton NW
    23. Westerhoff HV
    24. Kell DB
    25. Mendes P

    (2013) A model of yeast glycolysis based on a consistent kinetic characterisation of all its enzymes

    FEBS Letters 587:2832–2841.

    https://doi.org/10.1016/j.febslet.2013.06.043

    • PubMed
    • Google Scholar
31. 1. Snoep JL
    2. Yomano LP
    3. Westerhoff HV
    4. Ingram LO

    (1995) Protein burden in Zymomonas mobilis: negative flux and growth control due to overproduction of glycolytic enzymes

    Microbiology 141:2329–2337.

    https://doi.org/10.1099/13500872-141-9-2329

    • Google Scholar
32. 1. Sopko R
    2. Huang D
    3. Preston N
    4. Chua G
    5. Papp B
    6. Kafadar K
    7. Snyder M
    8. Oliver SG
    9. Cyert M
    10. Hughes TR
    11. Boone C
    12. Andrews B

    (2006) Mapping pathways and phenotypes by systematic gene overexpression

    Molecular Cell 21:319–330.

    https://doi.org/10.1016/j.molcel.2005.12.011

    • PubMed
    • Google Scholar
33. 1. Stoebel DM
    2. Dean AM
    3. Dykhuizen DE

    (2008) The cost of expression of Escherichia coli lac operon proteins is in the process, not in the products

    Genetics 178:1653–1660.

    https://doi.org/10.1534/genetics.107.085399

    • PubMed
    • Google Scholar
34. 1. Tuller T
    2. Carmi A
    3. Vestsigian K
    4. Navon S
    5. Dorfan Y
    6. Zaborske J
    7. Pan T
    8. Dahan O
    9. Furman I
    10. Pilpel Y

    (2010) An evolutionarily conserved mechanism for controlling the efficiency of protein translation

    Cell 141:344–354.

    https://doi.org/10.1016/j.cell.2010.03.031

    • PubMed
    • Google Scholar
35. 1. Vavouri T
    2. Semple JI
    3. Garcia-Verdugo R
    4. Lehner B

    (2009) Intrinsic protein disorder and interaction promiscuity are widely associated with dosage sensitivity

    Cell 138:198–208.

    https://doi.org/10.1016/j.cell.2009.04.029

    • PubMed
    • Google Scholar
36. 1. Vind J
    2. Sørensen MA
    3. Rasmussen MD
    4. Pedersen S

    (1993) Synthesis of proteins in Escherichia coli is limited by the concentration of free ribosomes. Expression from reporter genes does not always reflect functional mRNA levels

    Journal of Molecular Biology 231:678–688.

    https://doi.org/10.1006/jmbi.1993.1319

    • PubMed
    • Google Scholar
37. 1. Youn J-Y
    2. Friesen H
    3. Nguyen Ba AN
    4. Liang W
    5. Messier V
    6. Cox MJ
    7. Moses AM
    8. Andrews B

    (2017) Functional analysis of kinases and transcription factors in Saccharomyces cerevisiae using an integrated overexpression library

    G3: Genes|Genomes|Genetics 7:911–921.

    https://doi.org/10.1534/g3.116.038471

    • Google Scholar
38. 1. Young ET
    2. Pilgrim D

    (1985) Isolation and DNA sequence of ADH3, a nuclear gene encoding the mitochondrial isozyme of alcohol dehydrogenase in Saccharomyces cerevisiae

    Molecular and Cellular Biology 5:3024–3034.

    https://doi.org/10.1128/MCB.5.11.3024

    • PubMed
    • Google Scholar
39. 1. Zampar GG
    2. Kümmel A
    3. Ewald J
    4. Jol S
    5. Niebel B
    6. Picotti P
    7. Aebersold R
    8. Sauer U
    9. Zamboni N
    10. Heinemann M

    (2013) Temporal system-level organization of the switch from glycolytic to gluconeogenic operation in yeast

    Molecular Systems Biology 9:651.

    https://doi.org/10.1038/msb.2013.11

    • PubMed
    • Google Scholar
40. 1. Zhang J
    2. Yang JR

    (2015) Determinants of the rate of protein sequence evolution

    Nature Reviews Genetics 16:409–420.

    https://doi.org/10.1038/nrg3950

    • PubMed
    • Google Scholar

Thank you for submitting your article "Estimating the Protein Burden Limit of Yeast Cells by Measuring Expression Limits of Glycolytic Proteins" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Detlef Weigel as the Senior Editor. The following individual involved in review of your submission has agreed to reveal his identity: Claus O Wilke (Reviewer #3).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

Moriya et al., attempt to determine the protein expression limits of glycolytic proteins. They measure the growth rate, protein expression and copy number of glycolytic proteins using both high copy and low copy plasmids. They go on to estimate the contributions of various factors that might determine the upper limit of protein expression such as metabolic activity, codon usage, membrane localization and disulfide bond formation. The authors conclude that while metabolic activity has no role in determining the expression limit, disulfide bond formation, membrane localization and sub-optimal codon usage limit the extent of protein expression. The authors assert the ability of their pipeline to differentiate between proteins that can be overexpressed from those that will be harmful upon overexpression.

Essential revisions:

1) One of the main findings of the paper is that protein expression limits are not determined by the metabolic activity. The authors base this primarily on the lack of a difference between expression levels of proteins and their catalytic mutants (proteins with mutations in their catalytic sites). The authors need to verify that metabolism is indeed altered to unequivocally comment on the link between metabolic activity and protein expression levels. Instances that highlight the need to perform metabolic measurements in the manuscript are:

A) The authors measured metabolic activity for only one protein-mutant pair (Pfk2, Figure 4), which did not show any difference in protein expression levels (Figure 3A). The authors need to measure the metabolic levels for pairs of proteins that show a significant difference in their expression levels like Pfk1 or Tdh3 to be able to comment on the link between metabolism and protein levels.

B) In addition, there are several contradictions in the effect of these catalytic mutants on protein expression levels. Figure 3A shows that in two cases the mutant has higher expression (Pfk1 and Tdh3) but in other two cases, the wild-type has higher expression (Fba1, Eno1). How do the authors explain such differences? In addition, Tdh3 mutant has lower expression level compared to wild-type when expressed from a high copy plasmid. How do the authors explain this flip?

C) Discussion, fifth paragraph: The authors claim that one of the reasons why they don't see any association between metabolic activity and expression is that the majority of these enzymes are bidirectional. This is not true for all the enzymes as some glycolytic enzymes are unidirectional. In addition, the authors need to show a control example where unidirectional enzyme has a higher protein expression in order to make any claim between enzymatic directionality and protein expression.

2) The authors show that three mechanisms namely codon optimization, membrane localization and disulfide bond formation, determine the limit of expressions of several proteins. However, the differences observed by the authors are significant but really small. While it is likely that multiple mechanisms would contribute to determining the upper limit of protein expression, the authors need to be cautious about claiming them as the sole factors limiting expression levels (subsection “Mitochondrial localization restricting the expression limit of Adh3” and subsection “Lower expression of nonharmful glycolytic proteins explained by their codon optimality”, first paragraph). In addition, the authors claim that disulfide bonds limit the expression of Eno2 and Pgk1 by triggering aggregation. They show that addition of DTT removes the bond formation in Eno2, and changing cysteine to serine in a third protein (Tpi1) reduces its expression levels. Both these pieces of evidence are incomplete independently. Instead of performing the estimations in two different proteins, the authors need to alter cysteine to serine in Eno2 and Pgk1 and then show that the bands disappear in addition to increase in expression. Alternately, the authors need to show that Tpi1 also forms bonds which disappear upon treatment with DTT. The fact that expression of Tpi1 is independent of DTT contradicts that role of disulfide bonds in limiting expression.

3) Overall the authors play a bit fast and loose with their statistics. First, p values should be accurately reported. Don't write "p<0.05", write "p=0.032". Second, whenever p values are stated it should also be stated what test was used (see e.g. subsection “Mutations in catalytic centers not affecting expression limits of most glycolytic proteins”). Third, correlations should be reported with p values (e.g. subsection “Metabolic perturbations triggered upon overexpression of glycolytic proteins”, last paragraph).

4) While the authors describe an interesting system to estimate the limits of protein expression within a cell, there are several discrepancies between vector copy number and measured expression levels, which raises the concern the results can be reflective of the technical experimental setting instead of true limitations. In addition, the authors use GFP, an exogenous protein with a high expression, as control. Endogenous proteins, preferably unidirectional and bidirectional enzymes, that show high and low expression levels, will make for better controls to the glycolytic enzymes. The following are specific examples of such discrepancies:

A) In Figure 1B and C, why is the maximum growth rate between high and low copy number vector control so different?

B) Subsection “Measurement of expression limits of glycolytic proteins”, second paragraph: The two explanations mentioned by the authors are not mutually exclusive. The authors argue that proteins with low expression and low copy number are harmful for the cell and the ones with low expression and high copy number are repressed due to their high copy number. This is a circular argument and doesn't explain why the copy numbers are high in the first place. Finally, is the Pearson correlation of 0.3 significant? The authors have dismissed it but they need to show that it is statistically not significant.

C) The authors claim that 15 percent of the total cellular protein is the limit for overexpression of protein. However, the authors do not observe any correlation between molecular weight and protein expression levels. How do the authors explain this lack of correlation?

5) The results depend on how exactly one defines "growth defect" and how accurately one measures it. The paper does not discuss this issue. "growth defect" needs to be defined precisely, and the authors also need to argue that they can measure it with sufficient accuracy. One way by which one could get the result that there are no growth defects even at high levels of overexpression is by using a very insensitive assay.

6) A simple summary table presenting the expression limit and proposed mechanism of toxicity (or not) and the evidence for this for all 29 proteins would be very helpful. This could replace some of the information in Table 1 which could be moved to the supplement.

7) In places it is not entirely clear why the authors are only performing mechanistic experiments on a specific subset of the proteins. Again a summary table might help to better communicate what has been tested for which proteins and why.

8) 'Repression' implies an active mechanism to lower protein concentration whereas it is just that these proteins are not using optimised codons that increase translation like the other enzymes. It is better to avoid this word and simply talk about 'lower expression'.

9) The metabolic profiling is rather inconclusive. Is the conclusion simply that changes in the quantified metabolites cannot be causing the growth defect? There also doesn't seem to be much of a connection between the results of the computational simulations and the metabolic profiling, so it's not at all clear how useful the simulations are.

10) There is an obvious other source of potential growth defects that have been discussed widely in the literature but that aren't mentioned at all: Toxic effects due to protein misfolding or misinteractions. For example, Geiler-Samerotte et al. measured the effect of overexpressed, misfolded GFP on yeast growth and found an effect in proportion to the amount of misfolded protein (https://doi.org/10.1073/pnas.1017570108). Also concentration-dependent liquid demixing e.g. Bolognesi et al., 2016. Similar topics have been discussed in the literature for a long time, see e.g. this review: https://www.nature.com/articles/nrg2662. No additional work required, but discussion is needed.

Essential revisions:

1) One of the main findings of the paper is that protein expression limits are not determined by the metabolic activity. The authors base this primarily on the lack of a difference between expression levels of proteins and their catalytic mutants (proteins with mutations in their catalytic sites). The authors need to verify that metabolism is indeed altered to unequivocally comment on the link between metabolic activity and protein expression levels. Instances that highlight the need to perform metabolic measurements in the manuscript are:

A) The authors measured metabolic activity for only one protein-mutant pair (Pfk2, Figure 4), which did not show any difference in protein expression levels (Figure 3A). The authors need to measure the metabolic levels for pairs of proteins that show a significant difference in their expression levels like Pfk1 or Tdh3 to be able to comment on the link between metabolism and protein levels.

We agree the reviewer’s comment that we need a positive example of which metabolic perturbation mainly determines its expression limit. Overexpression of that protein should cause strong growth defects, and thus expression limit is low, and its CC mutation dramatically increases its expression limit up to the level of 15% of total protein (the protein burden limit). We expected to obtain this type of protein during our analysis of glycolytic proteins because such enzyme had never been identified as far as we know. However, we could not obtain it. We need to further survey whether our finding, “catalytic activity does not determine the expression limit of a metabolic enzyme”, can generally be applicable for other metabolic enzymes.

From the comparison of expression limits between wild-type proteins and catalytic center (CC) mutant proteins (Figure 3A and 3B), we currently believe that metabolic perturbations triggered by the overexpression of Pfk1 and Pfk2 (but not Tdh3) partially restrict their expression limits. We agree that difference in the expression limits between the wild-type and CC mutant is not a direct way to show whether the metabolic perturbation is a significant determinant of the expression limit, and measurement of metabolic levels would strengthen our arguments.

We first carefully re-examined our metabolite measurements of the cells overexpressing wild-type and CC mutant of Pfk2.The levels of some glycolytic metabolites were significantly lower than the those of the vector control (new Figure 5A, p < 0.05, Welch’s t-test). We further found that the level of F16bP in the cells overexpressing wild-type Pfk2 was >3-fold higher than CC mutant Pfk2 (p < 0.05, Welch’s t-test). Because F16bP is the product of Pfk catalysis and the simulation predicted the dramatic increase in the F16bP level upon overexpression of Pfks, the catalytic activity of Pfk2 might trigger this metabolic difference. However, as discussed later, it is not easy to conclude whether this minor F16bP change causes growth defects.

We then measured metabolic changes in the cells overexpressing Pfk1 and Tdh3 (and their CC mutants) in low-copy conditions where CC mutants showed increased expression limits. As shown in the new Figure 5B, C, levels of glycolytic metabolites in the cells overexpressing wild-type Pfk1 and Tdh3 were not changed more than threefold over the vector control. We did not observe reproducible higher F16bP level in the cells overexpressing wild-type Pfk1 than CC mutant Pfk1. Moreover, the metabolic changes were higher in the cells overexpressing CC mutant than wild-type Pfk1 (Figure 5B). We did not observe any reproducible difference between the metabolic changes between the cells overexpressing wild-type and those overexpressing CC mutant Tdh3 (Figure 5C). We thus concluded that overexpression of Pfk1 and Tdh3 did not trigger significant metabolic changes through their catalytic activities at least in the detected glycolytic metabolites.

We added these results in new Figure 5 and Figure 5—figure supplement 1 and thoroughly rewrote relevant descriptions in Results as follows:

“To further characterize physiological conditions triggered by the overexpression of Pfks, we next analyzed metabolic changes in yeast cells overexpressing wild-type and CC mutant Pfk2 over the vector control by measuring 35 metabolites (Figure 5–source data 1), because the CC mutants showed increased expression limits (Figure 3B). […] We thus concluded that overexpression of Pfk1 and Tdh3 did not trigger significant metabolic changes through their catalytic activities at least in the detected glycolytic metabolites.”

B) In addition, there are several contradictions in the effect of these catalytic mutants on protein expression levels. Figure 3A shows that in two cases the mutant has higher expression (Pfk1 and Tdh3) but in other two cases, the wild-type has higher expression (Fba1, Eno1). How do the authors explain such differences? In addition, Tdh3 mutant has lower expression level compared to wild-type when expressed from a high copy plasmid. How do the authors explain this flip?

Higher expression limits in CC mutants could be explained as they cause less perturbation in the metabolism, which is the reason why we constitute the mutants as described.

During the revision process, we found a mistake in the construction of the CC mutant of PGK1 (we had introduced a synonymous mutation (115A>C) which did not change the target amino acid). We thus re-constituted a correct CC mutant of PGK1 (R39A, 115A>G, 116G>C). We are very sorry for this mistake (we checked and confirmed that we constructed all others as we had intended). We measured the expression limit of the mutant and noticed it also showed significantly lower expression limit than the wild-type (Figure 3A). We also performed statistical analysis again by more strictly (Welch’s t-test instead of Student’s t-test), and noticed the difference between the wild-type and the CC mutant of Eno1 was not significant. The CC mutants of Fba1 and Pgk1 thus showed significant lower expression limits than the wild-types.

We currently do not have any substantial and consistent explanation why these CC mutants have lower expression limits. We can assume some general mechanisms; CC mutant proteins sequester the wild-type enzymes into inactive complexes; CC mutant proteins sequester substrate molecules for the wild-type enzymes; mutation in the catalytic center destabilize the structure of the enzyme. For example, Fba1 is an essential homodimeric enzyme (UniProtKB: P14540). Overexpressed of Fba1 CC mutant molecules might sequester active wild-type Fba1 molecule into inactive complexes.

The strange behavior of Tdh3 CC mutant might be related to its moonlighting function. As described above, the difference in the expression limits between wild-type and CC mutant Tdh3 seemed not to be explained its catalytic activity. Beside its metabolic function, Tdh3 directly bind to Sir2 protein to promote transcriptional silencing (Ringel AE., et al., 2013). A mutation in the catalytic center (C150G) reduces the silencing. It is thus possible that our CC mutant of Tdh3 (C150S) affect the silencing in a dose-dependent manner by competing with the wild-type Tdh3 for binding with Sir2.

We added these discussion in Discussion section as follows:

“The CC mutants of Fba1 and Pgk1 showed lower expression limits than their wild-types (Figure 3A). […] It is thus possible that the CC mutant Tdh3 (C150S) affect the silencing in a dose-dependent manner by competing with the wild-type Tdh3 for binding with Sir2.”

C) Discussion, fifth paragraph: The authors claim that one of the reasons why they don't see any association between metabolic activity and expression is that the majority of these enzymes are bidirectional. This is not true for all the enzymes as some glycolytic enzymes are unidirectional. In addition, the authors need to show a control example where unidirectional enzyme has a higher protein expression in order to make any claim between enzymatic directionality and protein expression.

Our claim is opposite. We are sorry for our confusing descriptions. We tried to claim that overexpression of a bidirectional enzyme does not strongly affect the metabolite levels because the substrate/product levels by themselves should determine the flux. Conversely, overexpression of a unidirectional enzyme should strongly affect the metabolite levels because the enzymatic activity should determine the flux. We thus think that unidirectional enzymes (Hxks, Pfks, and Cdc19) should instead have lower expression limits. The idea is based on the results of the simulation shown in Figure 4. In the simulation, overexpression of 14 bidirectional enzymes showed minor metabolic changes (Figure 4A), while all six unidirectional enzymes showed strong metabolic changes (Figures 4B–D). At least in theoretical level, above hypothesis that overexpression of a bidirectional enzyme does not strongly affect the metabolite levels seemed right. This hypothesis, however, was not supported by the experiment because both wild-type and CC mutant of Hxts had high limits and CC mutant Cdc19 did not increase its expression limits (Figure 3A). We need further analysis about this issue, but it is beyond the scope of this study.

We thus added descriptions about this issue in Discussion:

“To support this idea, overexpression of 14 bidirectional enzymes showed minor metabolic changes, while overexpression of 6 unidirectional enzymes (Hxks, Pfks, Cdc19, and Pdc1) showed strong metabolic changes in the simulation (Figure 4). […] These results suggest an additional mechanism that is not implemented into the model to avoid big metabolic changes upon overexpression of Hxks; a mechanism that prevents these metabolic perturbations from occurring, or a mechanism that prevents these metabolic perturbations from causing growth defects.”

2) The authors show that three mechanisms namely codon optimization, membrane localization and disulfide bond formation, determine the limit of expressions of several proteins. However, the differences observed by the authors are significant but really small. While it is likely that multiple mechanisms would contribute to determining the upper limit of protein expression, the authors need to be cautious about claiming them as the sole factors limiting expression levels (subsection “Mitochondrial localization restricting the expression limit of Adh3” and subsection “Lower expression of nonharmful glycolytic proteins explained by their codon optimality”, first paragraph).

We agree that multiple mechanisms would contribute to determining expression limit of a protein, and in this study, we tried to reveal them one by one using glycolytic proteins as model proteins. We are sorry that our descriptions gave an impression that we are claiming our findings as the sole factors limiting expression levels.

We thus added some arguments about this in Discussion section as follows:

“As described above, we revealed mechanisms restricting the expression limits of some glycolytic proteins. We, however, do not think that these mechanisms are the sole factors restricting expression limits of these proteins. The expression limits of ΔMTS-Adh3 (0.45 AU, Figure 3D) and CoGlkl (1.07 AU, Figure 6A) are still lower than the other high limit proteins such as Pgk1 and Gpm1 (2.26 AU and 2.63 AU, Figure 2B). It is thus likely that multiple mechanisms would restrict the expression limits of these proteins.”

In addition, the authors claim that disulfide bonds limit the expression of Eno2 and Pgk1 by triggering aggregation. They show that addition of DTT removes the bond formation in Eno2, and changing cysteine to serine in a third protein (Tpi1) reduces its expression levels. Both these pieces of evidence are incomplete independently. Instead of performing the estimations in two different proteins, the authors need to alter cysteine to serine in Eno2 and Pgk1 and then show that the bands disappear in addition to increase in expression. Alternately, the authors need to show that Tpi1 also forms bonds which disappear upon treatment with DTT. The fact that expression of Tpi1 is independent of DTT contradicts that role of disulfide bonds in limiting expression.

As described above, we found a mistake in the construction of the CC mutant of Pgk1. We thus withdrew the identification of proteins within the aggregate observed when Pgk1 CC mutant was overexpressed from Figure 6D (now Figure 7D) as it was not the mutant protein. We are sorry for our mistake, but our finding is not primarily affected by this.

We constructed mutants of Eno2 (C248S) and Pgk1 (C98S) whose cysteine was changed into serine and analyzed their effects. We confirmed that the aggregation bands were disappeared as shown in Figure 7—figure supplement 1. We, however, did not observe any increase in the expression limits of Eno2 and Pgk1 even when their cysteines were removed. We thus concluded that S-S bond-connected proteins aggregation does not restrict the expression limits of Eno2 and Pgk1. We are not surprised by this result because only small part of overexpressed Eno2 and Pgk1 constitute the aggregation bands (the bands are only observed with a long-exposure), and expression limits of wild-type Eno2 and Pgk1 are the highest levels among other glycolytic proteins (Figure 2B); hence their expression limits seems not restricted by other mechanisms than the protein burden effect. We thus concluded that deleterious effect of this aggregation might be protein-specific.

Detection of the S-S bond connected protein aggregation was just a hint to think the reason why the expression limits of Tpi1 is far lower than Gpm1 who does not contain any cysteine (Figure 2B). Tpi1 was detected in both Eno2 and Pgk1 overexpression-triggered aggregates although Tpi1 itself was not overexpressed in that situation, suggesting that Tpi1 is a naturally-aggregative protein. As the reviewer has suggested, we tried to detect aggregation band upon overexpression of Tpi1, and if the band is disappeared when cysteines are substituted.

We did not identify any aggregation band upon overexpression of Tpi1 by the total protein staining (new Figure 8A). We want to emphasize that our critical finding in our identification of the proteins in the aggregates is that the aggregation bands contain many different proteins in addition to the overexpressed proteins themselves. Hence, the aggregation bands visible in the gel are constructed by chance, and other invisible aggregation bands with different sizes could exist in the gel. If Tpi1 constitutes aggregation with many different proteins, it is no wonder even if no visible aggregation band is observed. We thus performed Western blotting using Tpi1-specific antibodies to detect invisible aggregation bands upon total protein staining, and confirmed the existence of many aggregation bands that were disappeared by the removal of cysteines from Tpi1 or the DTT treatment (Figure 8A).

We added these results in Figure 7—figure supplement 1 and 2, and Figure 8 and added relevant descriptions in Results and Discussion as follows:

In Results:

“We confirmed that cysteines were responsible for creating these bands because they were disappeared when cysteine residues were removed from Pgk1 and Eno2 (Figure 7—figure supplementary 1).”

“This aggregation seemed not to affect their expression limits because expression limits of wild-types and cysteine-less mutants (Eno2-C248S and Pgk1-C98S) were indistinguishable (Figure 7—figure supplementary 2). […] To test whether the aggregation restricts the Tpi1 expression limit, we measured expression limits of cysteine-less Tpi1. As shown in Figure 8B, the expression levels of cysteine-less Tpi1 significantly increased.”

In Discussion:

“Overexpression of Eno2, Pgk1, and Tpi1 triggered S–S-bond-connected aggregation (Figure 7 and 8).”

“The deleterious effect of this aggregation, however, seems protein-specific because expression limits of Pgk1 and Eno1 were among highest (Figure 2A), and removal of their cysteine did not increase expression limits of them (Figure 7—figure supplement 1).”

3) Overall the authors play a bit fast and loose with their statistics. First, p values should be accurately reported. Don't write "p<0.05", write "p=0.032". Second, whenever p values are stated it should also be stated what test was used (see e.g. subsection “Mutations in catalytic centers not affecting expression limits of most glycolytic proteins”). Third, correlations should be reported with p values (e.g. subsection “Metabolic perturbations triggered upon overexpression of glycolytic proteins”, last paragraph).

We added statistics as the reviewer suggested. We initially used Student’s t-test to calculate p-value for most of our measurements, but re-calculated p-values using Welch’s t-test for more strict significance evaluations. All p-values are listed in each figure source data.

4) While the authors describe an interesting system to estimate the limits of protein expression within a cell, there are several discrepancies between vector copy number and measured expression levels, which raises the concern the results can be reflective of the technical experimental setting instead of true limitations.

We answered to this comment in the following specific examples.

In addition, the authors use GFP, an exogenous protein with a high expression, as control. Endogenous proteins, preferably unidirectional and bidirectional enzymes, that show high and low expression levels, will make for better controls to the glycolytic enzymes.

As we mentioned in Introduction, we previously found that GFP can be overexpressed up to the protein burden level which is 15% of the total cellular protein (Kintaka et al., 2016). As far as we recognize, there was no report that any endogenous or exogenous protein is expressed up to this level in S. cerevisiae. In other words, we did not know any endogenous protein whose expression limit is this high, and it was impossible to use an endogenous protein as a control in this study. Therefore, we can instead consider this study as the first identification of endogenous unidirectional and bidirectional enzymes that show high and low expression levels that can be used for future research.

The following are specific examples of such discrepancies:

A) In Figure 1B and C, why is the maximum growth rate between high and low copy number vector control so different?

We are sorry that our experimental system used here is a bit unusual and difficult to understand. The detailed explanation of this method is omitted because it is previously published (Moriya et al., 2006 and 2012). In this experiment, we used pTOW40836 who carries URA3 and leu2-89 (LEU2 with a truncated promoter) as the selection markers. Level of the Ura3 protein expressed from the plasmid is sufficient to fully recover the growth of the host strain (ura3 and leu2 deletion) in –uracil (+leucine) conditions even if the plasmid copy number is low because URA3 has a full strength promoter. In contrast, levels of the Leu2 protein is insufficient to recover the growth in –leucine conditions if the plasmid copy number is low because leu2-89 has a large deletion in the promoter. This becomes a bias to increase the plasmid copy number up to 150 copies per cell in –leucine conditions (to be more precise, a bias to select cells with higher plasmid copies). However, even in this condition, the growth rate is not fully recovered compared to +leucine conditions probably because the level of Leu2 protein is not sufficient enough.

We added a description below in Results:

“In this experimental system, maximum growth rates of the cells with the vector in +leucine conditions is much higher than those in –leucine conditions (see Figures 1B–C), probably because the copy number of leu2-89 is not sufficient to fully support the leucine requirement in –leucine conditions.”

B) Subsection “Measurement of expression limits of glycolytic proteins”, second paragraph: The two explanations mentioned by the authors are not mutually exclusive. The authors argue that proteins with low expression and low copy number are harmful for the cell and the ones with low expression and high copy number are repressed due to their high copy number. This is a circular argument and doesn't explain why the copy numbers are high in the first place.

We agree those two explanations are not mutually exclusive. We thus remove ‘independent’ from the sentence. For the latter part, we are sorry for our confusing descriptions; we used ‘because’ not for the mechanistic explanation but the evidence. To avoid confusion, we added exact citations to the data points on the graph in the sentences as follows:

“There could be two reasons that the expression level of a protein is low: (i) its strong overexpression is harmful to cellular growth and (ii) its expression is repressed. […] In contrast, the expression of Glk1, Pyk2, and Pdc6 seemed to be repressed because their copy numbers were higher than the others’ (blue circles in Figure 2D).”

Finally, is the Pearson correlation of 0.3 significant? The authors have dismissed it but they need to show that it is statistically not significant.

We added p value (0.11) that was not significant.

C) The authors claim that 15 percent of the total cellular protein is the limit for overexpression of protein. However, the authors do not observe any correlation between molecular weight and protein expression levels. How do the authors explain this lack of correlation?

In this study, we calculate the unit for the expression level of a protein (AU) from the total intensity of fluorescently-stained protein band, which reflects the total amino acid number within the band (not the molecule number of the protein). Therefore, if two proteins give the same expression AUs, the molecule number in the band of the larger protein is lower as the reviewer claims. For example, Gpm1 and Pgk1 had their expression levels 2.63 AU and 2.27 AU. When those units are normalized with their lengths (248 a.a. and 415 a.a.), their expression levels are 0.010 AU/a.a and 0.005 AU/a.a., so the larger protein expresses lower in the molecule number.

We can think the reason why we did not see any correlation between the expression levels (AU) and lengths is that the cost for protein production but not the number of the protein molecule determines the expression limit. Fifteen percent of the total cellular protein thus means ‘15% of total amino acid incorporated into cellular protein’, but not ‘15% molecules of total protein molecules’. We also want to emphasize that a protein can be expressed up to 15% of total protein only if overexpression of the protein does not cause additional harmful effect than the protein burden. So the expression limit of a harmful protein should be determined independently by the size.

We added these explanations in Results and Discussion.

In Results:

“The AU is considered to reflect the total amino acid number within the band, and the relative number of the protein molecule can be estimated by dividing AU by the protein length. When two proteins with different sizes give the same AUs, the molecule number of the larger protein in the band should be lower.”

In Discussion:

“Among the glycolytic proteins studied here, Pgk1, Gpm1, and Eno2 gave highest expression limits. […] It also suggests that their expression limits are not determined by the molar concentration of the proteins but by the cost of the protein production.”

5) The results depend on how exactly one defines "growth defect" and how accurately one measures it. The paper does not discuss this issue. "growth defect" needs to be defined precisely, and the authors also need to argue that they can measure it with sufficient accuracy. One way by which one could get the result that there are no growth defects even at high levels of overexpression is by using a very insensitive assay.

We define growth defect based on the significance in the difference of maximum growth rates (MGRs) between the cells overexpressing the target protein and the cells with the control vector. We also use the copy numbers of plasmids expressing target proteins as indicators for growth defects upon overexpression of target proteins. It is not easy to argue how accurate our measurements are, because we did not compare our measurements with other more sensitive measurements like competitive growth assay. However, these measurements are sufficiently reproducible to argue growth defects with statistical significance. We observed significant reductions of MGRs of the cells overexpressing most of the glycolytic proteins (p < 0.01, Welch’s t-test, Figure 1B). Plasmid copy numbers seemed more sensitive indicators for growth defects because all of them overexpressing glycolytic proteins but Pyk2 were less than half of that of the vector control with larger statistical significance (p < 0.001, Welch’s t-test, Figure 1D). We thus think our experimental system is sufficiently sensitive to discuss growth defects triggered by the overexpression of glycolytic proteins.

We added the explanation for our definition of the growth defect in Materials and methods as follows:

“The maximum 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 biological triplicates. We define growth defect based on the significance in the reduction of maximum growth rate of the cells overexpressing a target protein compared with that with the control vector (p < 0.01, Welch’s t-test).”

6) A simple summary table presenting the expression limit and proposed mechanism of toxicity (or not) and the evidence for this for all 29 proteins would be very helpful. This could replace some of the information in Table 1 which could be moved to the supplement.

We are thankful for the reviewer’s suggestion. We made a summary table (Supplementary file 1).

7) In places it is not entirely clear why the authors are only performing mechanistic experiments on a specific subset of the proteins. Again a summary table might help to better communicate what has been tested for which proteins and why.

We made a summary table (Supplementary file 1).

8) 'Repression' implies an active mechanism to lower protein concentration whereas it is just that these proteins are not using optimised codons that increase translation like the other enzymes. It is better to avoid this word and simply talk about 'lower expression'.

In the earlier part of Results, we used ‘repress’ to imply an active mechanism to lower protein concentration because we did not know the mechanism to lower the expression of Glk1, Pyk2, and Pdc6. The mechanism was later turned out to be low codon-optimality. Although we think that lowering codon optimality (deoptimizing) could be an evolutionally active mechanism to lower protein concentration, we agree this word is confusing. We thus changed the description in Discussion as follows:

“The translational rate of some glycolytic proteins, including Glk1, seemed low due to their lower codon optimality (Figure 6).”

9) The metabolic profiling is rather inconclusive. Is the conclusion simply that changes in the quantified metabolites cannot be causing the growth defect? There also doesn't seem to be much of a connection between the results of the computational simulations and the metabolic profiling, so it's not at all clear how useful the simulations are.

We agree that the metabolic profiling is inconclusive. A big issue we have recognized through the metabolic analysis was we hardly know how much change of which metabolite triggers growth defect. For example, does a threefold change in the F16bP level cause growth defect? As far as we know, this issue has never been assessed. Moreover, we also have recognized that we currently do not have any systematic way to identify metabolic changes directly triggered by the overexpression of an enzyme because metabolism is interconnected and overexpression of a protein could cause non-specific perturbations that ultimately affect metabolism. To answer this issue, we have to wait until a large-scale survey to connect metabolic changes and growth defects, preferably with systematic perturbations such as deletion and overexpression is performed. As described above, we observed a threefold difference between wild-type and CC mutant Pfk2 (Figure 5A). However, we cannot conclude this causes the difference in their expression limits with our current knowledge. We thus do not want to simply conclude that changes in the quantified metabolites cannot be causing the growth defects.

Above reasons are why we used a mathematical model as a reference. Using a mathematical model, we can predict potential metabolic changes triggered by overexpression of an enzyme without considering other unknown effects than its metabolic activity. As in the case of Pfks and Hxks overexpression in the simulation (Figures 4B, D), we can predict the pattern of metabolic changes. We again cannot define how much changes in these metabolites could trigger growth defects in the simulation (and in vivo). However, if the metabolic changes are divergent and almost catastrophic as observed in the Pfks and Hxks simulations (~1,000 fold increase in some metabolites, Figures 4B, D), we could claim metabolic changes triggered by the overexpression of these enzymes should cause growth defects. Hence, we expected to obtain similar metabolic changes upon overexpression of Pfks whose catalytic activities seemed to trigger growth defects. However, we did not observe such a big changes, and the pattern was not consistent (Figures 5, Figure 5—figure supplement 1). We also agree that the usefulness of the simulations is unclear with these results, but we at least could show one approach to attack above uncertain biological issue.

We agree that this is a critical argument need to be included in Discussion and described as follows:

“Through the metabolic analysis, we realized that we currently do not have any systematic way to identify metabolic changes directly triggered by the overexpression of an enzyme, because metabolism is interconnected and overexpression of a protein could cause non-specific perturbations that ultimately affect metabolism. […] To precisely answer these issues, we need a much deeper understanding of the connection between metabolite levels and the cellular growth.”

10) There is an obvious other source of potential growth defects that have been discussed widely in the literature but that aren't mentioned at all: Toxic effects due to protein misfolding or misinteractions. For example, Geiler-Samerotte et al. measured the effect of overexpressed, misfolded GFP on yeast growth and found an effect in proportion to the amount of misfolded protein (https://doi.org/10.1073/pnas.1017570108). Also concentration-dependent liquid demixing e.g. Bolognesi et al., 2016. Similar topics have been discussed in the literature for a long time, see e.g. this review: https://www.nature.com/articles/nrg2662. No additional work required, but discussion is needed.

We thank the reviewer for noticing us a critical argument. We had intended to include these mechanisms as promiscuous interaction in Introduction, but we recognized that it was not sufficient and needed to discuss them more.

We thus added the following descriptions in Discussion:

“Protein misfolding or misinteraction is considered to cause toxicity upon high-level expression of a protein with low translational robustness, low folding stability, or a high propensity for misinteraction (Drummond and Wilke, 2009; Zhang and Yang, 2015). […] We do not think this mechanism caused growth defects upon overexpression of glycolytic proteins studied here because they are less structurally-disordered (Moriya, 2015) and not nucleic-acid-binding proteins.”

Author details

1. Yuichi Eguchi

   Graduate School of Environmental and Life Science, Okayama University, Okayama, Japan

   Contribution

   Data curation, Investigation, Writing—original draft, Writing—review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-4809-1402

2. Koji Makanae

   Research Core for Interdisciplinary Sciences, Okayama University, Okayama, Japan

   Contribution

   Data curation, Writing—review and editing

   Competing interests

   No competing interests declared

3. Tomohisa Hasunuma

   Graduate School of Science, Technology and Innovation, Kobe University, Kobe, Japan

   Contribution

   Data curation, Validation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

4. Yuko Ishibashi

   Department of Life Sciences, School of Agriculture, Meiji University, Kawasaki, Japan

   Contribution

   Data curation, Methodology

   Competing interests

   No competing interests declared

5. Keiji Kito

   Department of Life Sciences, School of Agriculture, Meiji University, Kawasaki, Japan

   Contribution

   Data curation, Validation, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

6. Hisao Moriya

   1. Graduate School of Environmental and Life Science, Okayama University, Okayama, Japan
   2. Research Core for Interdisciplinary Sciences, Okayama University, Okayama, Japan

   Contribution

   Conceptualization, Data curation, Supervision, Funding acquisition, Investigation, Writing—original draft, Writing—review and editing

   For correspondence

   hisaom@cc.okayama-u.ac.jp

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7638-3640

• 5,892

  Page views

• 690

  Downloads

• 33

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.