Introduction

Traditional conceptions of ageing implicate a progressive accumulation of damage leading to systemic degradation in performance until death, with evolutionary pressures acting to maximise early life fitness and fecundity at the expense of ageing health (1–3). The healthspan problem – the societal need to increase the fraction of human life spent in good health – is intractable under such models, since poor health late in life is an inevitable intermediate stage in the damage accumulation leading to death. However, ageing is not uniform across populations and from yeast to humans some members of any population maintain excellent fitness late in life (4–6); increasing the population fraction of such low-senescence individuals is a primary aim of ageing research. Interventions such as caloric restriction and mTOR inhibition can improve late-life fitness (7–11), but caloric restriction is an unrealistic societal intervention and mTOR inhibition has side effects, so new interventions are required.

Replicative ageing of the budding yeast Saccharomyces cerevisiae is a widely used model of ageing. Budding yeast cells divide asymmetrically into a mother cell and a daughter cell, with the mother cell undergoing only a limited number of divisions before permanently exiting the cell cycle (12). Mother cells maintain a rapid and uniform cell cycle for most of their replicative lifespan but the cell cycle usually slows dramatically in the last few divisions indicating loss of fitness (12–14), an effect known as ‘senescence’ following the classical definition of the term ‘senescence’ to denote a period of declining fitness for the organism at the end of life, which should not be equated with cellular senescence, the permanent cell cycle arrest caused by irreparable damage in mammalian cells. A sharp transition termed the senescence entry point (SEP) marks the change from rapid to slow, heterogeneous cycling and coincides with the appearance of apparent pathological markers (14,15). Not all cells senesce, and populations are extremely heterogeneous in some experimental systems, with individual cells following either a short-lived trajectory marked by an early SEP and accumulated foci of a mitochondrial Tom70-GFP marker when present, or a long-lived trajectory with little evidence of senescence (6,16,17). We recently reported that ageing yeast on galactose instead of glucose prevented senescence without extending lifespan across the whole population, providing a massive improvement in late-life fitness without caloric restriction (18).

The relevant metabolic differences caused by ageing on galactose are unclear, but suppression of senescence required both respiration and AMPK (18), a highly conserved kinase that reconfigures energy utilisation in response to poor nutrient availability by increasing respiration, mobilising internal energy stores, inducing autophagy and moderating fatty acid biosynthesis (reviewed in (19,20)). Due to its pivotal role in the response to low energy, AMPK has long been a focus of efforts to understand caloric restriction (21,22) and is required for the benefits of caloric restriction in yeast, worms and flies (23–25). Overexpression of the AMPK catalytic subunit extends lifespan in C. elegans and D. melanogaster, and high doses of the indirect AMPK activator metformin have been reported to extend lifespan and healthspan of mice (26–30). However, the mechanism underlying these lifespan effects is unclear, and the generality of these effects is uncertain, as other studies of metformin have reported no lifespan improvement in flies, mice or humans (31,32).

Caloric restriction robustly extends lifespan of budding yeast (33–35), but AMPK activity has different effects on yeast lifespan depending on the ageing paradigm employed, shortening replicative lifespan but extending chronological lifespan (36,37). Yeast AMPK is a central regulator of energy metabolism and shares close mechanistic and functional parallels with its counterpart in higher eukaryotes. Snf1, the catalytic subunit of AMPK in yeast, is activated by phosphorylation at a structurally conserved site, Thr210, by kinases Sak1, Elm1 and Tos3, and is regulated in response to glucose availability through Snf1 dephosphorylation, pH and subunit binding (38–40).

Suppression of fatty acid synthesis through inhibitory phosphorylation of acetyl coenzyme A carboxylase (ACC) is a highly conserved function of AMPK; this has been described for yeast Acc1, Drosophila ACC and mammalian ACC1/ACC2 (41–44). Fatty acids are synthesised from cytosolic Acetyl-CoA, which ultimately derives from glycolysis either through respiration in mammals or through oxidation of fermentation intermediates to acetate in yeast. Regardless of the source, fatty acid synthesis above immediate requirements acts as a sink of energy and carbon that can be mobilised during future scarcity by AMPK through suppression of fatty acid synthesis and promotion of fatty acid catabolism.

Here, we determine the basis of senescence and fitness loss in replicatively ageing yeast and demonstrate that senescence can be almost completely avoided by rewiring the conserved AMPK-fatty acid metabolic regulatory system.

Results

AMPK suppresses senescence through mitochondrial Acetyl-CoA import pathways

AMPK is active and necessary for growth on galactose but not on glucose (45,46) and is required for cells ageing on galactose to avoid senescence (18). We therefore asked whether activation of AMPK during ageing on glucose can recapitulate the benefits of ageing on galactose. For this, we compared log phase populations of S. cerevisiae (average age 0-1 generation) to purified cohorts aged on rich glucose media for 24 or 48 hours (average replicative ages ∼12 and ∼18 generations, at which points cells are ∼95% and ∼30% viable, respectively). Experiments were performed in batch culture with diploid cells using the mother enrichment program (MEP) to enrich aged cells, a well-characterised system that recapitulates lifespan effects of short- and long-lived mutants (47). Cell wall biotin labelling was used to purify aged mother cells, with density gradient separation to remove dead cells and debris (48,49). Cells carried Tom70-GFP as a marker of senescence (6,14), which we have previously shown to be strongly associated with other marks of senescence in the MEP system including vacuolar size, transcriptional dysregulation and chromosomal aberrations (18,50). Cells were co-stained for bud scars using WGA to quantify replicative age (51–53), and an Rpl13a-mCherry control marker that is unaffected by senescence was included in most experiments (54) (Figure 1A).

Figure 1:

AMPK was activated upon glucose by constitutive overexpression of the upstream kinase Sak1 using a P_GPD promoter as previously described (55,56) (Figure S1A). After 48 h ageing, only a subtle decrease in the Tom70-GFP senescence mark was detected in P_GPD-SAK1 cells by bulk measurements (Figures 1B, S1B), but a new subpopulation of aged cells with low Tom70-GFP and high WGA emerged (Figures 1A, 1C top left quadrant). This subpopulation has a greater replicative age (marked by WGA) since non-senescent cells continue to cycle at normal speed rather than slowing down after the SEP, and so reach a higher replicative age in 48 h. Low Tom70-GFP and high replicative age therefore mark a sub-population of aged cells with reduced senescence that is specific to the P_GPD-SAK1 mutant (Figure 1C right).

To understand how P_GPD-SAK1 reduces senescence, we performed RNA-seq comparing P_GPD-SAK1 to wild type cells at log and 48 h. Although AMPK is a major regulator of gene expression (57), only 21 genes were significantly differentially expressed at log phase in P_GPD-SAK1 (Figure 1D, left), but a greater effect was observed in aged P_GPD-SAK1 cells: 472 genes were significantly different from wild type (p<0.01) amongst which the up-regulated genes were enriched for respiration and cytoplasmic translation functions, while the down-regulated genes were not enriched for any GO terms. However, only 17 genes were substantially different (>3-fold: 8 down, 9 up including SAK1, Figure 1D right): the down-regulated genes shared no apparent connection, but the 9 up-regulated genes included all 5 of a co-regulated gene set that directs cytosolic Acetyl-CoA into gluconeogenesis via the glyoxylate cycle - MLS1, ICL1, PCK1, SFC1 and FBP1 (58,59). This pathway is not used during normal growth on glucose, though plentiful cytosolic Acetyl-CoA is available, but it is required during growth on carbon sources such as ethanol and acetate when Acetyl-CoA becomes the fuel for respiration as well as the feedstock for gluconeogenesis and associated anabolic pathways.

The glyoxylate cycle converts cytosolic or peroxisomal Acetyl-CoA into succinate for transport into mitochondria by Sfc1. Deletion of MLS1 or SFC1 significantly reduced the low senescence sub-population in P_GPD-SAK1 (Figure S1C), however Acetyl-CoA can also be transported from the cytoplasm into mitochondria after conjugation to carnitine by carnitine acetyl transferases (Figure 1E). Although multiple carnitine acetyl transferases exist outside mitochondria, this carnitine shuttle depends on the mitochondrial carnitine acetyltransferase Cat2 and deletion of CAT2, either alone or in combination with mls1Δ almost completely abrogated the effect of P_GPD-SAK1, with very few cells achieving a low senescence phenotype at 48 h (Figure 1F).

The transport of cytosolic Acetyl-CoA into mitochondria fuels respiration during growth on acetate and ethanol, and our results indicate that this process also fuels respiration during ageing on glucose. Indeed, the low senescence population in P_GPD-SAK1 is dependent on the electron transport chain and does not form in P_GPD-SAK1 cox9Δ cells, but is independent of pyruvate derived from glycolysis entering mitochondria directly via the mitochondrial pyruvate carrier Mpc1 (Figure S1D). Therefore, AMPK activation on glucose maintains mitochondrial activity and prevents the decline of mitochondria marked by high Tom70-GFP intensity, which was previously described as the cause of senescence on one yeast ageing trajectory (6,60).

Combining AMPK and Acc1 activity to prevent senescence

AMPK only suppresses senescence in a sub-population of cells, defining two classes of ageing cells – those in which senescence is reduced by AMPK activation and those in which AMPK activation is ineffective. This led us to ask whether AMPK activation also has a negative impact that dominates in the latter class. Having implicated Acetyl-CoA in senescence, we investigated other fates of this metabolite that are also regulated by AMPK. Cytosolic Acetyl-CoA is the building block for fatty acids, being processed into malonyl coenzyme A by ACC to initiate fatty acid synthesis, however ACC is inhibited by AMPK to suspend energy storage in fatty acids under low energy conditions. Yeast ACC (Acc1) is inhibited by AMPK through phosphorylation at two serines, with Ser1157 being the main regulatory site (Figure 2A) (61,62). To test whether Acc1 inhibition drives senescence in the sub-population of cells that do not respond to Sak1 over-expression, we combined P_GPD-SAK1 with an acc1^S1157A mutation that prevents Acc1 inactivation by AMPK, thereby creating the A2A mutant (AMPK and Acc1 active).

Figure 2:

Young A2A cells proliferate equivalently to wild-type cells in liquid culture (Figure S2A), and do not have an extended replicative lifespan (Figure S2B), but do show a significant and substantial increase in the low senescence population relative to P_GPD-SAK1 (Figure 2B). Separate analysis of Tom70-GFP and WGA revealed that the A2A and P_GPD-SAK1 populations were equivalent on average at 24 hours, but A2A consistently presented lower Tom70-GFP and higher WGA by 48 h (Figures 2C, S2C, S2D). Population measurements of Tom70-GFP and WGA do vary between experiments, but the differences were consistent across many experiments (Figure S2D, n>25), and bud scar counting confirmed that A2A cells reach a higher age than wild-types (Figure S2E). The combination of AMPK activation and unfettered Acc1 activity therefore allows the majority of cells to age without senescence. A2A cells cultured for 40 h reached an average replicative age comparable to wild-type at 48 h and, at this point, formed a very clear low senescence population, confirming that SEP onset is delayed or absent in age-matched cells (Figure 2D). This reduction of senescence occurred despite the viability of A2A cells (based on the ability to form a colony on a YPD plate) being lower at 40 h than wildtype at 48 h (Figure 2E). Importantly, at 40 h, >50% of the A2A cells are in the low Tom70-GFP, high age population even though <20% are viable, meaning that many cells reach the end of their replicative lifespan without passing through the SEP.

Tom70-GFP and WGA signals are only indirect indicators of cell fitness, whereas the growth rate of colonies formed by individual young or old cells provides a direct physiological measure of cell fitness. While young A2A cells formed slightly smaller colonies than did wild-type cells, the colonies formed by aged A2A cells were much larger than those formed by wild-type cells of equivalent replicative age (Figure 2F). Indeed, aged A2A cells formed colonies of similar size to young A2A cells, showing equivalent fitness, whereas most aged wild-type cells divided only once or not at all across the 24-hour assay period. Only viable cells that formed colonies within 3 days were included in this analysis, so the non-dividing aged wild-type cells were viable but had a cell division time greater than 24 hours. This demonstrates that A2A cells maintain excellent fitness late in life, whereas the fitness of aged wild-type cells is very low. A comparison of the fitness of the wild type, acc1^S1157A, P_GPD-SAK1 and A2A strains at a fixed 48 h time point revealed that A2A maintained the best fitness despite also having the highest median age, while P_GPD-SAK1 showed variable fitness, and acc1^S1157A was similar to wild type (Figure 2G). It is worth noting that colony formation by log phase cells carrying the P_GPD-SAK1 construct was consistently slightly slower than wildtype (5% longer doubling time on average) although this does not manifest in liquid growth assays.

These experiments show that the beneficial effects of AMPK activation during ageing can be increased substantially if Acc1 inhibition is prevented. High AMPK activity combined with normal Acc1 activity results in a robust low senescence ageing phenotype and high late life fitness without decreasing mortality rate.

Different classes of A2A cells depend on respiration and Acc1 activity

We then asked whether all A2A cells avoid senescence through the same mechanism. Given that the required factors (Acc1, the carnitine shuttle, the glyoxylate cycle) each deplete the cytosolic Acetyl-CoA pool, a prime candidate is the negative feedback loop involving the AMPK subunit Sip2, which inhibits AMPK when cytosolic Acetyl-CoA availability is high (63) (Figure 3A). It is not hard to imagine that the metabolic balance emerging from this negative feedback loop varies between cells, with some cells being more dependent than others on cytosolic Acetyl-CoA removal by Acc1 to maintain AMPK activity. Depleting Acetyl-CoA reduces bulk histone acetylation (64), and we observed this effect in A2A cells, indicating that cytosolic-CoA availability is indeed decreased (Figure 3B). However, AMPK activation with P_GPD-SAK1 yielded the same split population in the sip2Δ strain as in wildtype (compare Figs. 1C, 3C), and the progressive increase in the low senescence population achieved with P_GPD-SAK1 and P_GPD-SAK1 acc1^S1157A (A2A) was also observed in a sip2^3R mutant that does not respond to Acetyl-CoA (63)(Figure S3A). Therefore, the differential responses to AMPK activation in the ageing population cannot be attributed to Sip2-mediated feedback, but depletion of cytosolic Acetyl-CoA may be beneficial independent of Sip2.

Figure 3:

To determine whether reducing the availability of Acetyl-CoA is beneficial in ageing, we deleted ALD6, which encodes the primary cytoplasmic acetaldehyde reductase. Loss of Ald6 reduces the supply of acetate from fermentation to be processed into Acetyl-CoA (65,66). Although this did not have a large effect on the low Tom70-GFP, high WGA population as defined in Figure 1, the deletion of ALD6 was effective in decreasing Tom70-GFP and increasing age at harvest (Figure 3D). These findings showed that ald6Δ cells maintain high fitness for a longer portion of life. However, fitness measured at 48 h was very low (Figure 3E) meaning that the fitness gains from reducing cytosolic Acetyl-CoA availability are temporary and are lost by 48 h; therefore, the A2A phenotype cannot be attributed solely to a reduction of Acetyl-CoA availability.

Conversely, the low senescence of A2A populations may have different causes in different cells, with some cells requiring AMPK-driven respiration fuelled by cytosolic Acetyl-CoA and other cells requiring the combination of high AMPK and high Acc1 activity. To test this hypothesis, we deleted CAT2 and MLS1 in A2A, which should impair ageing fitness of only those cells requiring AMPK-driven respiration: this resulted in the population dividing between low and high senescence states based on flow cytometry and fitness (Figure 3F,G). Similarly, deleting COX9 to prevent respiration removed a subset of cells from the low senescence population (Figure S3B). Therefore, cells that achieve a low senescence state through respiration in P_GPD-SAK1 avoid senescence by the same mechanism in A2A, but the cells that avoid senescence due to the acc1^S1157A mutation in A2A use a mechanism independent of the carnitine shuttle and glyoxylate cycle. We further asked whether respiration itself or reduction in Acetyl-CoA availability is important for the first group of cells by deleting ALD6 in P_GPD-SAK1; if respiration is important, this should suppress the benefit of P_GPD-SAK1 by reducing the fuel for respiration, whereas if Acetyl-CoA removal is important then this should be neutral or beneficial. Working in sip2Δ to avoid an increase in AMPK activity due to reduced Acetyl-CoA availability, we observed that ald6Δ increased the low senescence population through decreasing Tom70-GFP (S3C), and therefore the beneficial impact of P_GPD-SAK1 on this pathway arises from Acetyl-CoA removal rather than a benefit of respiration.

These experiments define two classes of ageing cells with distinct metabolic needs, coherent with the model of two ageing trajectories previously proposed (6). In one class, respiration fuelled by cytosolic Acetyl-CoA is required to suppress senescence, while in the other class respiration is not required but maintaining Acc1 activity is critical. These processes all consume cytosolic Acetyl-CoA, and reducing the availability of this metabolite is beneficial but insufficient to prevent senescence.

Fatty acid supply is limiting in ageing cells that do not respond to AMPK activation

Acc1 converts Acetyl-CoA to Malonyl-CoA as the first step in fatty acid synthesis, a biosynthetic activity that may or may not be important during ageing. We asked whether fatty acids or perhaps the Malonyl-CoA intermediate prevent senescence onset in A2A, against a null hypothesis that Acc1 simply provides an alternative pathway to respiration for moderating cytosolic Acetyl-CoA levels (Figure 4A). To test this hypothesis, we attempted to reproduce the low senescence phenotype of A2A by manipulating fatty acid synthesis in P_GPD-SAK1, again in the absence of Sip2 to avoid feedback on AMPK activity (Figure 4B).

Figure 4:

Malonyl-CoA was recently found to be an endogenous mTOR inhibitor (67), and mTOR inhibition is well known to increase ageing health and lifespan (11). However, a low dose of the fatty acid synthase inhibitor cerulenin, which causes Malonyl-CoA to accumulate (67,68), profoundly decreased the low senescence population (Figure 4B, conditions 2,3). A higher dose of cerulenin was also tested in combination with oleic acid supplementation to complement the complete blockade of fatty acid synthesis, but this combination had very little effect (Figure 4B, condition 4). Therefore, Acc1 activation in A2A cells does not work through increased Malonyl-CoA levels. In contrast, supplementation with only the monounsaturated fatty acid oleic acid increased the low senescence population of P_GPD-SAK1 cells even more than acc1^S1157A, showing that fatty acid scarcity is the primary driver of senescence in ageing cells that do not respond to AMPK activation alone.

None of these treatments had a strong impact on Tom70-GFP however, with increases in the low Tom70-GFP, high WGA population being driven almost entirely by increased WGA (Figure S4A), indicating that the strongest effect of fatty acid supplementation was to overcome the known short lifespan of the sip2Δ background (37). After validating the effect of oleic acid by bud scar counting (Figure S4B), we therefore asked whether oleic acid supplementation also reduces senescence in a wild-type background. The effects were dramatic, with the low senescence population increasing 2.5-fold through decreased population average Tom70-GFP and increased WGA (Figures 4C, S4C). The low senescence population in P_GPD-SAK1 also almost doubled with oleic acid supplementation, but only a small, non-significant effect on A2A was observed (Figure 4C). Notably, oleic acid supplementation and P_GPD-SAK1 each allow approximately half of the population to age with low senescence, while the combination allows most cells to avoid senescence - this is as expected if P_GPD-SAK1 and oleic acid supplementation suppress senescence in different classes of ageing cells.

At 48 h the fitness of wildtype cells supplemented with oleic acid also increased dramatically, P_GPD-SAK1 cells improved somewhat while fitness was significantly but not substantially decreased in A2A (Figure 4D). In fact, the oleic acid-supplemented wild-type cells were fitter than the mutants, which likely reflects the decrease in baseline fitness caused by P_GPD-SAK1 in young cells as noted above. The results for P_GPD-SAK1 confirmed that a subset of cells in which AMPK is activated still undergo senescence because of fatty acid starvation, which can be compensated by the acc1^S1157A mutation or by oleic acid supplementation.

However, the results for the wild type are more complicated as ∼40% of oleic acid-supplemented cells undergo senescence according to flow cytometry whereas all oleic acid supplemented wild-type cells are extremely fit (compare Figures 4C and 4D). This difference arises because fitness assays exclude inviable cells, such as the cells that senesce and lose viability before 48 h due to lack of respiration, which are exactly the cells for which senescence is rescued by AMPK activation. We further noted that the low fitness of aged ald6Δ cells was rescued by oleic acid to wild-type levels, showing that the observed fitness defect simply arose from fatty acid starvation rather than from any other function of Acetyl-CoA (Figure S4D). Overall, in wild-type cells, as in P_GPD-SAK1 cells, high Acc1 activity or oleic acid supplementation prevents senescence in one class of ageing cells while AMPK activation prevents senescence in the other class.

Discussion

Our findings indicate a major role for Acetyl-CoA metabolism in the yeast replicative ageing process. Combining transport of cytosolic Acetyl-CoA into mitochondria with ongoing cytosolic fatty acid synthesis together ensures that cells rarely enter senescence. This is achieved without reducing age-linked mortality and with little effect on the fitness of cells when young, showing that lifespan and fitness decline are separable in the yeast replicative ageing system, and that high fitness can be maintained to the end of life.

Here we have characterised two classes of ageing cells seemingly differentiated by high and low availability of cytosolic Acetyl-CoA, consistent with a previous demonstration that ageing follows two trajectories in yeast (6). Acetyl-CoA is the feedstock for fatty acid synthesis, which is necessary both for membrane production and for energy storage in lipid droplets. However, cytosolic Acetyl-CoA production consumes as much ATP as is generated in glycolysis and thus competes directly with growth. The existence of two classes of cell that follow different ageing trajectories makes sense because yeast cells do not know when food availability will change and so the population employs a bet hedging strategy (69–71): part of the population synthesises excess cytosolic Acetyl-CoA to store as much energy as possible in lipids against future scarcity, while other cells maximise growth from the current food supply by synthesising the bare minimum of cytosolic Acetyl-CoA needed for membrane formation (Figure 5). Our findings indicate that though both strategies are useful they eventually decrease fitness as cells age, forming a classic antagonistic pleiotropy (3).

Figure 5:

Multiple pathways can promote senescence during yeast ageing; which pathway dominates is dependent on the experimental system and therefore the set of stresses experienced by cells (6,15,17,72,73). The SEP in yeast was originally characterised in microfluidic assays, where it is attributed to the accumulation of extrachromosomal rDNA circles (ERCs) (14,15,73). In contrast, ERC accumulation is completely unrelated to senescence during ageing in the batch culture system employed here (18,48,50), which is sensitive to metabolic changes such as a galactose diet or Acetyl-CoA processing; the galactose diet has no effect on cells in microfluidic systems and the impact of caloric restriction has remained controversial, so we do not necessarily expect A2A to prevent senescence in microfluidic systems (74,75). However, our findings parallel studies of yeast chronological ageing which have highlighted the importance of respiration and lipid synthesis, though chronological ageing measurements are performed in glucose-depleted conditions and it has been difficult to translate these findings to ageing under normal nutrient availability (23,64,76). Whatever the root of the difference, ageing in batch culture allows the exploration of ageing biology with little impact from ERC accumulation, a phenomenon that does not occur during ageing in animals.

AMPK is broadly considered to have a pro-longevity effect, however the benefits of AMPK activation under normal diets are limited, and some studies in budding yeast conversely found that AMPK activity is negative for lifespan (77,78). Our work shows that both positions can be true, with AMPK having both beneficial and detrimental impacts under nutrient replete conditions. It is not too surprising that issues arise when AMPK is ectopically activated on high glucose given that AMPK is a central energy regulator that evolved to manage metabolism under low nutrient conditions, and it is perhaps more surprising that a single point mutation largely corrects these issues.

The AMPK-Acc1 regulatory system is conserved from yeast to humans, but the source and fate of cytosolic Acetyl-CoA are different so the combination of AMPK activation with lipid supplementation is unlikely to be universally effective. In human cells, the only major route for removal of excess cytosolic Acetyl-CoA is lipid synthesis to palmitoyl coenzyme A followed by carnitine-dependent transport into mitochondria for oxidation. Ectopic activation of AMPK under high nutrient availability would promote lipid catabolism but inhibits lipid synthesis through ACC1/2 phosphorylation. Preventing inhibition of ACC1/2 by AMPK is ineffective as ACC2 also inhibits fatty acid catabolism so fatty acid synthesis and storage increase, resulting in non-alcoholic fatty liver disease (79). However, we speculate that activating AMPK while selectively preventing AMPK inhibition of ACC1 alone may be effective.

Materials and Methods

Detailed and updated protocols are available at https://www.babraham.ac.uk/our-research/epigenetics/jon-houseley/protocols.

Yeast strain construction, culture and labelling

Yeast deletion mutants and fusion constructs were constructed by standard methods and are listed in Table S1, oligonucleotide sequences are given in Table S2. Plasmid templates were pFA6a-GFP-KanMX4, pYM-N14, pAW8-mCherry, pFA6a-HIS3 and pFA6a-URA3 (80–82). To create the acc1^S1157A point mutation, ACC1 was amplified with ACC1 F3/R3 and cloned in pRS314, then the S1157A mutation introduced by site directed mutagenesis. MEP a ±Kan-P_GPD-SAK1 cells carrying this plasmid were transformed with pGSKU (83) amplified with ACC1 CORE UP45 1 and ACC1 CORE DN45 2 to replace the chromosomal ACC1 S1157 codon with a CORE cassette flanked on each side by I-SceI sites. Cells were plated on YPGal to induce I-SceI then FOA to select for loss of the CORE cassette, then TRP negative cells lacking the pRS314-acc1^S1157A plasmid were verified by sequencing. sip2^3R was constructed by a simpler method as SIP2 is not an essential gene: a CORE cassette flanked by I-SceI sites was amplified from pGSKU using SIP2 3Q-R CORE UP45 and SIP2 3Q-R CORE DN45 v2 and transformed in destination strains. These were transformed with pRS413 containing an EagI-SacI fragment of SIP2 with the sip2^3R sequences, then cells plated on YPGal then FOA.

All cells were grown in YPD media (2% peptone, 1% yeast extract, 2% glucose) at 30°C with shaking at 200 rpm. Media components were purchased from Formedium and media was sterilised by filtration. MEP experiments were performed as described with modifications (84): cells were inoculated in 4 ml YPD and grown for 6-8 hours then diluted in 25 ml YPD and grown for 16-18 hours to 0.2-0.6×10⁷ cells/ml in 50 ml Erlenmeyer flasks. 0.125×10⁷ cells per aged culture were harvested by centrifugation (15 s at 13,000 g), washed twice with 125 µl PBS and re-suspended in 125 µl of PBS containing ∼3 mg/ml Biotin-NHS (Pierce 10538723). Cells were incubated for 30 min on a wheel at room temperature, washed once with 125 µl PBS and re-suspended in 125 µl YPD then inoculated in 125 ml YPD at 1×10⁴ cells/ml in a 250 ml Erlenmeyer flask (FisherBrand FB33132) sealed with Parafilm. 1 µM β-estradiol (from stock of 1 mM Sigma E2758 in ethanol) was added after 1.5h. An additional 0.125×10⁷ cells were harvested from the log phase culture while biotin labelling reactions were incubating at room temperature. Cells were harvested by centrifugation for 1 min, 4600 rpm, immediately fixed by resuspension in 70% ethanol and stored at −80°C. To minimise fluorophore bleaching in culture, the window of the incubator was covered with aluminium foil, lights on the laminar flow hood were not used during labelling and tubes were covered with aluminium foil during biotin incubation. Lifespan assays in the MEP background were performed as previously described (47,84).

Cell purification

Percoll gradients (1-2 per sample depending on harvest density) were formed by vortexing 1 ml Percoll (Sigma P1644) with 110 µl 10x PBS in 2 ml tubes and centrifuging 15 min at 15,000 g, 4°C. Ethanol fixed cells were defrosted and washed once with 1 volume of cold PBSE (PBS + 2 mM EDTA) before resuspension in ∼100 µl cold PBSE per gradient and layering on top of the pre-formed gradients. Gradients were centrifuged for 4 min at 2,000 g, 4 °C, then the upper phase and brown layer of cell debris removed and discarded. 1 ml PBSE was added, mixed by inversion and centrifuged 1 min at 2,000 g, 4 °C to pellet the cells, which were then re-suspended in 1 ml PBSE per time point (re-uniting samples where split across two gradients). 25 µl Streptavidin microbeads (Miltenyi Biotech 1010007) were added and cells incubated for 5 min on a wheel at room temperature. Meanwhile, 1 LS column per sample (Miltenyi Biotech 1050236) was loaded on a QuadroMACS magnet and equilibrated with cold PBSE in 4 °C room. Cells were loaded on columns and allowed to flow through under gravity, washed with 1 ml cold PBSE and eluted with 1 ml PBSE using plunger. Cells were re-loaded on the same columns after re-equilibration with ∼500 µl PBSE, washed and re-eluted, and this process repeated for a total of three successive purifications. After addition of Triton X-100 to 0.01% to aid pelleting, cells were split into 2 fractions in 1.5 ml tubes, pelleted 30 s at 20,000 g, 4 °C, frozen on N2 and stored at −70 °C.

Fitness and viability assays

For live purification, cells were pelleted, washed twice with synthetic complete 2% glucose media, then resuspended in 2 ml final volume of the same media and incubated for 5 min on a rotating wheel with 10 µl MyOne streptavidin magnetic beads (ThermoFisher), isolated using a magnet and washed five times with 1 ml of the same media. Cells were streaked on a YPD plate and individual cells moved to specific locations using a Singer MSM400 micromanipulator. Colony size was measured 24 hours later on the screen of the micromanipulator imaging with a 4x objective, sizes were calibrated using a haemocytometer. Cells that did not form a colony within 72 h were assumed to have been inviable and were not included in the analysis.

For viability assays, log phase cells were inoculated in YPD at 1×10⁴ cells/ml and 60 µl spread on a YPD plate. 1 µM β-estradiol was then added to the cultures, and after 40-48 h 1.5 ml of cells were harvested, washed with 100 µl of water and 40 µl spread on a YPD plate. Colonies were counted after 2 d (young) or 3 d (aged).

RNA extraction

Cells were re-suspended in 50 µl Lysis/Binding Buffer (from mirVANA kit, Life Technologies AM1560), and 50 µl 0.5 µm zirconium beads (Thistle Scientific 11079105Z) added. Cells were lysed with 5 cycles of 30 s 6500 ms⁻¹ / 30 s on ice in Next Advance Bullet Blender Storm 24 (ThermoFisher) in cold room, then 250 µl Lysis/Binding buffer was added followed by 15 µl miRNA Homogenate Additive and cells were briefly vortexed before incubating for 10 minutes on ice. 300 µl acid phenol : chloroform was added, vortexed and centrifuged 5 min at 13,000 g at room temperature before extraction of the upper phase. 400 µl room temperature ethanol and 2 µl glycogen (Sigma G1767) were added and mixture incubated for 1 hour at −30 °C before centrifugation for 15 minutes at 20,000 g, 4 °C. The pellet was washed with cold 70% ethanol and re-suspended in 10 µl water. 1 µl RNA was glyoxylated and analysed on a BPTE mini-gel, and RNA was quantified using a Qubit RNA HS Assay Kit.

150 ng RNA was used to prepare libraries using the NEBNext Ultra II Directional mRNA-seq kit with poly(A)+ purification module (NEB E7760, E7490) as described with modifications: Reagent volumes for elution from poly(T) beads, reverse transcription, second strand synthesis, tailing and adaptor ligation were reduced by 50%; libraries were amplified for 13 cycles using 2 µl each primer per 50 µl reaction before two rounds of AMPure bead purification at 0.9x and elution in 11 µl 0.1x TE prior to quality control using a Bioanalyzer HS DNA ChIP (Agilent) and quantification using a KAPA Library Quantification Kit (Roche).

Sequencing and bioinformatics

Libraries were sequenced by the Babraham Institute Sequencing Facility using a NextSeq 500 instrument in 75 bp single end mode. After adapter and quality trimming using Trim Galore (v0.6.6), RNA-seq data was mapped to yeast genome R64-1-1 using HISAT2 v2.1.0 (85) by the Babraham Institute Bioinformatics Facility. Mapped data was imported into SeqMonk v1.47.0 (https://www.bioinformatics.babraham.ac.uk/projects/seqmonk/) and quantified for log₂ total reads mapping to the antisense strand of annotated open reading frames (opposite strand specific libraries), excluding the mtDNA and the rDNA locus. Read counts were adjusted by Size Factor normalisation for the full set of quantified probes (86).

GO analysis of individual clusters performed using GOrilla (http://cbl-gorilla.cs.technion.ac.il/) (87). Quoted p-values for GO analysis are FDR-corrected according to the Benjamini and Hochberg method (q-values from the GOrilla output), for brevity only the order of magnitude rather than the full q-value is given (88).

Flow cytometry

Cell pellets were re-suspended in 240 µl PBS and 9 μl 10% triton X-100 containing 0.3 µl of 1 mg/ml Streptavidin conjugated with Alexa Fluor 647 (Life technologies) and 0.6 μl of 1 mg/ml Wheat Germ Agglutinin (WGA) conjugated with CF405S (Biotium). Cells were stained for 10 min at RT on a rotating mixer while covered with aluminium foil, washed once with 300 μl PBS containing 0.01% Triton X-100, re-suspended in 200 μl PBS and immediately subject to flow cytometry analysis on a Cytek Aurora. Cells for single colour controls were prepared in the same manner as the fully stained samples. On the Aurora, time-matched unmixing reference files were required for different ages as cell size and autofluorescence varied with age. Therefore, numerical differences in fluorescence measurements between different ages are not absolutely quantitative. Cell populations were gated on FSH-H and SSC-H to remove debris, on FSC-A and FSC-H to select for single cells, and streptavidin positive (AF647) cells were selected, all in a hierarchical manner and 10,000 events acquired. Low senescence aged populations were defined using quadrants of low GFP and high WGA, with quadrants being defined on the wild-type and applied uniformly across samples within an experiment. As is visible in the representative flow plots, quadrants are almost invariant across most datasets, with differences arising over time through changes in machine performance.

A subset of samples was acquired on an Amnis ImageStream X Mk II with the following laser power settings: 405 = 25 mW, 488 = 180 mW, 561 = 185 mW, 642 = 5 mW, SSC = 0.3 mW. On the ImageStream, cell populations were gated for; (1) single cells based on area and aspect ratio (>0.8), (2) in-focus cells were gated based on a gradient RMS value (>50), and (3) streptavidin positive (AF647) cells, all in a hierarchical manner and 1,000 events acquired. Before data analysis, compensation was applied according to single-colour controls and auto-generated compensation matrix, see (18) for additional parameters.

As absolute measured fluorescence intensities varied somewhat over time and between machines, median normalisation has been applied across all samples of a given age in certain datasets to allow comparison of Tom70-GFP and WGA intensities between datasets. This is a single transformation applied equally to all samples such that differences between samples are maintained, and is not applied to the quadrant analysis described above.

Statistical analysis

All statistical analysis was performed in GraphPad Prism (v10.4.1), tests and n values are given in Figure legends.

Supplementary information

Figure S1:

Figure S2:

Figure S3:

Figure S4:

Table S1:

Table S2:

Data availability

RNA-seq data is deposited at the Gene Expression Omnibus (GEO), Accession GSE308402.

Acknowledgements

We would like to thank Sophie Trefely and Mark Cheong for experimental help and insightful discussions. We also thank Sophie Trefely for critical reading of the manuscript. We would like to thank Rachael Walker, Sam Thompson, and Christopher Hall of the Babraham Institute Flow Facility (Flowcyt04) for extensive experimental advice and troubleshooting. RNA-seq library sequencing and processing were performed by Babraham Institute Genomics (Geno06) and Babraham Institute Bioinformatics (Bioinf01), which receive financial support from the Institute Core Capability Grant (BBSRC CCG).

Additional information

Funding

JH,MU,MKAM - BBSRC (BI Epigenetics ISP; BBS/E/B/000C0523), HHM – UKRI (Horizon Europe Marie Skłodowska-Curie fellowship; EP/Y027892/1), DH - BBSRC (DTP PhD award 1947502).

Funding

UKRI | Biotechnology and Biological Sciences Research Council (AFRC) (BBS/E/B/000C0523)

• Jonathan Houseley

UKRI | Biotechnology and Biological Sciences Research Council (AFRC) (1947502)

• Jonathan Houseley

UKRI | Engineering and Physical Sciences Research Council (SRC) (EP/Y027892/1)

• Jonathan Houseley