Toxicity of the model protein 3×GFP arises from degradation overload, not from aggregate formation

Most proteins need to fold into the correct three-dimensional structures to perform their functions properly (Dobson et al., 1998). However, mutations, translation errors, high temperatures and oxidative stress can affect the protein folding process and induce misfolding. Misfolded proteins are not only useless but sometimes form cytotoxic aggregates (Dill and MacCallum, 2012; Lindquist and Kelly, 2011). These aggregates are associated with cellular dysfunction caused by high temperature and oxidative stress, as well as aging and neurodegenerative diseases (Aguzzi and Lakkaraju, 2016; Sweeney et al., 2017). Thus, there are intracellular protein quality control systems to maintain proteostasis, prevent protein misfolding and to deal with aggregates (Hartl et al., 2011; Sinnige et al., 2020).

Aggregates have long been considered cytotoxic, but it is now clear that aggregates are not immediately cytotoxic, but can be toxic in a variety of different ways depending on their size, structure and composition (Iadanza et al., 2018; Bucciantini et al., 2002). In recent years, it is also believed that aggregates are actively formed as a quality control mechanism to isolate cytotoxic misfolded proteins (Rothe et al., 2018; Kaganovich et al., 2008). The nature of protein aggregates has been investigated for disease-related proteins, particularly using budding yeast as a model eukaryotic cell. Polyglutamine (PolyQ), which is linked to Huntington's disease, forms amorphous/mesh-like aggregates or amyloid fibrils, leading to endoplasmic reticulum (ER) and heat-stress responses (Klaips et al., 2020). Similarly, intracellular aggregates of α-synuclein, which is associated with Parkinson's disease, disrupt transport between the ER and Golgi (Cooper et al., 2006; Lázaro et al., 2014). Studies of VHL, UBC9-TS and TDP-43 have shown that toxicity is caused by highly reactive misfolded proteins and very small aggregates (Kaganovich et al., 2008; Bolognesi et al., 2019). Interestingly, aggregates of polyQ protein, which normally have an amyloid-like structure, become a mesh structure capable of dynamically trapping Hsp70 family proteins (Ssa1 and Ssa2 in yeast; hereafter denoted Hsp70) upon overexpression of the cochaperone Ydj1 (Klaips et al., 2020). In this state, a productive heat-shock response (HSR) can be triggered. Thus, the aggregates themselves can be cytoprotective. Furthermore, cells have mechanisms to sequester toxic misfolded proteins into intracellular compartments called JUNQ and IPOD, which are observed as large aggregates (Miller et al., 2015; Hill et al., 2016). In contrast to primary quality control, such as avoidance of protein misfolding and rapid degradation of misfolded proteins, the mechanism to isolate such misfolded proteins is called spatial protein quality control (Hill et al., 2017; Rothe et al., 2018). Thus, we do not fully understand how aggregates affect cellular physiology.

One factor complicating the study of protein aggregates is the difficulty in separating the physiological effects of aggregation from other effects (Schneider et al., 2018). For example, overexpression of proteins, a method often used in aggregation studies, has effects other than aggregation (Moriya, 2015). Overexpression can enhance intrinsic protein properties, such as enzymatic activity, or cause the sequestration of essential proteins through innate interactions with endogenous proteins. To isolate such effects from those due to aggregation, non-aggregating proteins that are structurally and functionally similar might be used as control proteins. However, creating a control protein with appropriate properties is not easy, and there are few aggregation model proteins with controls of the same molecular weight and structure.

Since its cloning and expression, green fluorescent protein (GFP) has been designed to minimize interactions with endogenous proteins in the cell (Prasher et al., 1992; Pédelacq et al., 2006; Costantini et al., 2015). As a result, it is considered there is almost no specific interaction with any cellular protein and no physiological function, and it is one of the non-harmful proteins that cells can express in large amounts (Eguchi et al., 2018). In addition, multiple GFP variants of the same structural and molecular mass exist (Lambert, 2019; Rodriguez et al., 2017). We hypothesized that these features of GFP might allow us to separate the physiological effects of aggregation from other effects.

We recently found that overproduction (op) of EGFP, a widely used green fluorescent protein in yeast (Saccharomyces cerevisiae) cells promotes an HSR, presumably by misfolding, and the formation of aggregates containing Hsp70 (Namba et al., 2022). By contrast, moxGFP (MOX), which has improved folding properties and lacks cysteine, did not produce such aggregates. Furthermore, the overproduction of three EGFP-linked proteins (3×GFP-op) causes stronger growth inhibition (i.e. they are more toxic) than EGFP-op (Kintaka et al., 2020). In 10–20% of 3×GFP-op cells, one giant aggregate, of ∼5 μm in size, with GFP fluorescence is produced. In the insoluble fraction of 3×GFP-op cells, the molecular chaperone Hsp70 (Ssa1 and Ssa2) and the glycolytic enzymes Fba1 and Eno2 are enriched, suggesting that the aggregates might be toxic because they sequester these proteins. In addition, ubiquitylated 3×GFP accumulates in the same insoluble fraction, and 3×GFP-op has a negative genetic interaction with proteasome mutants. This also suggests that the overload on degradation caused by misfolded 3×GFP is responsible for growth inhibition. Considering the above, the 3×GFP-op cells were analyzed in more detail in this study, as they are a good model case to investigate the cytotoxicity induced by protein misfolding and aggregates.

In this study, 3×MOX-op was used as a control because the molecular mass and structure of 3×GFP and 3×MOX are close, which should allow us to better discriminate between the effects of aggregation and other effects, such as the effect of high expression of 75 kDa proteins itself. We investigated the conditions for the generation of giant aggregates of 3×GFP, the components they contain, the formation process and the mechanism by which they exert cytotoxicity while using 3×MOX as a control. We show that 3×GFP-op creates large aggregates that trap Hsp70 but are not themselves the cause of toxicity, but rather that they cause an overload of degradation by the ubiquitin-proteasome system. We also show that the entrapment of essential proteins into aggregates can be a potential mechanism of toxicity.

To separate the effects of high-level protein expression from those caused by aggregation, we first generated 3×MOX, a control protein with a similar structure to 3×GFP but without aggregation potential. We overproduced 3×GFP and 3×MOX under the constitutive PYK1 promoter (PYK1_pro) or used the inducible WTC₈₄₆ system (Azizoglu et al., 2021) on the pTOW40836 plasmid (Makanae et al., 2013) (Fig. 1A). This plasmid is a multicopy plasmid whose copy number increases in cells cultured in synthetic complete medium without leucine and uracil (SC–LU medium) to give up to >100 copies per cell (Moriya et al., 2006; Makanae et al., 2013). We confirmed the development of single large aggregates in the 3×GFP-op cells as previously reported (Kintaka et al., 2020), whereas there was no aggregate formation in 3×MOX-op cells (Fig. 1B). Conditions triggering protein misfolding, such as 5-azacytidine (AZC; mistranslation) and hydrogen peroxide (H₂O₂; oxidative stress) treatment, enhanced the aggregation upon 3×GFP-op, but not 3×MOX-op (Fig. 1C–E). In the aggregate, it is likely that 3×GFP is not completely misfolded, but rather partially misfolded, because GFP fluorescence is still observed even when aggregated (Fig. 1B). These results suggest that the aggregates are formed with partially misfolded GFP, and not by the clustering of folded GFPs, and 3×MOX can be used as a control non-aggregative protein for 3×GFP.

We next compared the cytotoxicity of 3×GFP and 3×MOX. We also analyzed monomeric MOX to assess the increased cytotoxicity resulting from the linking of three MOX proteins. We first measured the maximum growth rate upon a stepwise increase in the expression using the WTC₈₄₆ system (Fig. 2A). We note that even without the inducer anhydrous tetracycline (aTc), 3×GFP-op and 3×MOX-op cells showed a reduced growth rate compared to that for cells with the vector control under the high-copy conditions (SC–LU, in Fig. 2A). This suggests that a ‘leaky’ expression of these proteins already causes some cytotoxicity. Upon induction up to 150 nM aTc, all three proteins caused gradual decreases in the growth rates. With the aTc concentration above 100 nM, 3×GFP-op cells almost completely halted growth, whereas 3×MOX-op and MOX-op cells maintained growth. In addition, the growth of MOX-op cells was always better than that of 3×MOX-op cells under all conditions, suggesting that cytotoxicity was higher for 3×GFP, 3×MOX and MOX, in that order.

To further investigate the toxic effect of 3×GFP-op, we next measured the number of dead cells within the cell population as shown in Fig. 2B. Specifically, we counted the number of dead cells for 1×MOX, 3×MOX and 3×GFP before induction, and 24 and 48 h after induction at different aTc concentrations. Before induction, ∼20% dead cells were observed in most samples. After 24 h of induction, a significant increase in cell death was observed specifically in 3×GFP-op compared to the vector control. After 48 h, the proportion of dead cells increased overall, but only 3×GFP-op showed a significantly higher percentage of dead cells compared to the vector. Note that the number of dead cells significantly decreased in 1×MOX-op as the concentration of aTc increased, compared to the vector. This might be an effect of the high expression of a less toxic protein. Therefore, it was confirmed that 3×GFP-op reduces cell viability. This reduction in viability was also verified by a spot assay (Fig. S1).

We next assessed the maximum protein levels at which the cells could maintain growth (expression limit). We analyzed the protein levels of 3×GFP at 50 nM aTc, 3×MOX at 50 nM, 250 nM and 500 nM aTc (Fig. 2C). These aTc concentrations were determined by the maximum viable concentration for 3×GFP and the maximum induced dose for 1×MOX and 3×MOX (Fig. 2A). We then measured the expression limits as a percentage of the 1×MOX limit (expressed under TDH3pro which is equivalent to the maximum induction of the WTC₈₄₆ system) (Fig. S2). The maximum measurable expression limit of 3×GFP (at 50 nM aTc; 9 unit) was about one-third of that of 3×MOX (500 nM aTc; 38 units), suggesting that 3×GFP has stronger cytotoxicity than 3×MOX. We note that the linking of three MOX itself increased cytotoxicity because the expression limit of 3×MOX is lower than that of 1×MOX (Fig. 2C).

We previously found that 3×GFP-op had negative genetic interactions with proteasome mutants and high-molecular-mass ubiquitylated proteins accumulated in the insoluble fraction upon 3×GFP-op (Kintaka et al., 2020). Those results suggest that 3×GFP-op perturbs the ubiquitin-proteasome system. Here, we further analyzed the genetic interactions with 3×MOX as a control. As shown in Fig. 3A, the growth rate of temperature-sensitive mutant strains for three proteasome genes (pre6-5001, pup1-1 and rpn12-1) showed significantly slower growth under 3×GFP-op than with the vector control or under 3×MOX-op. These cells showed a lower percentage of cells with aggregation (Fig. S3), probably because the deleterious effects of 3×GFP-op are increased. We also tried to confirm the ubiquitylation of isolated 3×GFP aggregates (Fig. 3B). As shown in Fig. 3C, aggregates were purified from 3×GFP-op cells using nickel (Ni) carriers; 3×GFP aggregates were shown to be bound to beads. For 3×MOX, small clumps were also observed to bind to the beads. This might occur during the crushing or enrichment of the experimental process. Then, we detected purified proteins with anti-GFP and anti-ubiquitin antibodies (Fig. 3D). High molecular mass proteins were detected by anti-GFP and anti-ubiquitin antibodies in purified insoluble fractions of 3×GFP-op, but not 3×MOX-op cells. These results confirm that 3×GFP-op perturbs the ubiquitin-proteasome system, probably through the massive degradation of the 3×GFP aggregates.

We previously found that Hsp70 (Ssa1 and Ssa2) and the glycolytic enzymes Eno2 and Fba1 were enriched in the insoluble fraction of 3×GFP-op cells using liquid chromatography tandem mass spectrometry (LC-MS/MS) (Kintaka et al., 2020). To confirm whether these three proteins were included in the aggregation of 3×GFP, here we used fluorescence microscopy. Among these three proteins with a red fluorescent protein (mScarlet-I; mSca) bound to the C-terminus, colocalization of Ssa1–mSca with 3×GFP aggregates was confirmed (Fig. 4A, Fig. S4). By contrast, Eno2–mSca and Fba1–mSca did not colocalize with 3×GFP aggregates. As Eno2 and Fba1 are known to fractionate into insoluble fractions under conditions that cause proteotoxicity (Geiler-Samerotte et al., 2011; Weids et al., 2016), these proteins might become insoluble due to the proteotoxic stress produced by 3×GFP-op.

Klaips et al. have shown that the kinetics of Hsp70 in aggregates distinguishes the aggregate properties (Klaips et al., 2020). We thus analyzed the dynamic properties of Ssa1–mSca colocalized with 3×GFP aggregates by fluorescence recovery after photobleaching (FRAP) analysis (Fig. 4B). The fluorescence recovery of 3×GFP was only ∼10% even after 60 s of bleaching, whereas the fluorescence of Ssa1–mSca recovered immediately after bleaching, reaching 50% in 40 s (Fig. 4C). This result suggests that 3×GFP molecules do not form dense aggregates, but rather create a mesh-like structure in which accumulated Hsp70 can dynamically move.

We next investigated the process by which 3×GFP and Hsp70 grow into one large aggregate by time-lapse observation after induction of 3×GFP expression (Fig. 5A; Movie 1). The 3×GFP was observed as multiple small bright spots at 1 h after induction. The dots were initially colocalized with Hsp70, but eventually, Hsp70 surrounded the exterior (after 3 h), and they clustered into a small number of dots (after 6 h). After a long time, Hsp70 and 3×GFP were mixed into one large aggregate (15 h). We also showed, by quantification of non-sequential time-series observations, that aggregates are formed upon induction of 3×GFP expression (Fig. 5B). As the aggregates clustered and became huge, the expression of Hsp70 also increased (Fig. 5C). By contrast, expressed 3×MOX did not induce aggregates (Fig. S5B).

Combined with the results of the FRAP analysis above, a possible mechanism for the aggregation process is that the initial 3×GFP aggregates are gradually dissolved by the surrounding Hsp70 and fuse into a large mesh-like aggregate containing dynamic Hsp70 (Fig. 5D). This progression is thought to be similar to the formation process of mesh-like aggregates of polyQ protein with dynamic Hsp70 (Klaips et al., 2020). There, polyQ protein aggregate triggers a productive HSR by trapping Hsp70. Indeed, 6 h after induction of 3×GFP expression, Ssa1–mSca fluorescence increased (Fig. 5C), suggesting that an HSR was triggered. Induction of HSR upon 3×GFP-op (but not 3×MOX) was confirmed by the RNAseq analysis (Fig. S6, Table S1). Thus, the large aggregation upon 3×GFP-op itself seemed not toxic but protective to the cell.

To further investigate whether 3×GFP aggregation itself is protective, we investigated the relationship between 3×GFP aggregation and growth inhibition in 15 gene deletion strains involved in the secondary quality control (Hill et al., 2017) (Fig. 6A). We first observed 3×GFP aggregation in each deletion strain (Fig. 6B; Fig. S7A). The deletion strains showed both increases and decreases in the numbers of cells with aggregates upon 3×GFP-op compared with the wild type, whereas no aggregation was observed in any of the deletion strains upon 3×MOX-op (Fig. S7B). We next measured the growth rate of the deletion strains upon 3×GFP-op, 3×MOX-op and with the vector control (Fig. 6C). Again, the deletion strains showed an increase or decrease in the growth rates compared with the wild type. No significant growth reduction was observed in the ssa1Δ and ssa2Δ strains compared to the wild type (P-value 0.99 for ssa1Δ and 0.10 for ssa2Δ; Dunnett's test), suggesting that the trapping of Hsp70 (Ssa1 and/or Ssa2) by the 3×GFP aggregates did not cause growth inhibition through Hsp70 depletion. We note that 3×MOX-op mitigates the growth defects of some deletion strains with the vector control. This could be a potential deleterious effect of the experimental system using a high-copy vector (Kintaka et al., 2020).

Next, we examined how an increase or decrease in aggregation affects growth (Fig. 6D,E). The growth rates and percentage of cells with aggregation in the 3×GFP-op dataset, when the vector was used as a control, showed a weak positive correlation (Pearson's r=0.33). The correlation was higher (r=0.63) when ydj1Δ, which had significantly poorer proliferation, was excluded. Furthermore, the correlation was even higher when 3×MOX-op was used as a control (r=0.79). These results mean that the more aggregates formed, the more growth inhibition was mitigated, suggesting that aggregation works in the direction of mitigating cytotoxicity.

We further investigated the relationship between aggregate size and growth rate in each mutant. Our analysis using average aggregate sizes showed a positive correlation between growth rate and aggregate size (r=0.69) (Fig. S7C). We note that due to significant variation in aggregate size within the same strain, no statistically significant differences were found between strains. Fig. 6F and G present the cellular images and quantitative analysis of aggregate sizes in specific deletion strains, namely rkr1Δ and cur1Δ, which showed notably larger aggregates compared to the wild type, and hsp82Δ, which exhibited smaller aggregates. Upon 3×GFP-op, the growth of rkr1Δ and cur1Δ mutants was faster than that of the wild type, whereas growth was slower in the hsp82Δ mutant (Fig. 6C). These results further support the possibility that the increase in aggregate size is mitigating the toxicity of 3×GFP aggregates.

The above results suggest that the cytotoxicity caused by 3×GFP aggregates is not due to sequestration of essential proteins. However, there are known cases in which specific aggregates show cytotoxicity due to the sequestration of essential proteins (Treusch and Lindquist, 2012; Park et al., 2013). Therefore, we tested whether these aggregates have the potential to exhibit cytotoxicity when they trap essential proteins.

We accidentally found that monomeric EGFP is incorporated into aggregates when non-fluorescent 3×GFP is overproduced (3×GFP–Y66G) (Fig. 7A). We thus consider that co-expression of 3×GFP with GFP-tagged proteins can entrap GFP-tagged proteins into aggregates and deplete them (Fig. 7B). To test this idea, 3×GFP was overproduced in strains expressing the GFP-linked Cdc20 and Cdc28, which are essential cell cycle regulators. As shown in Fig. 7C,D, cells were highly enlarged upon 3×GFP-op (and not 3×MOX-op) in Cdc20–GFP and Cdc28–GFP strains, as is also seen for temperature-sensitive mutants of cdc20 and cdc28 at the restrictive temperature (Tavormina and Burke, 1998; Iida and Yahara, 1984). These results indicate that trapping essential proteins into 3×GFP aggregates potentially causes cytotoxicity.

This study investigated how aggregation can be toxic or protective using 3×GFP as a model. In particular, we used 3×MOX, which is structurally similar but does not aggregate, as a control (Fig. 1). This allowed us to separate the effects caused by protein overproduced from those caused by aggregates (Fig. 2). Summarizing the results of the study, the mechanism of the large 3×GFP aggregation and exertion of toxicity can be explained as follows (Fig. 8). 3×GFP-op induces aggregation, probably by partial misfolding during translation (Fig. 1). The misfolded 3×GFP is ubiquitylated and degraded by the proteasome (Fig. 3). This degradation is so active that it overloads the proteasome system, which causes growth defects. During the formation process of the giant aggregates, smaller aggregates are surrounded and fused through the action of the chaperone Hsp70, eventually transforming into a giant static mesh structure (Fig. 5). In addition, during the process of aggregate enlargement, Hsp70 is trapped inside, triggering a protective HSR. This HSR constitutes a positive feedback loop that induces the expression of Hsp70 and leads to further enlargement of the aggregates. Although Hsp70 is trapped in the aggregates, it does not cause growth inhibition due to Hsp70 depletion, either because it is dynamic (Fig. 4) or because the amount is sufficient (Fig. 6). Therefore, the aggregates formed by 3×GFP-op themselves do not cause growth inhibition because they do not deplete essential proteins. By contrast, it might be viewed as a necessary structure to capture Hsp70 to induce a productive HSR, as proposed by Klaips et al. (Klaips et al., 2020). The positive correlation between the rate of aggregation and the decrease in growth inhibition also suggests that aggregation is cytoprotective. The formation of aggregates that are inaccessible to the proteasome might also play a role in reducing cytotoxicity by preventing the overloading of the proteasome.

The process of transformation of small aggregates into large dynamic Hsp70-capturing aggregates is considered to be equivalent to the observed transition between the two states of polyQ protein (Klaips et al., 2020). In normal yeast cells, polyQ protein forms rigid aggregates that do not trap enough Hsp70 to induce HSR. By contrast, in cells overexpressing the co-chaperone Sis1, polyQ protein aggregates form a mesh structure that can trap dynamic Hsp70 (Fig. 5). This causes a productive HSR. Time series analysis shows that the overproduced of 3×GFP initially results in the formation of small aggregates that colocalize with Hsp70. This structure is probably similar to that of polyQ protein aggregates when Sis1 is not overexpressed. Eventually, the small aggregates are surrounded by Hsp70 and fuse to form large aggregates that dynamically capture Hsp70 (Fig. 4). This structure is probably similar to that of polyQ protein aggregates when Sis1 is overexpressed.

Protein overexpression results in growth inhibition (cytotoxicity) for a variety of reasons. We have categorized the mechanisms by which overexpression leads to toxicity into four categories – pathway modification, resource overload, stoichiometry imbalance and promiscuous interactions (Moriya, 2015). The formation of protein aggregates is thought to cause promiscuous interactions – loss of function by the non-specific trapping of essential proteins into the aggregates. The 3×GFP aggregates trap essential Hsp70, but this does not appear to cause growth inhibition. By contrast, overloading of the proteasome (i.e. degradation of resources) might be the cause of growth inhibition. There are two possible mechanisms by which overloading of the proteasome occurs: failure of the ability of the proteasome to process degradation due to excess ubiquitylated proteins and/or reduced activity of the proteasome itself. Further analysis will be required to solve these possibilities. We also observed that artificial trapping of essential cell cycle proteins into the 3×GFP aggregates caused cell cycle abnormalities (Fig. 7), indicating that the toxicity of the aggregates is strongly context dependent - what proteins can bind to them.

Strains, plasmids, assay kits, reagents and software used in this study are listed in Table S2.

BY4741 (MATa his3Δ1 leu2Δ0 met15Δ0 ura3Δ0) were used as the budding yeast host strains (Brachmann et al., 1998). The GFP collection, a temperature-sensitive mutant collection was created based on BY4741 (Z. Li et al., 2011; Huh et al., 2003). Cultivation and transformation of S. cerevisiae were performed as previously described (Burke et al, 2000). Synthetic complete (SC) medium without leucine and uracil (CSM-Leu-Ura, ForMedium) as indicated was used for yeast culture. Cells were cultivated at 30°C otherwise noted.

The plasmids were constructed by homologous recombination activity of yeast cells (Oldenburg et al., 1997) and their sequences were verified by DNA sequencing.

For the short-term induced heat shock and H₂O₂ stress, log phase cells cultured to an optical density at 660 nm (OD660)=0.8-1.0 were used. Heat-shock stress was applied at 42°C for 10 min using a heat block. For H₂O₂, cells were left in SC–LU with H₂O₂ (0.3%) for 10 min. For AZC, yeast was cultured in liquid SC–LU with AZC (0.75 mM; L-azetidine-2-carboxylic acid, Fujifilm Wako), and cells with OD660=0.8–1.0 were used.

Cell growth was measured every 10 min using a microplate reader (680XR Bio-Rad) for OD595. For temperature-sensitive proteasome mutants, an Infinite F200 microplate reader (Tecan) was utilized for temperature control. The maximum growth rate (MGR) was calculated as described previously (Moriya et al., 2006).

The Zombie Red™ Fixable Viability Kit (Bioregend) was used to determine whether cells with living or dead. Cells were cultured in 200 μl of SC–LU overnight (‘0 h’). The cells (50 µl) were then cultured in 200 μl of SC–LU with or without aTc (anhydrotetracycline, Cayman Chemical) and analyzed after 24 and 48 h. Fluorescence measurements were performed using flow cytometry (Gallious, Beckman colter). To distinguish and count live and dead cells in each sample, we used the staining intensity of proliferating cells and dead cells (cells treated at 70°C for 10 min) as a standard. Each time point consisted of 10,000 cells across four biological replicates.

Measurements were taken at two time points, before and 24 h after the aTc induction. Cells at 0 h, before the start of induction, were cells cultured in 200 μl of SC–LU and 96-well plates overnight in a stationary phase. For cells after the start of induction, 50 μl of cultured cells were cultured in 200 μl of SC–LU with or without aTc.

Once it was verified that the OD of each sample was within the desired range (OD660=0.5±0.1), a serial dilution of the cell culture was performed. Specifically, 5 μl aliquots of the cell culture were diluted in a ten-fold series. These diluted samples were then spotted onto SC–LU solid medium plates that did not contain aTc. Following a 24-h incubation period post spotting, images of the colonies were captured using a scanner (GT-X 980, Epson).

Log-phase cells cultured to OD660=0.8–1.0 were observed with a fluorescence microscope (Leica DMI6000B). GFP fluorescence was observed with a GFP filter cube (Leica cat. #11513899) and RFP fluorescence with an RFP filter cube (Leica cat. #11513894) for observation. Image capture was performed using image acquisition software (Leica Application SuiteX).

Cell recognition was conducted using YeastSpotter (Lu et al., 2019) to identify yeast cells from microscopy images (Fig. S8A). Subsequent quantification of the size and brightness of each cell was performed using the raw fluorescence image data in CellProfiler (Version 4.2.0) (Stirling et al., 2021). For cell size quantification in CellProfiler, the ‘Area’ parameter from the ‘MeasureObjectSizeShape’ module was employed, whereas brightness assessment was performed using the ‘Mean Intensity’ parameter from the ‘MeasureObjectIntensity’ module.

Aggregate recognition was primarily conducted through visual inspection. However, for screening involving spatial protein quality control-related genes, where analysis of over 100 images was necessary, aggregate segmentation was performed using Trainable Weka Segmentation within Fiji (Arganda-Carreras et al., 2017) (Fig. S8A). The model was developed based on manually Segmented images (at least 100 cells), following the steps outlined in the official protocol. As a specific example, representative images are shown (Fig. S8B).

Samples were prepared as for fluorescence microscopy observation. A confocal microscope (Olympus FV-3000) and a 100× oil lens were used. GFP was detected at an excitation of 488 nm and emission of 540 nm, and RFP was detected at an excitation of 561 nm and emission of 670 nm. Circular regions of a fixed size were bleached three frames after the start of the observation and then observed for 200 frames over 60 s. Bleaching was undertaken using excitation light at 488 nm. Image analysis was performed using Fiji software (Schindelin et al., 2012). Fluorescence intensity data were normalized to 1 before light bleaching and 0 immediately after bleaching.

Samples were prepared as for fluorescence microscopy observation. Glass-bottomed 96-well plates (Matsunami) were coated with 1 mg/ml of concanavalin A (Caloca et al., 2022). The observation was conducted using a confocal microscope (Olympus FV-3000) equipped with a 100× oil lens. Imaging was performed continuously for ∼16 h with an interval of about 10 min. The imaging was carried out in Z-stacks to adjust for focus shift. The temperature was maintained at 25°C during observation.

Cells were cultured in 25 ml of SC–LU medium overnight using 50 ml conical tubes. Cultured cells were washed with phosphate-buffered saline with 0.1% Tween 20 (PBST) with protease inhibitors (Thermo Fisher Scientific) and then crushed with a bead beater (TOMMY) (1 min at 4000 rpm) using glass beads. After crushing, cell extracts were centrifuged (20,630 g for 10 min) and separated into soluble (sup) and insoluble fractions (ppt). The insoluble fraction was washed with 1 ml of PBST and centrifuged (20,630 g for 5 min) three to five times. The insoluble fraction was then suspended in 1 ml of PBST and 50 μl of nickel carrier (His buffer kit, GE Healthcare) was added. The samples were kept on ice for 30 min with occasional inverted mixing. At this stage, the binding of the aggregates to the nickel carrier could be confirmed by fluorescence microscopy. The carrier was then centrifuged at low speed (300 g for 1 min) and washed with PBST three times. 100 μl of PBST with 200 mM imidazole was added to the carrier and the aggregates were eluted. The carrier was then treated with 100 μl of NuPAGE LDS sample buffer (Thermo Fisher Scientific) at 70°C for 5 min, and subjected to SDS-PAGE.

Cells overexpressing the target protein were cultured in SC–LU. Then, 1 ml of cells in the log phase were collected (OD660=0.9–1.0). Cells were treated with 0.2 mol/l NaOH for 10 min (Kushnirov, 2000) and then with 100 μl of NuPAGE LDS sample buffer (Thermo Fisher Scientific) for 5 min at 70°C.

For protein visualization, total proteins were labeled with Ezlabel FluoroNeo (ATTO) as described in the manufacturer's protocol. As a loading control using fluorescent dyes for total protein staining, a stable, prominent band of ∼50 kDa, which does not overlap with the molecular mass of the protein measured in this study, was selected (Eguchi et al., 2018). SDS-PAGE was then performed using 4–12% gradient gel. Proteins were detected on a LAS-4000 (GE Healthcare) in SYBR-green fluorescence detection mode and Image Quant TL software.

Western blotting was performed to detect specific proteins. Proteins separated by SDS-PAGE were transferred to a PVDF membrane (Thermo Fisher Scientific). The membrane was then blocked with blocking buffer (PBST with 4% skim milk) for 1 h, followed by incubation of the membrane with primary antibody diluted in PBST for 1 h. The membrane was then incubated with PBST containing peroxidase-conjugated secondary antibody (Nichirei bioscience) for 1 h. GFP was probed with anti-GFP antibody (Roche; 1:1000) and ubiquitin was probed with anti-ubiquitin antibody (Santa Cruz Biotechnology; 1:1000). The chemiluminescent image was acquired with a LAS-4000 image analyzer in chemiluminescence detection mode. Full presentations of immunoblots are shown in Fig. S9.

RNAseq analysis was performed essentially according to Namba et al. (2022). Yeast overexpressing 3×GFP or 3×MOX was grown in SC–LU medium and collected in a log phase. For vector controls, we used previous RNAseq data (Namba et al., 2022; accession number: GSE178244). RNA extraction was performed according to Köhrer and Domdey (1991). Purified RNA was quality-checked by Multina (Shimazu). cDNA libraries were prepared using the TruSeq Stranded Total RNA kit (Illumina). Paired-end sequencing was performed using the Illumina NextSeq 550 (Illumina). Three biological duplications were analyzed for all strains. Sequences were checked for read quality by FastP (Chen et al., 2018) and aligned using Hisat2 (Kim et al., 2019). Aligned data were formatted into bam files by Samtools (Li et al., 2009) and quantified by StringTie (Pertea et al., 2015). Analysis of expression levels was performed by EdgeR (Robinson et al, 2010). GO enrichment analysis was performed using the Gene Lists function on the SGD website (www.yeastgenome.org/). The raw data are available in the DNA Data Bank of Japan (accession number: PRJDB17991).

We would like to thank the members of the Moriya Laboratory (Okayama University) for their helpful discussions. We also thank the Division of Instrumental Analysis, Okayama University for the FRAP measurements.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.261977.reviewer-comments.pdf