Proteostasis collapse halts G1 progression and delimits replicative lifespan

Loss of proteostasis and cellular senescence are key hallmarks of cell aging, but whether they are subject to direct cause-effect relationships is not known. We show that most yeast cells arrest in G1 before death with low nuclear levels of cyclin Cln3, a key activator of Start extremely sensitive to chaperone status. Chaperone availability is seriously compromised in aged cells, and the G1 arrest coincides with massive aggregation of a metastable chaperone-activity reporter. A mathematical model integrating autocatalytic protein aggregation and a minimal Start network recapitulates empirical observations. As key predictions, G1-cyclin overexpression increases lifespan in a chaperone-dependent manner, and lifespan reduction by enforced protein aggregation is greatly alleviated by increased expression of specific chaperones or cyclin Cln3. Overall, our data indicate the crucial role of chaperone malfunction in setting lifespan in yeast cells, and configure a molecular pathway whereby proteostasis breakdown acts as a direct effector of cell senescence.


Introduction 1
Like most other cell types, individual yeast cells display a finite lifespan as they undergo subsequent 2 replication cycles and, due to their relative simplicity, have become a very fruitful model to study the 3 causal interactions among the different hallmarks of cell aging. Since yeast daughter cells are 4 rejuvenated during most of the mother cell lifespan, it is generally accepted that aging is the result of 5 asymmetric segregation of factors such as extrachromosomal rDNA circles (ERCs), dysfunctional 6 mitochondrial and vacuolar compartments, or resilient protein aggregates (Denoth Lippuner et al., 7 2014;Janssens and Veenhoff, 2016). Proteostasis deterioration is a universal hallmark of cellular aging 8 (Kaushik and Cuervo, 2015;Klaips et al., 2018;Labbadia and Morimoto, 2015;López-Otín et al., 2013), 9 and yeast cells have been the paradigm to study the mechanisms of asymmetric segregation of 10 protein aggregates or deposits and their relevance in aging (Hill et al., 2017). 11 Molecular chaperones play key roles in proteostasis by folding nascent polypeptides, refolding 12 misfolded proteins, and facilitating their degradation or accumulation in different types of aggregates 13 and deposits if they cannot be properly recycled (Hartl et al., 2011). Yeast cells display an age-14 dependent protein deposit, termed APOD (Saarikangas and Barral, 2016), that appears early in their 15 replicative lifespan and is retained in the mother cell compartment in every division cycle (Hill et al., 16 2017). Several chaperones including Hsp104, Ssa1 and Ydj1 co-localize with the APOD (Andersson et 17 al., 2013;Hill et al., 2014;Saarikangas and Barral, 2015), where they are thought to play a concerted 18 role in disaggregation and recycling of deposited proteins. Regarding asymmetric segregation of 19 protein aggregates during cell aging, farnesylated Ydj1 has been shown to be important for proper 20 retention of the APOD at the ER in the mother cell compartment (Saarikangas et al., 2017). The 21 functional relevance of chaperones at the crossroads of protein aggregation and replicative aging is 22 supported by the fact that Hsp104 and Ydj1 are required for a normal replicative lifespan and, when 23 overexpressed, Hsp104 restores proteasome activity in aging cells (Andersson et al., 2013) and 24 suppresses lifespan defects of sir2 mutants (Erjavec et al., 2007). Moreover, by counteracting protein 25 aggregation, overexpression of metacaspase Mca1 extends the lifespan of yeast mother cells in a 26 Hsp104-and Ydj1-dependent manner (Hill et al., 2014). 27 The interdivision time of yeast cells increases during the last cycles before death (Fehrmann et al., 28 2013;Lee et al., 2012;Lindstrom and Gottschling, 2009) and most aging cells accumulate in the 29 unbudded period before death (Delaney et al., 2013;Mcvey et al., 2001), suggesting that aging-30 related processes interfere with the mechanisms that trigger Start to drive cells into the cell cycle. 31 The Cln3 cyclin is a rate-limiting activator of Start that is maintained at low but nearly constant levels 32 during G1 (Tyers et al., 1993). Nuclear accumulation of Cln3 is driven by a constitutive C-terminal 33 nuclear-localization signal (NLS) (Edgington and Futcher, 2001;Miller and Cross, 2001), but involves 34 the essential participation of Ssa1 and Ydj1 chaperones (Vergés et al., 2007) and the segregase 1 activity of Cdc48 to release the G1 cyclin from the ER (Parisi et al., 2018). In addition, Ssa1 and Ydj1 2 also affect Cln3 stability (Truman et al., 2012;Yaglom et al., 1996), and their availability modulates 3 the execution of Start as a function of growth and stress (Moreno et al., 2019). Here we study the 4 effects of proteostasis decline during aging on the availability of Ssa1 and Ydj1 chaperones and, 5 hence, on G1 cyclin function, aiming to uncover the processes that restrain proliferation in aged cells. 6 7 Results 8 Aging cells arrest mostly in G1 with low nuclear levels of cyclin Cln3 after the last budding event 9 To analyze cell-cycle entry kinetics in the last generations prior to death, we first examined wild-type 10 cells expressing Whi5-GFP (Costanzo et al., 2004) in a CLiC microfluidics device ( Figure 1A and Movie 11 1) that had been developed for high-throughput analysis of single mother cells during aging 12 (Fehrmann et al., 2013;Goulev et al., 2017). As previously observed, the average interdivision time 13 was rather constant during aging until the senescence-entry point (SEP) (Fehrmann et al., 2013), 14 when it displayed an abrupt increase that was maintained for ca. 2-3 generations on average prior to 15 cell death ( Figure 1B). The SEP concurred with an increase in the length of both unbudded (G1) and 16 budded (S-G2-M) phases of the cycle. However, as assessed by the localization of Whi5 in the nucleus 17 to inhibit the G1/S regulon (de Bruin et al., 2004;Costanzo et al., 2004), the G1 period prior to Start 18 (T1) of the last three cycles before death displayed the largest relative increase compared to young 19 mother cells ( Figure 1C). Accordingly, while only about 15% of young mother cells are found in T1 in 20 asynchronous cultures, the percentage of cells dying in this G1 subperiod increased up to ca. 75% 21 ( Figure 1D). Finally, old cells selected with the mother-enrichment program (MEP) displayed a larger 22 fraction in G1 compared to young mother cells (Figure 1-figure supplement 1A,B). These data point to 23 the notion that the deleterious effects of aging on cell cycle progression are particularly severe in G1 24 and prior to Start. 25 Execution of Start is particularly sensitive to growth, and cells arrest in G1 when deprived of 26 essential nutrients. Nonetheless, old cells grew in volume after the last budding event at same rate as 27 in the previous cycle ( Figure 1-figure supplement 1C) and, as a result of progressive lengthening of G1, 28 their size rapidly increased during the last cycles before death (Figure 1-figure supplement 1D). Our 29 results agree with recent precise measurements of cell volume in aging cells until death in a different 30 microfluidics device (Sarnoski et al., 2018). Overall, these data would rule out possible indirect effects 31 of growth impairment on cell cycle progression in aging cells. On the other hand, the coefficient of 32 variation in volume at the last budding event was 41.2%, while young cells displayed a reduced 20.2%, 1 suggesting that cell size control at Start becomes less efficient as cells age. 2 To further characterize the observed defects in G1 progression in old mother cells, we carefully 3 quantified the levels and localization of Whi5-GFP during the last cycles before death. In agreement 4 with previous analyses of aging cells at the mRNA level (Janssens et al., 2015;Yiu et al., 2008), the 5 overall cellular concentration of Whi5 did not change much during the last cycles ( Figure 1-figure  6 supplement 1E). The nuclear/cytoplasmic ratio oscillated normally in the previous cycles to the last 7 budding event ( Figure 1E), being high in G1 and low in the budded phases of the cycle due to Cdk-8 dependent phosphorylation and nuclear export of Whi5 (de Bruin et al., 2004;Costanzo et al., 2004). 9 However, after the last budding event nuclear levels of Whi5-GFP remained low for about 200 min on 10 average, and rose again to stay high in 76.8% of cells ( Figure 1F), indicating that most cells completed 11 the last cell cycle and arrested in the next G1 prior to death. Although the differences were not as 12 large as previously reported (Neurohr et al., 2018), Whi5 levels in the nucleus displayed a 3-fold 13 increase during the last cycles and the final arrest in G1. 14 Cln3 is the most upstream G1 cyclin acting in the positive feedback loop that inactivates Whi5 and 15 executes Start (Skotheim et al., 2008;Tyers et al., 1993). Since Cln3 is too short-lived to be detected 16 as a fluorescent-protein fusion in single cells, we used a hyperstable and hypoactive Cln3 11A protein 17 fused to mCitrine (mCtr-Cln3 11A ) that can be detected by fluorescence microscopy with no gross 18 effects on cell cycle progression (Schmoller et al., 2015). As expected from its essential role in the 19 nucleus, mCtr-Cln3 11A displayed a distinct nuclear signal during the last cycles before the final budding 20 event ( Figure 1G,H, and Movie 2); however, the nuclear/cytoplasmic ratio decreased to very low 21 levels afterwards and remained low until death. In agreement with the fact that CLN3 mRNA levels do 22 not show significant changes in aged cells (Janssens et al., 2015;Yiu et al., 2008), overall cellular levels 23 of mCtr-Cln3 11A remained rather constant and similar to young cells ( Figure 1-figure supplement 1E), 24 ruling out major effects due to transcriptional or translational regulation of Cln3. In summary, our 25 data suggest that aging cells would undergo profound alterations in the mechanisms that drive 26 nuclear accumulation of cyclin Cln3 and, hence, delay G1 progression as observed in the last cycles 27 before cell death. 28 29 Ssa1/Ydj1 chaperone function is compromised in aging cells 30 We have previously shown that chaperones play a key role in the mechanisms that regulate Cln3 31 localization (Moreno et al., 2019;Parisi et al., 2018;Vergés et al., 2007). Ssa1 and Ydj1, with the 32 participation of Cdc48, are important for releasing the G1 Cdk-cyclin complex from the ER and 33 promoting its nuclear accumulation to trigger Start. On the other hand, it is generally assumed that 1 aged cells display severe defects in protein homeostasis, thereby leading to the accumulation of 2 misfolded-protein aggregates (Kaushik and Cuervo, 2015;Klaips et al., 2018;Labbadia and Morimoto, 3 2015). Thus, we decided to analyze the levels of Ssa1, Ydj1 and Hsp104 fused to fluorescent proteins 4 during aging in the CLiC microfluidics chamber. Levels of Ssa1 and Ydj1 chaperones were only slightly 5 reduced during the last cycles before death when compared to young cells (Figure 2A). By contrast, 6 Hsp104 concentration rose steadily during aging until the last budding event (Movie 1), when it 7 reached a two-fold increase compared to young cells, and continued to increase afterwards during 8 the posterior G1 arrest at an even higher average rate (Figure 2A). To confirm this result with a 9 different experimental approach we used the mother enrichment program (MEP) (Lindstrom and 10 Gottschling, 2009) to select cells aged for ca. 20 generations and also observed an increase in Hsp104 11 concentration ( Figure 2-figure supplement 1A). Observed changes in chaperone concentrations agree 12 with previous analysis at the mRNA (Yiu et al., 2008) andprotein (Janssens et al., 2015) levels, and13 suggest that cells sense proteostasis defects and, regarding to Hsp104, react during aging similarly to 14 other stress instances in which chaperone availability is assumed to be temporarily compromised (De 15 Nadal et al., 2011). As their engagement in protein interactions must cause a decrease in the diffusion 16 coefficient of chaperones, their mobility has been used as a proxy of availability (Lajoie et al., 2012;17 Moreno et al., 2019;Saarikangas et al., 2017). Thus, we used MEP-aged cells to analyze the mobility 18 dynamics of Ssa1 and Ydj1 chaperones as GFP fusions by fluorescence-loss in photobleaching (FLIP). 19 Notably, we detected a dramatic drop in mobility of both Ssa1 and Ydj1 when we compared aged cells 20 with their young counterparts ( Figure 2B,C). This decrease was similar to that caused in young cells by 21 L-azetidine-2-carboxylic acid (AZC), which induces the accumulation of misfolded proteins with 22 chaperones into disperse cellular aggregates (Escusa-Toret et al., 2013), thus compromising 23 chaperone availability. By contrast, free GFP did not display significant changes in its mobility in aged 24 or AZC-treated cells ( Figure 2C). Since AZC treatment rapidly hindered nuclear localization of mCtr-25 Cln3 11A (Figure 2-figure supplement 1B), these data point to the notion that aged cells would be 26 impaired in their ability to accumulate Cln3 in the nucleus due to severe limitations in chaperone 27 availability. 28 To further analyze chaperone mobility during cell aging we used Raster-Image Correlation 29 Spectroscopy (RICS) (Digman and Gratton, 2012) as an orthogonal approach. Briefly, RICS provides 30 information on moving molecules from raster-scan confocal images by obtaining an autocorrelation 31 function (ACF) from small arrays of pixels within the cell ( Figure 2D). After fitting a free-diffusion 32 model to the autocorrelation functions of Ssa1-GFP from young and aged cells ( Figure 2E), a 33 significant drop in the coefficient of diffusion (D) of Ssa1-GFP was detected in aged cells ( Figure 2F), 34 which was again similar to that observed in AZC-treated young cells. Moreover, a similar behavior was 1 observed for Ydj1-GFP ( Figure 2F). 2 The intersection value of autocorrelation functions obtained by RICS depends on a second 3 parameter related to the number of fluorescent molecules in the moving particles, termed brightness 4 (B). Interestingly, aged cells displayed lower B values for both Ssa1-GFP and Ydj1-GFP compared to 5 young cells (Figure 2-figure supplement 1C), which would reinforce the notion that the behavior of 6 these two chaperones is altered in aged cells, perhaps as a result of different transient interaction 7 dynamics. Contrary to the diffusion coefficient, which can only be robustly estimated after pooling 8 data from many cells and images per cell, particle brightness can be determined rather consistently at 9 a single-pixel resolution in every image to generate B maps. As shown in Figure 2G, Ssa1-GFP 10 produced rather uneven B maps in young cells, displaying moving particles with more Ssa1-GFP 11 molecules in compartments of the cell that did not particularly match the nucleus or the ER as 12 assessed with an Ole1-mCh fusion ( Figure 2G). We have recently described a procedure, called 13 coincidence analysis (Moreno and Aldea, 2019), that uses B maps to study the spatio-temporal 14 colocalization of molecular pairs undergoing transient interactions when performing their function, 15 such as Ssa1 and Ydj1. As previously observed, Ssa1-mCh and Ydj1-GFP displayed a much higher 16 coincidence coefficient compared to free GFP and mCherry and, giving support to its application as a 17 functional indicator of these two chaperones, their coincidence coefficient strongly decreased in the 18 presence of AZC. Notably, B maps of Ssa1-mCh and Ydj1-GFP were more dissimilar in aged cells, and 19 displayed a much lower coincidence coefficient compared to young cells ( Figure 2H,I), suggesting that 20 these two chaperones form less dynamic complexes in aged cells. All in all, these data point to the 21 existence of important defects in the availability and concerted activity of Ssa1 and Ydj1, two key 22 chaperones in the mechanisms that maintain protein homeostasis, in aged cells. 23 24 Firefly luciferase aggregation takes place during the G1 arrest preceding cell death 25 Firefly luciferase (FFL) refolding and enzymatic activity recovery has been widely used to assay 26 chaperone activity in vitro (Glover and Lindquist, 1998;Schumacher et al., 1996) and in vivo (Nollen et 27 al., 1999), and an FFL-GFP fusion has been used as a single-cell reporter of chaperone activity after 28 protein denaturation by heat shock (Abrams and Morano, 2013). We first compared the aggregation 29 state of FFL-GFP in young and MEP-aged cells and found that, while we were unable to detect clear 30 FFL-GFP foci in young cells, ca. 40% of cells aged for 20-25 generations showed a variable number of 31 FFL-GFP foci ( Figure 3A,B), confirming the notion that aged cells accumulate misfolded-protein 32 aggregates. We then analyzed the dynamics of FFL-GFP aggregation during aging in the CLiC 33 microfluidics chamber, and developed the required algorithms in BudJ (Ferrezuelo et al., 2012) to 34 delimit and quantify fluorescent-protein aggregates with precision ( Figure 3C). We detected the first 1 visible FFL-GFP foci around the last budding event, followed by an accelerated increase in the amount 2 of FFL-GFP present in foci until death ( Figure 3D,E). It is important to note that, while Hsp104-mCh 3 colocalized with FFL-GFP foci induced by heat shock in young cells as expected, most FFL-GFP foci in 4 aged cells did not colocalize with Hsp104-mCh in the APOD (Figure 3-figure supplement 1), indicating 5 that the normally operating mechanisms of misfolded protein recycling are altered in advanced aging. 6 As previously mentioned, overall Hsp104-mCh levels increased much faster after the last budding 7 event ( Figure 3F). However, Hsp104-mCh levels in the APOD remained constant, leading to a 8 reduction of the Hsp104-mCh fraction in the APOD relative to total levels. Notably, the fraction of 9 Hsp104-mCh in the APOD correlated at a single-cell level with the appearance of FFL-GFP foci in the 10 following 180 min ( Figure 3G), pointing to a causal relationship between Hsp104 dysfunction and FFL-11 GFP aggregation during the G1 arrest before death. Since Hsp104 levels increase under stress 12 conditions known to affect protein folding, our data reinforce the notion of proteostasis defects 13 becoming increasingly important after the last budding event. 14 15 Asymmetric aggregate inheritance predicts a decrease in chaperone availability and a G1 arrest in 16 aging cells 17 The asymmetric distribution of protein aggregates to the mother cell during cytokinesis is a key 18 safeguard mechanism to produce rejuvenated daughter cells (Hill et al., 2017). Thus, we established a 19 stochastic model based on the asymmetric distribution of protein aggregates that appear 20 stochastically during consecutive cycles of division, taking into account that chaperones are key 21 factors in two mechanistic modules: (1) counteracting protein aggregation reactions and (2) 22 facilitating nuclear accumulation of cyclin Cln3 to phosphorylate Whi5 and trigger Start ( Figure 4A). average. First, we ran the model to simulate independent single cells, and stored all variables during 30 consecutive cycles until a permanent G1 arrest was achieved, or up to a maximum time equivalent to 31 75 generations in wild-type cells under regular growth conditions. As shown in Figure 4B, simulated 32 protein aggregates increased around the last budding event, causing a sharp decrease in available 33 chaperones ( Figure 4C) and free nuclear Cln3 ( Figure 4D). Notably, all these simulated variables 34 displayed kinetics qualitatively similar to the experimental data ( Figure 4B,E insets). Simulated 1 interdivision time in consecutive cycles remained rather constant, but progressively increased during 2 the last generations before the final G1 arrest ( Figure 4E and Figure 4-figure supplement 2A), thus 3 recapitulating the SEP (Fehrmann et al., 2013). Interestingly, the time when simulated levels of 4 protein aggregates, available chaperones and free nuclear Cln3 initiated their respective changes 5 closely correlated with the SEP (Figure 4-figure supplement 2B-D). Particularly for free nuclear Cln3 6 levels, which could be more precisely measured during the last division cycles before death, we 7 observed a similar decrease to that predicted by the integrative model before and after the SEP 8  (Hill et al., 2014;Yang et al., 2011), while the whi5 knockdown 13 mutant showed the opposite behavior and lived longer than wild-type (Yang et al., 2011). We also 14 tested the effect of high and low protein synthesis rates in the model to simulate fast-and slow-15 growing cells. As experimentally observed (Kaeberlein et al., 2005;Yang et al., 2011), lifespan was 16 strongly reduced by high protein synthesis rates ( those obtained by conventional procedures (Hill et al., 2014;Yang et al., 2011). Next we used this 29 approach to estimate the lifespan of cells overexpressing CLN3 from a regulatable promoter and 30 observed a remarkable increase in the relative lifespan compared to wild-type cells as predicted by 31 the model ( Figure 5A). Daughter cells overexpressing CLN3 execute Start prematurely and bud at a 32 smaller cell size ( Figure 5B), which has been shown to have an effect on lifespan (Yang et al., 2011). 33 To avoid these effects, we activated CLN3 expression at different times after MEP induction, and 34 compared the effects of CLN3 overexpression in young cells and cells pre-aged for 24 h (12-15 1 generations) and 48 h (25-30 generations), respectively. Overexpressing CLN3 in pre-aged cells did 2 not affect their budding size ( Figure 5B), but produced a similar relative increase in lifespan ( Figure  3 5C). Thus, higher levels of Cln3 were able to increase lifespan independently of cell size. Next we 4 tested whether this effect was mediated by Ydj1 as one of the prominent chaperones acting on Cln3 5 at Start (Vergés et al., 2007). As shown in Figure 5D, overexpression of CLN3 was able to suppress 6 most of the lifespan reduction of the ydj1 mutant compared to wild-type cells, these effects being 7 qualitatively similar to those predicted by the integrative model. These data indicate that the 8 molecular deficiencies produced by lack of Ydj1 with regards to lifespan can be greatly corrected by 9 an excess of Cln3, and suggest that this G1 cyclin is a relevant chaperone client involved in cell aging. 10 On the other hand, the effects of CLN3 overexpression were also clearly attenuated by the ydj1 11 deletion, indicating that higher levels of Cln3 require the Ydj1 chaperone to extend lifespan. 12 13 Ssa1-GFP, and to a much lesser extent Ydj1-GFP, also accumulated ( Figure 6A). We then analyzed the 26 effects of these synthetic peptides on chaperone mobility by FLIP as above, and found that only PFD 27 expression caused a clear reduction in the mobility of both Ssa1-GFP and Ydj1-GFP ( Figure 6B), which 28 decreased even further for Ssa1-GFP in cells displaying PFD aggregates (PFD*). These data suggest 29 that PFD expression was able to compromise chaperone availability by sequestering Ssa1 in 30 aggregates with low exchange rates. Next we analyzed the effects of heterologous protein 31 aggregation on the nuclear localization of Cln3, and found that PFD overexpression was sufficient to 32 decrease the nuclear levels of mCtr-Cln3 11A in a dose-dependent manner ( Figure 6C,D). Consistent 33 with these results, PFD overexpression increased the average budding size ( Figure 6E). Sup35 is an 34 endogenous yeast prion that accumulates in the APOD in aging cells (Saarikangas and Barral, 2015). 1 Thus, we overexpressed the yeast prion Sup35N domain and observed an increase in the budding 2 volume of cells that showed Sup35N aggregates similar to those with PFD aggregates ( Figure 6E). In 3 marked contrast, a Sup35N m3 mutant that does not form aggregates ( Figure 6F). We observed that 9 the presence of PFD aggregates did not alter ostensibly the average budding size of first-time mother 10 cells, but the dependence on growth rate in G1 was greatly decreased as it had been observed in the 11 ydj1 mutant (Ferrezuelo et al., 2012), further suggesting that PFD aggregation affects Ydj1 availability. To confirm the notion that proteotoxic aggregates limit replicative lifespan we expressed the 16 abovementioned synthetic peptides in wild-type cells in the CLiC microfluidics chamber. Notably, PFD 17 caused a dramatic decrease in lifespan, which was accentuated even more in mother cells showing 18 PFD aggregates ( Figure 7A). The frequency of cells in G1 at death also increased about 4-fold relative 19 to young mother cells ( Figure 7B), and there was a strong correlation between PFD concentration and 20 the occurrence of death in the following 180 min ( Figure 7C). By contrast, CD levels did not correlate 21 at all with the timing of cell death ( Figure  Our data point to the notion that premature protein aggregation shortens replicative lifespan by 27 compromising chaperone availability which, in turn, would hinder nuclear accumulation of cyclin Cln3 28 and progressively delay Start, leading the cell to an irreversible G1 arrest and death. To test this 29 possibility further, we decided to analyze the effects of enforced chaperone or Cln3 expression in the 30 lifespan of PFD-expressing cells. Notably, as predicted by our model ( Figure 7E), we found that 31 overexpression of SSA1 and YDJ1 from the dual GAL1-10 promoter partially suppressed the lifespan 32 reduction caused by PFD ( Figure 7F), the lifespan being even closer to control CD-expressing cells 33 when the copy number of seven chaperone genes (SSA1, YDJ1, HSP82, CDC37, CDC48, UFD1, NPL4) 34 that cooperate in ER-release and proper Cdk-cyclin complex formation was duplicated ( Figure 7G). 1

Protein aggregation in young mother cells delays G1 progression and hinders Cln3 accumulation in
Finally, also as predicted by the model (Figure 7E), the lifespan was totally comparable to control cells 2 when PFD-expressing cells were subject to CLN3 overexpression ( Figure 7H). These results give 3 additional support to the notion that protein aggregation in young cells leads to a premature G1 4 arrest by specifically inhibiting chaperone-and G1 cyclin-dependent execution of Start. 5 6 Discussion 7 It is generally accepted that aging cells undergo many different deleterious processes that somehow 8 restrain proliferation and ultimately lead to cell death. However, their specific relevance and cause-9 effect relationships are just starting to emerge. Here we show that most yeast cells arrest in G1 10 before death and display low nuclear levels of cyclin Cln3, a key activator of Start that is particularly 11 sensitive to chaperone status (Moreno et al., 2019;Parisi et al., 2018;Vergés et al., 2007). By using 12 several independent approaches, we show that chaperone availability is seriously compromised in 13 aged cells, and we find that blockade of cell-cycle entry finely correlates with the appearance of 14 visible aggregates of a chaperone client reporter. A mathematical model integrating the role of 15 chaperones in proteostasis and cyclin Cln3 activation is able to recapitulate our observations in aging 16 cells. Notably, overexpression of Cln3 increases lifespan in a chaperone-dependent manner. As also 17 predicted by the model, overexpression of aggregation-prone proteins in young cells decreases 18 chaperone availability and restrains nuclear accumulation of Cln3 and, hence, cell-cycle entry. Finally, 19 lifespan shortening by enforced protein aggregation can be suppressed by increased expression of 20 specific chaperones or cyclin Cln3. Overall, these data establish a molecular mechanism linking loss of 21 protein homeostasis to proliferation arrest in aged yeast cells. 22 Our data agree with the recent observation that expression of G1/S genes is greatly compromised 23 in aged cells (Neurohr et al., 2018), concurrently with an increase in the nuclear levels of Whi5. Since 24 Whi5 is phosphorylated and exported to the cytoplasm by G1 Cdk-cyclin complexes (de Bruin et al., 25 2004;Costanzo et al., 2004), our observation that nuclear accumulation of mCtr-Cln3 11A is hindered in 26 aging cells would explain, at least in part, the increase in nuclear Whi5 and the deficiencies in the 27 activation of the G1/S regulon. Cells lacking Cln3 display a dramatic delay in G1, but do not arrest at 28 Start unless CLN1 and CLN2 are disrupted (Richardson et al., 1989). Thus, the observed final G1 arrest 29 should also involve other mechanisms restraining Cln1/2 levels or activity. It has been recently 30 proposed (Neurohr et al., 2018) that the accumulation of ERCs in aged cells (Shcheprova et al., 2008;31 Sinclair and Guarente, 1997) could have a direct inhibitory role on the CLN2 promoter at the nuclear 32 pore (Kumar et al., 2018). Interestingly, ERCs increase their levels around the time when G1 33 progression displays a clear delay (Morlot et al., 2018). As an alternative view, since Cln2 interacts 34 with Ssa1,2 chaperones (Gong et al., 2009) and likely requires chaperoning activities similar to Cln3 1 (Ferrezuelo et al., 2012;Moreno et al., 2019), our observations on Cln3 could also apply to basal levels 2 of Cln1 and Cln2 and, hence, contribute to explaining the final G1 arrest. 3 Mitochondrial membrane potential has been shown to play a key role in dissolution of protein 4 aggregates (Ruan et al., 2017;Zhou et al., 2014) and loss of membrane potential correlates with the 5 SEP in a fraction of aging cells (Fehrmann et al., 2013). On the other hand, vacuolar acidity declines 6 early during aging, and conditions that prevent this decline ameliorate mitochondrial function and 7 extend lifespan (Hughes and Gottschling, 2012). Related to this, the vacuolar protein Vac17 has been 8 shown to be involved in asymmetric segregation of protein aggregates (Hill et al., 2016). Interestingly, 9 we have observed that enforced protein aggregation in young cells increased vacuolar pH (our 10 unpublished observations). These findings support the notion that mitochondrial defects, vacuolar 11 dysfunction and accumulation of protein aggregates during aging would exhibit multiple functional 12 interactions (Hill et al., 2017). 13 In our experiments with the CLiC microfluidics chamber, a fraction of cells (ca. 30%) were not 14 arrested in G1 at death and showed a slightly shorter lifespan, which agrees with published findings 15 (Delaney et al., 2013). PFD-overexpressing cells also displayed a similar percentage of death outside 16 G1. Interestingly, enforced aggregation of the Rnq1 yeast prion causes a G2/M arrest with monopolar 17 spindles (Treusch and Lindquist, 2012). In all, these observations suggest that proteostasis defects 18 would also hinder cell-cycle progression and limit replicative lifespan after bud emergence. 19 Ydj1 cooperates with Hsp104 and Hsp70 chaperones to recycle misfolded proteins (Glover and 20 Lindquist, 1998), and improper recruitment of chaperones to misfolded proteins has very negative 21 effects in lifespan (Hanzén et al., 2016). In addition, Ydj1 associates with the APOD playing a key role 22 in its asymmetric segregation to the mother compartment during cell division (Saarikangas and 23 Barral, 2015;Saarikangas et al., 2017). Ydj1-deficient cells are particularly short-lived (Hill et al., Proteostasis defects have been associated with cell aging in many different model organisms 1 (Klaips et al., 2018), and the key factors that maintain the proteome in a conformationally-active state 2 are exquisitely conserved. On the other hand, as in yeast, human G1 Cdk-cyclin complexes require the 3 participation of chaperones also involved in general proteostasis (Diehl et al., 2003;Hallett et al., 4 2017). Thus, we foresee that mechanisms similar to those shown here for yeast cells could also play a 5 prominent role in restraining proliferation in aging human cells. 6 Acknowledgments 7 We thank E. Rebollo, J. Comas and M. Kerexeta for technical assistance, and D.E. Gottschling and J. 8 Skotheim for providing strains. We also thank C. Rose for editing the manuscript, and F. Antequera, Y. 9 Barral, and C. Gallego for helpful comments. This work was funded by the Ministry of Economy and 10   Chaperones can bind to all forms of unfolded and misfolded proteins, but misfolded dimers and hexamers require two or six chaperones, respectively, to properly refold all monomers. Although not shown in the wiring diagram, all species are subject to degradation reactions, and Cln3, chaperone and unfolded protein synthesis is set as a function of growth rate. Created using CellDesigner (Funahashi et al., 2008).  Increases and decreases in parameters were selected by conducting parameter sensitivity analysis in deterministic mode at 1, 10, 25% intervals after which the most biologically accurate perturbation was selected for stochastic simulations in each scenario. Initial sizes of mutants are based upon experimental data (Ferrezuelo et al., 2012).  Ferrezuelo et al., 2012), an ImageJ (Wayne Rasband, NIH) plugin that can be obtained from ibmb.csic.es/groups/spatial-control-of-cell-cycle-entry to obtain cell dimensions and fluorescence data as described (Ferrezuelo et al., 2012); budding events were identified visually. Wide-field microscopy is able to collect the total fluorescence emitted by yeast cells and, consequently, cellular concentration of fluorescent fusion proteins was obtained by dividing the integrated fluorescence signal within the projected area of the cell by its volume. The nuclear compartment was delimited as described (Ferrezuelo et al., 2012). Briefly, the gravity center from brightest pixels in the cell was used as center of a projected circle with area equal to that expected for the nucleus (17% of the cell projected area). Since the signal in the nuclear projected area is influenced by both nuclear and cytoplasmic fluorescence, determination of the nuclear concentration required specific calculations as described (Moreno et al., 2019).

Simulation Parameter alterations
Intracellular foci were detected with BudJ as pixels with a fluorescence value above a certain threshold relative to the median cell fluorescence that produced a contiguous area with a minimum size (both set by the user). In a typical set up, pixels were selected if at least 30% brighter than the cell median, with a minimal size of 0.4μm. Photobleaching during acquisition was negligible (less than 0.1% per time point) and autofluorescence was always subtracted.
Biotin labeling of the cell wall for aged cells detection. MEP-derived cells were labeled with Sulfa-NHS-LC-Biotin (Pierce) as described (Lindstrom and Gottschling, 2009), and seeded in SC medium with 1µM β-estradiol. After ageing for 1 or 2 days, cells were collected using a 0.2μm pore centrifuge filter and a soft spin. Cells were washed twice with PBS in the column and stained with a Streptavidin-APC conjugate solution (2μg/ml in PBS) for 30min at 4º; simultaneously, bud scars were labeled with a compatible WGA-conjugated fluorochrome at 20μg/ml. Afterwards cells were washed twice with media and transferred to 35-mm glass-bottom culture dishes (GWST-3522, WillCo) before microscopy.

Chaperone mobility analysis by FLIP and RICS.
We used fluorescence loss in photobleaching (FLIP) to analyze chaperone mobility in a Zeiss LSM780 confocal microscope with a 40X/1.2NA water-immersion objective at room temperature. FLIP was used as a qualitative assay to determine Ssa1-GFP and Ydj1-GFP mobility in the whole cell. A small circular region of the cytoplasm (3.6 µm 2 ) was repetitively photobleached at full laser power while the cell was imaged at low intensity every 0.5 sec to record fluorescence loss. After background subtraction, fluorescence data from an unbleached cell region were made relative to the initial time point, and a mobility index was calculated as the inverse of the fluorescence half-life obtained by fitting an exponential function. We noticed a clear dependency of this mobility index on cell size. Using a dataset of cells with very wide size range and expressing free GFP, we obtained an expression to correct the mobility index for cell size with = 1.49 , where MIfit is the fitted mobility index, r is the cell radius and MI is the corrected mobility index (data available upon request).
Raster Image Correlation Spectroscopy (RICS) analysis was performed in a Zeiss LSM780 confocal microscope with a 63X/1.3NA water-immersion objective; specifically, we used a 35nm pixel size and 12.6μs dwell time, tacking a stack of 100 frames (2sec/frame) in photon-counting detection mode, at room temperature. To obtain the coefficient of diffusion, 64x64 pixel stacks were used to remove the immobile fraction with a 5 frame moving average and analyzed using a set of plugins written by Jay Unruh (Stowers Institute) for ImageJ. The resulting autocorrelation function (ACF) in the scanning direction, i.e. GRICS(ξ,0), for each cell was obtained and averaged for each group of cells to determine the diffusion coefficient (D) and the number of molecules in the focus (N) by fitting the data to a simple diffusion model. In order to assess the uncertainty of the predicted D values from pooled RICS data by non-linear regression methods we used a Monte Carlo approximation.
Coincidence analysis with RICS data was carried out with CoinRICSJ (Moreno and Aldea, 2019). Briefly, after removal of the immobile fraction as described above, the ACF of each pixel was obtained using a 16 pixel range only in the raster direction. The intercept obtained by linear regression of the ACF (no specific model of diffusion assumed) was used as an approximation of the inverse of the number of moving particles (N). Then, the fluorescence intensity (I) at each pixel was used calculate the brightness (B) parameter as B=I/N, which were assembled into B maps covering the whole image being analyzed. Finally, correlation between Bmaps was analyzed using the Pearson's correlation coefficient, setting the threshold as the mean value in the B map. These correlation coefficients assess the degree of spatiotemporal coincidence of moving particles of the two proteins analyzed as a function of the number of fluorescent molecules per particle (Moreno and Aldea, 2019).
Lifespan analysis by MEP-induced microcolony size. MEP strains were grown as above, diluted to OD600=0.01, and plated in 500μL at ~3·10 4 cfu/cm 2 onto 35mm 2% agar plates containing SC medium with 2% glucose or 2% galactose and 1µM β-estradiol. Once the plates were dry, they were incubated at 30ºC for 4days. Finally, the microcolonies were imaged using a Leica AF7000 microscope with a 20X/0.5NA dry objective. As a proof of concept for the method, we measured the microcolony size produced by cells pre-aged in liquid media with 1µM β-estradiol for increasing amounts of time, and we observed a progressive decrease in microcolony size as a function of pre-aging time in liquid media before plating ( Figure 5-figure supplement 1C). The microcolony area was determined semiautomatically using an ImageJ macro (microcolony_size.ijm). Briefly, after thresholding and binarization, segmentation of adjacent microcolonies with the watershed function, and exclusion of the objects at the edge of the image, the area of particles (holes included) was measured. Microcolonies that were too small (with less than 4-5 cell bodies) or too big (microcolonies where cells had likely escaped from the MEP) were filtered out.
Integrative mathematical model. The wiring diagram used to describe the interaction between Start and the protein folding/aggregation pathway is described in Figure 4-figure supplement 1. We chose to focus only upon execution of Start because experimentally we found that over 75% of aging cells arrest in G1 before death. To simulate the rest of the cell cycle we run a fixed timer. The Start network was also simplified to a constantly diluting (Schmoller et al., 2015).
Whi5 molecule that is phosphorylated and inactivated by fluctuating Cln3 (Liu et al., 2015), which requires the concerted action of Ssa1 and Ydj1 chaperones for full activation (Vergés et al., 2007). Thus, execution of Start was modelled to take place when a minimal Whi5 threshold was reached. Figure 4-figure supplement 1 also details the wiring diagram used to describe the protein aggregation pathway. This again is a simplified approximation and includes dimers (which are assumed to represent a pool of all non-nucleated oligomers) and hexamers (which are assumed to represent a pool of all nucleated oligomers). This drastically decreases the number of species in the model and the complexity of the system, and we specifically chose hexamers to represent the nucleated form as they appear to have a critical size for stabilization of the oligomer in aggregating proteins (Breydo and Uversky, 2015;Xue et al., 2008). The Hsp104 disaggregase is required for disassembling large aggregates and works in conjunction with Ssa1 and Ydj1 chaperones (Okuda et al., 2015). It is therefore included in the dissociation of nucleated aggregates (hexamers), but not in monomer or dimer refolding. Ssa1 and Ydj1 are also able to suppress aggregation, presumably by refolding monomers and other small oligomers that are not nucleated, so we allow the Ssa1/Ydj1 chaperones to bind and refold these states. It is assumed that refolding always adds to the folded protein pool and that chaperones are not released until the misfolded proteins are either degraded or obtain their correct conformation.
The wiring diagram was converted into a model using COPASI (Hoops et al., 2006) with chemical reactions shown in Supplementary table 1. All chemical reactions are assumed to follow mass-action kinetics and, hence, every catalyst is placed on both sides of the chemical equation. However, we explicitly create states where the Ssa1/Ydj1 chaperones are bound to the protein, as opposed to acting as a catalyst, as this allows us to follow the pool of chaperones bound to protein aggregates. Regarding the chemical equations involving the chaperone, we take into account that larger aggregates bind more chaperones and assume a 1:1 chaperone to protein ratio. Following experimental observations, Hsp104 disaggregase is given a background concentration that increases upon accumulation of chaperone-hexamer complexes. We choose to model Hsp104 as a catalyst that does not have multiple bound and unbound states for simplicity.
In order to create full cell cycles, events were included to simulate Start and cell division when specific conditions are met. Inactivation of the Whi5 inhibitor by Cln3 is what triggers Start in the model. We assume that Whi5 has a decaying concentration during G1 (Schmoller et al., 2015) and, arbitrarily, Start is executed when 75% of Whi5 molecules are inactivated. The remainder of the cell cycle is assumed to take a constant time for simplicity. At division nuclear Cln3 is set to zero and Whi5 to the same initial Whi5 concentration in the previous generation, thus replicating the result that as cells age, the number of Whi5 molecules observed at the beginning of G1 increases with the increase in cell size (Neurohr et al., 2018).
Parameter selection was completed by scanning the parameter space using deterministic simulations in COPASI so that the average replicative lifespan matched the average value observed in experiments with wild-type cells. We then perturbed the model parameters by 1, 10 and 25% to obtain various cell cycle mutants, as described in Supplementary   table 3. The most biologically accurate perturbation for each mutant in deterministic simulations was used for all subsequent simulations in stochastic mode. We also ensured that the parameters produced biologically accurate values for Ssa1/Ydj1, Wh5 and Cln3 (10000, 1000, and 100 molecules per cell, respectively). In order to speed up simulations, overall folded protein was limited to 10 5 molecules per cell by increasing degradation of the folded protein pool, which is the final product of the process and does not affect the results of the model. Once the most biologically accurate parameter set was selected, time course simulations were run using a direct stochastic method that implemented the Gillespie algorithm to simulate aggregation as a stochastic process. The stochastic time-course simulations were run 75 times for each condition, in order to obtain the distribution and average of replicative lifespans. Due to the existence of backup mechanisms for CLN3 (basal expression of CLN1,2) or YDJ1 (SIS1), genetic ablation of these genes was simulated by applying different degrees of reduction in the concentrations of the respective proteins (Supplementary table 3 Miscellaneous. DNA-content distributions were obtained by Fluorescence Activated Cell Sorting (Gallego et al., 1997), with slight modifications to identify aged mother cells with labeled cell wall (see above) in a Gallios Flow Cytometer.