Potassium ion homeostasis modulates mitochondrial function

The decline of mitochondrial function with age is observed across eukaryotes (Somasundaram et al., 2024; Chistiakov et al., 2014; Lima et al., 2022). This decline in function has been associated with cellular-level disorders such as reduced membrane potential, calcium cycling, and ATP production, increased production of reactive oxygen species, and loss of proteostasis (Sun et al., 2016; Haynes and Hekimi 2022; Pi et al., 2007). In metazoa, aging-related loss of mitochondrial function has been found in many tissues (Somasundaram et al., 2024) and is associated with general disorders such as decreased metabolic and muscle function, increased inflammation, and stem cell senescence (Sun et al., 2016; Grevendonk et al., 2021), as well as specific aging-related diseases of the lung (Cloonan et al., 2020), heart (Chistiakov et al., 2014), and brain (Coskun et al., 2012).

The complexity of mitochondrial biology and its connection to other subcellular systems (Gottschling and Nyström 2017) means that many biological processes currently not associated with mitochondrial health or function could nevertheless affect mitochondrial function within the context of the cell as a whole. For instance, a screen for genes that delayed the age-associated fragmentation of mitochondria found that this dysfunction was preceded by a loss of acidity in the lysosome-like vacuole in the budding yeast Saccharomyces cerevisiae (Hughes and Gottschling 2012). Mitochondria also engage in interorganelle communication through a wide variety of contact sites (Eisenberg-Bord et al., 2016). Loss of function in any of these other subcellular systems could directly affect mitochondrial function through a reduction or absence of these points of communication (Casler et al., 2025).

Of all the ways mitochondrial function can decline, the central role of the mitochondrial membrane potential (MMP) makes it a particularly good candidate for monitoring mitochondrial health (Nicholls 2004). In healthy cells, the MMP is primarily generated by the electron transport chain (ETC), but it is also influenced by any changes within the cell that alter the charge balance across the mitochondrial inner membrane (Zorova et al., 2018). Thus, the MMP integrates the state of cellular ion homeostasis, including the amounts and relative concentrations of charged molecules and ions in the cytosol relative to the mitochondrial matrix. These cellular properties are in turn determined by the amount and activity of channels and transporters (Palmieri and Monné 2016; Szabo and Szewczyk 2023).

The MMP is used to perform essential cellular functions. While generation of ATP via oxidative phosphorylation is probably the most well-known use of MMP, the mitochondria itself could not function without the MMP that it generates, as nearly all of the ∼1,000 proteins necessary for mitochondrial function are encoded in the nuclear genome, translated in the cytosol, and imported into the mitochondria in an MMP-dependent manner (Kutik et al., 2007; Sickmann et al., 2003). The molecular precursors necessary to generate ATP, such as pyruvate, ADP, and inorganic phosphate, all rely on the MMP-dependent transport to enter the mitochondria (Palmieri and Monné 2016), while the products of mitochondrial metabolism, such as ATP, nonessential amino acids, and other metabolites, rely on MMP-dependent export to be utilized by other parts of the cell (Palmieri and Monné 2016). Finally, apart from its role as a readout of overall mitochondrial health, the MMP appears to be a relevant target for intervention. Recent work in Caenorhabditis elegans demonstrated that artificially maintaining a high MMP using optogenetic tools was sufficient to improve mitochondrial function during stress and increase worm lifespan (Berry et al., 2020; Berry et al., 2023).

In an attempt to discover new pathways contributing to mitochondrial function, we developed a genome-wide screening platform based on MMP. With the information from this screen, we sought to find novel ways of intervening to delay or prevent the age-associated decline in MMP.

To investigate the relationship between MMP and lifespan, we genomically integrated a fluorescent protein–based, two-component reporter system (Fig. 1, A and B; Video 1) into S. cerevisiae (Materials and methods). This system contained an MMP-sensitive (MMP_s) component and an MMP-insensitive (MMP_i) component to allow longitudinal, ratiometric measurement of MMP in live cells, and circumvents many limitations of dye-based MMP reporters such as tetramethylrhodamine methyl ester (TMRM). The dyes cannot be fixed for fluorescence-activated cell sorting (FACS) analysis and cannot be left in live cells for extended periods of time, and their signal does not take into account changes in mitochondrial mass or the amount of dye taken up by cells, which is determined by the plasma membrane potential (Scaduto and Grotyohann 1999; Brand and Nicholls 2011).

The expression of the MMP_s component is driven by the promoter of TPI1 and contains the leader sequence of COX4 (“preCOX4”) fused to the fluorescent protein mNeonGreen and the degron SL17 (Gilon et al., 1998). Import of preCOX4-tagged fluorescent proteins is well characterized and specifically depends on MMP (Veatch et al., 2009; Fehrmann et al., 2013; Garipler et al., 2014; Vowinckel et al., 2015), including throughout replicative lifespan (Fehrmann et al., 2013). The SL17 degron is a well-characterized signal sequence that confers a half-life of minutes in the cytosol (Gilon et al., 1998), but is completely protected from degradation in the mitochondrial matrix (Shlevin et al., 2007; Papić et al., 2013), thus preventing accumulation of the mNeonGreen signal in the absence of MMP (Garipler et al., 2014; Vowinckel et al., 2015).

The MMP_i component is a C-terminal mCherry fluorescent tag of the mitochondrial inner membrane protein Tim50p, which is part of the Translocase of the Inner Mitochondrial membrane complex (TIM23) (Kutik et al., 2007). Tim50p is essential (Yamamoto et al., 2002), accurately represents mitochondria in aged cells lacking MMP ([Hughes and Gottschling 2012], Fig. S1, B and E), and, unlike mitochondrial outer membrane proteins, is not found in aging-related, vacuolar-associated, mitochondrial-derived compartments (Hughes et al., 2016). These attributes make Tim50p a good MMP_i reporter of mitochondrial mass. We used the ratio of the mNeonGreen signal to the mCherry signal (“MMP ratio”) as a readout of the cell’s MMP-dependent import capacity normalized by its mitochondrial mass.

We used the Yeast Lifespan Machine (YLM [Thayer et al., 2022, Preprint]) with fluorescence images taken every 2 h to track the MMP ratio of ∼12,000 cells as they aged (Fig. 1, C and D; Fig. S1; Materials and methods). While there was significant heterogeneity in individual cell trajectories, we found that, similar to Hughes and Gottschling (2012), but in contrast to Fehrmann et al. (2013), the MMP ratio of every cell declined with age. This decline in MMP ratio was driven by a decrease in the MMP_s signal (Fig. S1), and occurred regardless of whether it was measured by time (Fig. 1 C, Fig. S1, first column), number of divisions (Fig. 1 D, Fig. S1, second column), number of divisions before death (Fig. S1, third column), or fraction of divisions remaining (Fig. S1, fourth column). Using an orthogonal, microscopy-based imaging assay, we found that relative to a new mother cell, TMRM declined more rapidly with age than MMP_s (Fig. S2 B). Thus, the MMP_s reporter is somewhat delayed in responding to reduction of MMP with age. Interestingly, there was a significant correlation between a cell’s MMP ratio averaged over its first five divisions and the number of divisions it completed before death (r = 0.55, p ∼ 0; Fig. 1 E), suggesting that higher initial MMP is associated with a longer lifespan.

In order to discover processes that could delay or prevent the age-associated reduction in MMP, we set up a screening platform to identify loss-of-function mutants that altered MMP in aged cells (Fig. S3 A). Specifically, we introduced the reporter system (Fig. 1 A) into the yeast knockout collection (Winzeler et al., 1999), in which each of the ∼4,700 nonessential genes has been deleted and tagged with a unique barcode (Materials and methods). This deletion library was pooled and aged using Ministat Aging Devices (MADs) (Hendrickson et al., 2018) for 24 h (mean age 15.1 divisions; Fig. S3 C), a time at which wild-type (WT) cells have a dramatically reduced MMP as evaluated by either TMRM or the MMP reporter ratio (Fig. 1 C, Fig. S2 B). We then used FACS to sort mother cells from the bottom, middle, or top thirds of the distributions of the MMP ratio (calculated from the green/red signals) (Fig. S3 B; Materials and methods).

We focused our attention on the mutants enriched in the high MMP ratio group relative to the low MMP ratio group. Deletions of the top 35 mutant genes from the screen were remade in a prototrophic strain background that contained the MMP reporter system (Table S5). Using the YLM, we tracked the MMP of mother cells in each of these strains. Statistical modeling of the trajectories (Materials and methods) revealed that 26 (74%) of the deleted genes identified in the screen had a higher MMP ratio than WT after 24 h in the microfluidics device (at a significance threshold of P <0.05), indicating good concordance between the screen and microfluidics approaches. In young cells, TMRM accumulation was strongly correlated with the reporter MMP_s signal that was normalized to total mitochondrial mass (Pearson’s r = 0.79, P = 1.2 × 10⁻⁷; Fig. S3 F), thus providing independent evidence for the success of the MMP reporter and its utilization in this screen.

Of the 26 deletion mutants confirmed to have a higher MMP ratio than WT after 24 h, 21 had significantly longer lifespans than WT, but sis2Δ was by far the longest lived (Table S2). Despite several observations that deletion of SIS2 extends lifespan (McCormick et al., 2015; Ölmez et al., 2023; Smith et al., 2008), the mechanism by which this lifespan extension occurs remains unclear, and its connection to mitochondrial function has not been reported. The lifespan extension of the sis2Δ mutant was additive with one of the most well-characterized lifespan-extending deletions, fob1Δ, suggesting that the mechanism is not related to the extrachromosomal ribosomal DNA circle–mediated mechanism of lifespan determination (Sinclair and Guarente 1997) (Fig. 2, A and B). The MMP ratio trajectory of the sis2Δ mutant declined only slightly throughout the lifespan of the cell and leveled off at a value higher than WT (Fig. 2 C, orange), while the fob1Δ mutant had no effect on the MMP ratio (Fig. 2 C, purple). The mitochondrial oxygen consumption rate (OCR) was also elevated in sis2Δ (Fig. 2 D), indicating that increased MMP in this mutant was associated with more active mitochondria (Schmidt et al., 2021; Nicholls 2004).

Sis2p has been implicated in performing two distinct functions. It is a subunit of the heterotrimeric phosphopantothenoylcysteine decarboxylase complex involved in the biosynthesis of coenzyme A (Ruiz et al., 2009), and is an inhibitor of the serine/threonine phosphatase Ppz1p (de Nadal et al. 1998). We investigated the role of Sis2p in coenzyme A biosynthesis, as this has been proposed as a mechanism for the observed lifespan extension (Ölmez et al., 2023). When SIS2 was deleted, targeted metabolomics showed no change in the amount of coenzyme A, the coenzyme A precursor pantothenate, or the coenzyme A product acetyl-CoA (Fig. S4, A and B; Table S3). This was consistent with a recent report indicating no change in these metabolites in the sis2Δ mutant (Ölmez et al., 2023).

On the other hand, we did find evidence that the observed MMP and lifespan phenotypes of sis2Δ were related to a loss of Ppz1p inhibition. The overexpression of PPZ1 results in hyperpolarization of the plasma membrane due to reduced potassium uptake (Ariño et al., 2010; Navarrete et al., 2010; Kahm et al., 2012; Herrera et al., 2014). This causes sensitivity to toxic cations such as lithium. Consistent with this idea, sis2Δ had a growth defect on 300 mM LiCl, though it was not as severe as a deletion of the gene encoding the high-affinity potassium transporter, TRK1 (Fig. S4 C). In addition, we confirmed previous observations (Ölmez et al., 2023) that the ppz1Δ mutant had a shorter lifespan than WT and prevented the lifespan extension of sis2Δ (Fig. 2, E and F). Deleting PPZ1 also caused a reduction in young cell MMP compared with WT and prevented the MMP increase of sis2Δ, as measured by TMRM (Fig. 2 G). Taken together, these data suggested that one of the ways Sis2p was impacting MMP and lifespan was via its regulation of Ppz1p. Specifically, we hypothesized that deleting SIS2 resulted in increased activity of Ppz1p.

One of the better characterized functions of Ppz1p is the destabilization of nutrient and ion transporters at the plasma membrane (Ariño et al., 2010; Lee et al., 2019). The Npr/Hal-family kinases Sat4p (aka Hal4p) and Hal5p (Antunes and Sá-Correia 2022) phosphorylate and stabilize many of the same targets that Ppz1p dephosphorylates and destabilizes (Mulet et al., 1999; Pérez-Valle et al., 2007; Tumolo et al., 2020). Thus, if the increased phosphatase activity of Ppz1p was required for the increased MMP and lifespan of a sis2Δ mutant, then removing one or more of these antagonistic kinases should also increase MMP and lifespan. (We did not obtain information about sat4Δ or hal5Δ mutants from our screen, as they were underrepresented in the screening pool of deletion mutants.) Consistent with this hypothesis, we found that deleting SAT4 or HAL5 increased MMP (P = 2.5 × 10⁻¹³ and 8 × 10⁻⁷, respectively) and extended lifespan (Fig. 3, A and B). For both gene deletions, increased MMP and extended lifespan were additive with sis2Δ (Fig. 3 C, P = 1.2 × 10⁻¹³ for sis2Δ vs sis2Δ sat4Δ; p = 1 × 10⁻⁶ for sis2Δ vs sis2Δ hal5Δ). On the other hand, it has been shown that overexpressing Ppz1p can be toxic to the cell (Makanae et al., 2013; Casamayor and Ariño 2022), suggesting that unbalanced kinase/phosphatase activity favoring dephosphorylation by Ppz1p can be detrimental to the cell. Consistent with this, we and others (Pérez-Valle et al., 2010) found that the sat4Δ hal5Δ double mutant grew slowly (Fig. 3 D), and we showed that it had a short lifespan (Fig. 3, A and B). The results were even more extreme for the sis2Δ sat4Δ hal5Δ triple mutant (Fig. 3, A–D). Intriguingly, both the sat4Δ hal5Δ mutant and the sis2Δ sat4Δ hal5Δ mutant displayed an MMP higher than sat4Δ or hal5Δ alone (Fig. 3 C; all p < 2 × 10⁻¹³ for these comparisons). Collectively, these results supported the hypothesis that the target protein or proteins responsible for modulating MMP and lifespan were shared between the kinases Sat4p and Hal5p and the phosphatase Ppz1p, and were thus likely to be nutrient or ion transporters.

One of the most well-characterized targets of Ppz1p that is also a target of Sat4p and Hal5p is the high-affinity potassium transporter, Trk1p (Ariño et al. 2010; Ariño et al. 2019; Yenush 2016). Since deletion of TRK1 and potassium limitation have been shown to extend lifespan (Sasikumar et al., 2019), we speculated that sis2Δ could be acting as a TRK1 hypomorphic allele. Consistent with this hypothesis, several sis2Δ phenotypes were intermediate between WT and trk1Δ, including lower internal potassium (Fig. 4 A), increased plasma membrane potential (Fig. 4 B and Fig. S4 C), and increased MMP (Fig. 4 C). In all cases, the phenotypes of the sis2Δ trk1Δ double mutant were not significantly different from the trk1Δ mutant (Fig. 4, A–C), suggesting that for these phenotypes, SIS2 and TRK1 are in the same pathway. Deleting SIS2 increased lifespan more than deleting TRK1, while the lifespan of the sis2Δ trk1Δ double mutant was equivalent to the trk1Δ mutant (Fig. 4, D and E). This is also consistent with sis2Δ acting as a TRK1 hypomorphic allele, as it suggests that reducing Trk1p activity (by deleting SIS2 only) is more beneficial to lifespan than eliminating Trk1p activity completely (by deleting TRK1).

Since Ppz1p has other targets in addition to Trk1p (Velázquez et al., 2020), we tested whether potassium limitation was sufficient to increase MMP and extend lifespan. We grew WT and sis2Δ strains in media with different concentrations of potassium and measured their MMP ratios and lifespans (Fig. 5, dark colors and solid lines). The MMP ratio and lifespan of the WT and sis2Δ strains increased as external potassium was reduced (Fig. 5 A, dark colors and solid lines). At ∼0.5 mM external potassium, WT and sis2Δ had equivalent MMP ratios. Further reduction of external potassium to 0.1 mM caused a small decline in the MMP ratio (Fig. 5 A, dark colors and solid lines). Interestingly, both MMP_s and MMP_i increased as external potassium was reduced (Fig. 5, C and D, dark colors and solid lines). The reduction of the MMP ratio at very low levels of potassium reflected a greater increase in MMP_i relative to MMP_s.

Lifespan dramatically declined at 0.1 mM external potassium (Fig. 5 B, dark colors and solid lines). While this decrease in lifespan brought the sis2Δ strain below its lifespan in media with standard potassium, the lifespan of WT was still higher than its lifespan in standard potassium (Fig. 5 B, dark colors and solid lines).

While performing the initial experiments in ammonium sulfate, we observed a dramatic increase in culture doubling time as external potassium was reduced, and absolutely no growth below 0.1 mM external potassium (Fig. S4 D, dark colors and solid lines). Slower culture growth in limiting potassium has been attributed to toxicity associated with excessive ammonium uptake (Hess et al., 2006; Rodríguez-Navarro and Ramos 1984) because ammonium, the main nitrogen source in standard media, competes with potassium for uptake by Trk1p (Conway and Duggan 1958). To remove this effect, we repeated potassium limitation experiments with arginine in place of ammonium as the primary nitrogen source (Fig. 5, light colors and broken lines). Below standard medium potassium concentrations, cultures doubled more quickly in arginine than in ammonium, and cultures in arginine even grew in 0.01 mM potassium (Fig. S4 D). However, culture doubling times increased even with arginine as a nitrogen source, demonstrating a growth defect in extreme potassium limitation (Fig. S4 D).

While arginine had different effects on both the MMP ratio (Fig. 5 A, light colors and broken lines) and lifespan (Fig. 5 B, light colors and broken lines) than ammonium, growth in limiting potassium was sufficient for the WT strain to phenocopy the high MMP ratio and long lifespan of a sis2Δ mutant grown at the standard potassium concentration regardless of the nitrogen source. As in ammonium, growth in arginine and limiting potassium resulted in a large increase in mitochondrial mass (as measured by MMP_i) to support increased MMP_s import (Fig. 5, C and D).

The lifespan extension of potassium limitation has been attributed to increased acidification of the yeast lysosome-like vacuole (Sasikumar et al., 2019). We checked vacuolar acidification using a fluorescence protein–based, ratiometric, genomically integrated pH sensor specifically designed to report on vacuolar pH (Okreglak et al., 2023), and found that in fact, the vacuole generally became less acidic (more alkaline) when reducing external potassium below standard medium levels (Fig. S4). In our conditions, there was no difference between WT and sis2Δ vacuolar acidity at 1 mM external potassium (Fig. S4 E), despite large differences in the MMP ratio (Fig. 5 A) and lifespan (Fig. 5 B). Thus, we sought other explanations for the observed effects on MMP and lifespan.

The results so far demonstrated that changing external potassium concentration led to changes in MMP. Since in yeast (Kahm et al., 2012; Herrera et al., 2014; Masaryk and Sychrová 2022) and mammalian cells (Patrick 1978; Relman et al., 1961), internal and external concentrations of potassium are correlated, we hypothesized that the internal concentration of potassium might, directly or indirectly, determine MMP. We therefore measured the internal potassium of WT and sis2Δ strains grown in different potassium concentrations, in either ammonium or arginine. We confirmed that in both strains, lower external potassium resulted in lower internal potassium, regardless of nitrogen source (Fig. 6 A). Consistent with the idea that ammonium competes with potassium uptake, the decline in internal potassium was greater for cells grown in ammonium (Fig. 6 A). There was a general anticorrelation between MMP_s and external potassium (Fig. 5 C; Pearson’s r = −0.67, P = 0.001), but to a different degree depending on nitrogen environment and genotype (Fig. 5 C). In contrast, there was a very strong anticorrelation between MMP_s and internal potassium, which held across nitrogen environments and genotypes (r = −0.98, p < 4 × 10⁻⁹; Fig. 6 B). Increasing external potassium to 200 mM did increase internal potassium in both WT and sis2Δ strains (Fig. S5 A), but this had a negligible effect on MMP_s (Fig. S5 B) or lifespan (Fig. S5 C), suggesting some other processes, such as vacuolar sequestration (Herrera et al., 2013) or alternative feedback mechanisms, alter this relationship when external potassium is high. We also found that aged WT cells had a lower concentration of potassium than young cells (Fig. S5 E). Thus, for young cells growing in standard to limiting amounts of potassium, internal potassium determined MMP_s import capacity, regardless of genotype or nitrogen source.

In young WT cells grown in arginine as a nitrogen source, potassium-dependent increases in MMP correlated very well with lifespan (Fig 6 C). In sis2Δ cells, the lifespan was greater than WT at all MMP_s levels, suggesting that Sis2p increases lifespan by additional pathways. To investigate why sis2Δ cells have longer lifespans than WT in all external potassium concentrations (Fig. 5 B), we reasoned that genes whose expression was significantly higher or lower than WT in 7.35 and 0.5 mM KCl would be good candidates for the observed MMP- and potassium-independent lifespan extension of sis2Δ. As observed previously (Ölmez et al., 2023), the expression of most ETC transcripts was elevated in sis2Δ compared with WT in 7.35 mM KCl (Fig. S5 F). However, in 0.5 mM KCl, the expression of ETC transcripts in WT cells increased, such that they were not significantly different between WT and sis2Δ, with the exception of COX6 (Fig. S5 G). Thus, the expression of ETC components was unlikely to account for the extended lifespan of sis2Δ in low potassium. Overall, the expression of 61 genes was differentially expressed between sis2Δ and WT in both environments (Table S4). Among these, only the set of 22 genes lower in both conditions had any significant functional enrichments, and they were all related to amino acid biogenesis (Fig. S5 H). Consistent with this, of the 23 metabolites found to be significantly lower in sis2Δ vs WT, 13 were amino acids or amino acid intermediates (Table S3). This altered metabolism may contribute to, or reflect processes involved in, its longer lifespan in all potassium concentrations.

To determine whether potassium ion itself was directly responsible for altering MMP, we isolated mitochondria from cells and monitored TMRM in different conditions (Scaduto and Grotyohann 1999). We titrated the amount of KCl in the incubation buffer from 40 to 250 mM (Fig. 7 A), corresponding to the range of values observed in vivo (Fig. 6 B). Electron transport was stimulated by the addition of the metabolic substrates glutamate/malate, succinate, and glycerol-3-phosphate (Fraenkel 2011). MMP declined with increasing KCl in the buffer (Fig. 7 A, green), suggesting that the KCl concentration had a direct effect on MMP. This effect persisted after adding ADP to stimulate complex V, demonstrating that the isolated mitochondria were fully functional and capable of oxidative phosphorylation at all KCl concentrations (Fig. 7 A, orange) (Nicholls 2004). The addition of carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), which effectively makes the mitochondrial inner membrane permeable to protons (Nicholls 2013), reduced MMP to an equivalent level across all KCl concentrations (Fig. 7 A, purple), providing further evidence that we were monitoring MMP. These results were also seen for the sis2Δ mutant (Fig. S5 I). The effect of KCl on MMP was rapid and reversible: mitochondria preincubated in 250 mM KCl for 15 min and then diluted into a buffer containing 58 mM KCl had higher MMP than the same mitochondria diluted into buffer containing 250 mM KCl (Fig. 7 B).

We tested whether the effect of KCl on MMP was specific to potassium by incubating isolated mitochondria in 50 mM KCl plus either 200 mM choline chloride (Fig. 7 C, purple) or 200 mM NaCl (Fig. S5 J, purple) and found both salts reduced MMP as much as the 250 mM KCl condition (Fig. 7 C, orange; and Fig. S5 J, orange). This was not due to changes in osmolarity, as adding 400 mM sorbitol to 50 mM KCl (Fig. 7 C, pink) did not change MMP relative to 50 mM KCl alone (Fig. 7 C, green). Thus, MMP was sensitive to the ionic strength of the buffer, not the particular ion or the osmolarity of the buffer.

The effect of ionic strength was observed for each of the ETC substrates and appeared to affect the substrates in a similar way (Fig. 7 D; P >0.2 for all interactions between substrate and KCl concentration), indicating that the effect of ionic strength exerted its influence downstream of the dehydrogenases (e.g., on complex III, cytochrome c, or complex IV) (Rosenfeld and Beauvoit 2003; Fraenkel 2011), or had a more general effect on the inner membrane, such as increased inner membrane permeability to protons (Campbell et al., 1975; Lambert and Merry 2004). Increased membrane permeability would lead to increased respiration (Nicholls 2004; Schmidt et al., 2021). However, the OCR of isolated mitochondria declined with increasing ionic strength (Fig. 7 E). This ruled out the membrane leak hypothesis and instead suggested that ionic strength interferes with overall flux through the ETC. Consistent with this hypothesis, we found that in intact, isolated mitochondria, the activity of cytochrome c oxidase (complex IV) decreased with increasing ionic strength, using either KCl or NaCl (Fig. 7 F).

Taken together, these experiments suggested that, in vivo, internal potassium levels directly affect MMP by changing the ionic strength of the cytoplasm. These changes to MMP may then contribute to overall cell health as measured by lifespan.

In this study, we combined in vivo MMP sensors with technologies we recently developed (Thayer et al., 2022, Preprint; Hendrickson et al., 2018) to follow the decline in MMP in aging S. cerevisiae cells (Hughes and Gottschling 2012). This revealed that young cells with higher MMP are more likely to have a longer lifespan than their siblings with lower MMP (Fig. 1 E). This suggests that early mitochondrial health can have a persistent impact on overall cellular health, and complements the observation made in worms that artificially boosting MMP extends lifespan (Berry et al., 2023). It is worth noting that respiratory requirements of yeast cells appear to increase with age (Leupold et al., 2019). This age-associated change in respiration is similar to the observation that the respiratory capacity of CD8⁺ T cells increases as they age, though a mechanistic understanding of this change is unclear in T cells (Quinn et al., 2020).

Using the tools we developed in a genome-wide screen (Fig. S3 A), we identified gene deletion alleles that delayed or abrogated the age-associated decline in MMP (Table S2). We further characterized the allele that provided the greatest extension of lifespan, sis2Δ (Table S2). We found that the sis2Δ mutant appeared to act through its role as an inhibitor of Ppz1p activity (Fig. 2, E–G). One role of Ppz1p is destabilizing nutrient transporters through its phosphatase activity (Ariño et al., 2010; Lee et al., 2019). Consistent with this, we found that deleting the kinases SAT4 and HAL5, which stabilize nutrient transporters (Mulet et al., 1999; Pérez-Valle et al., 2007; Tumolo et al., 2020), had the same effect as deleting SIS2 (Fig. 3). We hypothesized that changes in MMP and lifespan are modulated by the activity of the high-affinity potassium transporter, Trk1p, since it is a known target of Ppz1p, Sat4p, and Hal5p (Mulet et al., 1999; Pérez-Valle et al., 2007; Tumolo et al., 2020). Consistent with this hypothesis, deleting TRK1 extended lifespan (Sasikumar et al., 2019), and increased MMP (Fig. 4 C).

Synthesizing these data led us to a model (Fig. S5 K), whereby increasing the ability of Ppz1p to phosphorylate the high-affinity potassium transporter, Trk1p, by eliminating either its inhibitor, Sis2p, or the kinases Sat4p and/or Hal5p, reduced the activity of Trk1p, leading to lower internal potassium, higher MMP, and longer lifespan. Indeed, reducing environmental potassium was sufficient to extend lifespan (as observed before [Sasikumar et al., 2019]), and increase the MMP of a WT strain to the same levels as a sis2Δ mutant grown in a standard potassium environment (Fig. 5, A and B). Interestingly, sis2Δ was longer-lived at all external potassium concentrations, suggesting that deleting SIS2 has other lifespan-extending effects in addition to limiting internal potassium. Transcriptional (Table S4) and metabolic (Table S3) evidence suggests this may be related to a chronic reduction in amino acid levels, which has been shown to extend lifespan in yeast, flies, and mammals (Orentreich et al., 1993; Johnson and Johnson 2014; Grandison et al., 2009; Fulton et al., 2024). Although vacuolar pH did not appear to explain the young cell phenotypic differences between WT and sis2Δ (Fig. S4 E), we note that vacuolar pH plays a role in aging more generally (Henderson et al., 2014; Antenor et al., 2025), and could further affect aging phenotypes.

Nitrogen source and genotype affected potassium homeostasis by modulating the relationship between external and internal potassium, but, ultimately, MMP was determined by the concentration of internal cellular potassium (Fig. 6 B). Using isolated mitochondria, we determined that this was a direct effect of potassium through its influence on the ionic strength of the mitochondrial environment (Fig. 7). Because ionic strength is determined by charge density and not the identity of the ion, in theory any cation could have a similar effect on mitochondrial function. However, since potassium is the most abundant cation in the cell by an order of magnitude (Satlin 2009; Mulkidjanian et al., 2012), it largely determines the ionic strength of the cytoplasm and the mitochondrial inner membrane space to which the mitochondrial inner membrane is exposed (Cortese et al., 1991).

Mechanistically, our experiments appear to rule out an effect of ionic strength on the ATP synthase (complex V) (Fig. 7 A), the dehydrogenases before complex IV (Fig. 7 D), and changes in membrane permeability (Fig. 7 E). Rather, we provide evidence that complex IV activity is sensitive to ionic strength (Fig. 7 F). This mirrors results in other systems (from beef heart and rat liver), where complex IV activity has been shown to be sensitive to ionic strength, both by directly modulating its interaction with (Wilms et al., 1980; Wilms et al., 1981) and by altering the diffusivity of (Gupte and Hackenbrock 1988) cytochrome c. As we show here, we can attribute reductions in MMP above ∼120 mM potassium to reduced activity of complex IV, but the further increase of MMP observed below 120 mM may be due to the modulation of another factor, or a limitation related to determining complex IV activity in isolated mitochondria.

Potassium is the most abundant intracellular cation across all domains of life (Stautz et al., 2021; Ariño et al., 2010; Satlin 2009; Vašák and Schnabl 2016), and is maintained at intracellular concentrations in the hundreds of millimolar—much higher than any other cation (Satlin 2009; Mulkidjanian et al., 2012). This reflects the highly conserved role of potassium as a critical determinant of fundamental cellular attributes, including cell volume, cytoplasmic pH, and plasma membrane potential (Ariño et al., 2010). Here, we suggest that MMP should be added to this list.

Cells maintain internal potassium within a relatively narrow range, but this range is correlated with external potassium in both yeast (Fig. 6 A, [Kahm et al., 2012; Herrera et al., 2014; Masaryk and Sychrová 2022]) and mammalian cells (Patrick 1978; Relman et al., 1961; Sterns et al., 1981). Regulation of intracellular potassium levels in an environment of fluctuating extracellular potassium is certainly relevant to microorganisms, but it is also important in mammalian tissues, such as the gut epithelium, the liver, and the kidney, which are exposed to widely varying levels of potassium (McDonough and Youn 2017; Satlin 2009). Cells of the immune system are also exposed to a wide variety of extracellular potassium concentrations, with potentially negative consequences. For instance, the high potassium tumor microenvironment can lead to suppression of T cell effector functions and their ability to differentiate (Vodnala et al., 2019). Intriguingly, this is associated with reduced MMP (Eil et al., 2016).

Loss of homeostatic balance in cellular systems is a principal characteristic of aging in every organism (Pomatto and Davies 2017). In humans, the ability to maintain potassium homeostasis declines with age, leading to a total body reduction of potassium content (as measured by isotope labeling), thought to be mostly due to loss of muscle mass (Andreucci et al., 1996; Cox and Shalaby 1981). Intracellularly, red blood cell potassium content also declines, but it is not known whether this is true of other cell types (Andreucci et al., 1996). Clearly, the cause-and-effect relationships between aging, potassium homeostasis, and mitochondrial function have not been worked out. More generally, failure to maintain ion homeostasis can lead to aging-related diseases, including channelopathies in cancer (Pi et al., 2007; Prevarskaya et al., 2018), defects in insulin signaling (De Marchi et al., 2021), Alzheimer’s disease (Song et al., 2024), and Parkinson’s disease (Zhuang et al., 2024).

In this study, we revealed a crucial role for potassium ion homeostasis in controlling and maintaining mitochondrial function as cells age, and we show that one way to intervene is through manipulation of the activity of potassium transport. There are a wide variety of small molecules that can be used to manipulate the activity of mammalian mitochondrial potassium channels, but their effects on membrane potential and longevity have not been investigated (Wrzosek et al., 2020). Based on our results, we believe this approach could provide new potential avenues for slowing the age-associated decline in mitochondrial function and, as a consequence, slow the onset of aging-related disease.

Reporter strains were derived from BY4700 (MATa ura3Δ0) (Baker et al. 1998). Nonreporter strains were derived from a cross between BY4707 (MATα met15Δ0) and BY4717 (MATα ade2Δ0::hisG). We found that BY4707 and BY4717 contain a null allele of PTR3 due to insertion of a TY element. Since ptr3^- alleles extend lifespan (Tsang et al., 2015) and are synthetic-lethal with leu^-, we corrected this defect in the following way. A prototrophic strain derived from a cross between BY4707 and BY4717 (CGY2.66) was crossed with BY4713 (MATα leu2Δ0), and a leu^- haploid was selected. A prototrophic, PTR3⁺LEU⁺ strain (CGY4.50) was generated by PCR amplification of full-length LEU and transformation followed by selection on plates lacking leucine. The genotype was confirmed by whole-genome sequencing.

To construct the reporter strain, a plasmid containing the mCherry coding sequence and natMX selection cassette (CGB4.09) was PCR-amplified using primers with homology to the C-terminal region of TIM50 and introduced into BY4700 by transformation. ClonNat-resistant colonies were selected and sequence-verified. The resulting strain was transformed with NotI-digested plasmid containing pTPI-preCOX4-SL17-mNeonGreen-URA3 (CGB2.36), which has homology to a gene-free region of chromosome I (Hughes and Gottschling 2012). The sequence of the preCOX4 leader was 5′-ATG​CTT​TCA​CTA​CGT​CAA​TCT​ATA​AGA​TTT​TTC​AAG​CCA​GCC​ACA​AGA​ACT​TTG​TGT​AGC​TCT​AG-3′. The sequence of the SL17 degron was 5′-TCG​ATT​AGT​TTC​GTA​ATA​CGT​TCA​CAT​GCT​TCG​ATT​AGA​ATG​GGG​GCA​TCC​AAT​GAT​TTT​TTC​CAC​AAG​TTG​TAC​TTC​ACG​AAA​TGC​CTC​ACA​TCA​GTT​ATT​TTA​TCT​AAA​TTT​CTG​ATC-3′.

We confirmed that the expression of both TPI1 and TIM50 did not change after 24 h of aging (log₂[old/young] = −0.07, P = 0.48; and log₂[old/young] = −0.1, P = 0.12, respectively; Table S1) and that reporter expression had a negligible effect on lifespan (mean lifespan 25.6 divisions without reporter vs 24.2 divisions with reporter, P = 0.003; Fig. S2 C).

Gene deletions were made using standard techniques, either by amplifying the indicated drug resistance marker from a plasmid with primers containing homology to the 5′- and 3′-regions immediately surrounding the ORF, or by amplifying the deletion cassette containing kanamycin resistance out of the library strain. Gene deletions were confirmed using PCR, and all strains were checked for functional mitochondria by growth on 2% agar plates containing 1.0% yeast extract, 2.0% peptone, 2.0% ethanol, and 3.0% glycerol. In general, at least two independent clones were analyzed. A full list of strains can be found in Table S5, and oligos can be found in Table S6.

Unless indicated otherwise, cells for all assays were prepared in the following way. Strains to be assayed were streaked from a −80°C freezer onto 2% agar plates containing 1% yeast extract, 2% peptone, 2% glucose and grown until colonies were visible (typically 2–3 days). Cells were inoculated into synthetic complete (SC) medium (2 g 10X SC mix [Sunrise], 6.7 g yeast nitrogen base (YNB) containing ammonium sulfate [Sunrise], 20 g glucose per liter, buffered to pH 4.5, filter-sterilized) and grown for 16–24 h to saturation. These “overnight” cultures were then diluted into the experimental media and grown for 24 h in the exponential phase, diluting as necessary so that the cell culture never exceeded an OD₆₀₀ of 0.1. For potassium titration experiments, we used a custom formulation of “Translucent K+ free medium” (Formedium), with 960 mg/l (8 mM) sodium dihydrogen phosphate instead of 920 mg/l (8 mM) ammonium dihydrogen phosphate. To this, either 5 g/l (38 mM) ammonium sulfate or 2.1 g/l (10 mM) arginine HCl was added as a nitrogen source.

Mitochondria were isolated by the fractionation and differential centrifugation method (Albers et al., 2006; Liao et al., 2018). Specifically, cells were inoculated into SC media and grown for 16–24 h to saturation (OD₆₀₀ ∼4). The OD₆₀₀ of the culture was measured just prior to sampling. 400 OD₆₀₀ of cells were collected on a filter and washed with water. Cells were resuspended in 35 ml water and spun down at 1,500 g for 10 min, and the supernatant was removed. Cells were resuspended in 20 ml cell wall destabilization buffer (100 mM Tris-HCl, pH 9.5, 50 mM β-mercaptoethanol) and incubated at 30°C for 15 min. Cells were then washed with 20 ml sorbitol buffer (1.2 M sorbitol, 5 mM MgCl₂, 20 mM potassium phosphate, pH 7.5) and digested in 4 ml sorbitol buffer with 12 μl 200 mg/ml 100T zymolyase at 30°C for 90 min. Digested cells were washed with 20 ml ice-cold sorbitol buffer and resuspended in 4 ml ice-cold mitochondrial isolation buffer (0.6 M sorbitol, 20 mM HEPES, pH 7.5) to which protease inhibitor (cOmplete protease inhibitor cocktail, Roche) was added. Cells were lysed with a Dounce homogenizer using 15 strokes with the tight pestle (Kimble Kontes 2 ml Dounce Tissue Grinder, DWK Life Sciences). Lysed cells were collected, and 4 ml ice-cold mitochondrial isolation buffer with protease inhibitor was added. Lysed cells were spun down at 3,000 g for 5 min. The supernatant was collected and spun again at 12,000 g for 10 min. The supernatant was discarded, and the mitochondrial pellet was resuspended in ice-cold mitochondrial isolation buffer with protease inhibitor. The mitochondrial concentration was quantified using a Pierce BCA protein assay (Thermo Fisher Scientific). All assays on isolated mitochondria were performed in mitochondrial assay solution (MAS) containing 220 mM mannitol, 70 mM sucrose, 10 mM NaH₂PO₄ (dihydrate), 5 mM MgCl₂, 2 mM HEPES, 1 mM EGTA, 0.2% (w/v) fatty acid–free BSA.

We crossed the strain containing the MMP reporter system into the yeast knockout collection (Winzeler et al., 1999), in which each of the ∼4,700 nonessential genes has been replaced with a unique barcode. This deletion library was pooled and aged using MADs (Hendrickson et al., 2018) for 24 h, and the protocol was adapted as follows. Cells were grown for 24 h at low OD₆₀₀ (<0.05) in SC medium containing 2% methyl-alpha-D-mannopyranoside (MP, Sigma-Aldrich) at pH 4.5. The presence of MP in the media prevents age-related aggregation of cells in the MAD (Hotz et al., 2022), similar to mannose used in the original protocol. For cell beading, one tube of Super Mag NHS-activated 100-nm beads (Ocean NanoTech) was equilibrated to room temperature for 30 min and then resuspended in 400 μl resuspension buffer (25 mM MES, pH 6.0, 0.01% Tween-20). Then, beads were vortexed at 1,600 rpm until needed. Four ODs of cells per MAD were harvested, washed extensively in labeling medium (YNB, 2% glucose, 2% MP, pH 7.3), and resuspended in 400 μl labeling medium. Magnetic beads were placed on a magnet, resuspension buffer was taken off, 200 μl of labeling medium was added, and 40 μl of beads was immediately mixed with each aliquot of resuspended cells. 600 μl PEG-3000 was added, and the tubes were incubated at room temperature on a rotating wheel for 15 min. Then, cells were washed twice on a magnet with growth medium and the quality and uniformity of beading were checked under a microscope.

MADs were loaded in triplicate and run for 24 h (“T24”). After harvesting, cells were immediately fixed in 100 mM Tris base, 20 mM sodium azide, 20 mM sodium fluoride, adjusted to pH 7.6 (TAF [Olmo and Grote 2010]) buffer to prevent further changes to the MMP ratio, and then washed several times on magnets to remove daughter cells. As a control, we used mother cells from an unaged population, which was bead-labeled as above and harvested after 3 h (“T3”) of growth in SC medium. The resulting aged and unaged samples were stained with wheat germ agglutinin (WGA) conjugated to a 647-nm Alexa Fluor (Thermo Fisher Scientific; WGA647) to mark bud scars, and SYTOX Blue (Thermo Fisher Scientific) to mark dead cells. These enriched, stained populations were sorted on a BD FACSAria (BD Biosciences). To further purify the mother cell population and remove dead cells, only cells with high WGA, low SYTOX Blue signals were considered for analysis (Fig. S3 B). From the resulting live mother population, the distribution of the mNeonGreen/mCherry ratio was divided into “low,” “medium,” and “high” populations, and 10⁵–10⁶ cells were collected for each subpopulation. In total, 10 sets of populations were collected for young and old mother cells.

From these populations, barcode libraries were generated in 2 or 3 technical replicates as follows. Genomic DNA was extracted using YeaStar Genomic DNA Kit (Zymo Research), and DNA yield was quantified using the Qubit dsDNA High Sensitivity kit (Invitrogen). Barcodes were amplified by PCR using 10-ng genomic DNA and custom oligos containing Illumina indices (98°C for 1 min; 15 cycles of 98°C for 20 s, 66°C for 30 s, 72°C for 30 s; 72°C for 2 min). The PCR product was purified with RNAClean XP beads (Beckman Coulter) and run on Agilent Bioanalyzer for quantification and quality control. Samples were pooled in equal concentrations and sequenced on an Illumina HiSeq 2000.

The entire screen was performed twice with slight differences in the protocol. In the first experiment, the low, medium, and high populations were defined as being below the T3 mode, above the T3 mode, and the top 25% of the T3 population. In the second experiment, each population distribution was split into thirds to make the low, medium, and high groups. The method of sorting had very little influence on the results, as shown by the fact that both the mean counts (Fig. S3 D) and significant log fold changes (Fig. S3 E) were highly correlated between the two experiments. We used the second experiment for further analysis.

To identify barcodes with high recovery and high accuracy, we developed an algorithm that utilized information from both the sense and antisense strands of the paired-end reads. This algorithm takes advantage of the additional information available from paired-end reads. First, the common sequence outside of the (variable) barcode is identified in the sense and antisense direction using regular expression matching and allowing up to two mismatches on either side of the barcode. This allowed us to identify 98% of reads in at least one direction. Once the common sequence was identified, the associated barcode was extracted. If the barcode was present in both directions and the sequences were identical, or if the barcode was only identified in the sense or antisense direction, we used that sequence as the barcode. If the barcode was present in the sense and antisense direction, but the barcode sequences did not agree, we chose the bases with the highest associated quality score to create the barcode. Using this technique, we recovered a barcode in at least one direction from >87% of sequences.

Next, the cosine similarity of each extracted barcode to the set of known barcodes in the collection was calculated, and the highest scoring member from the barcode collection was selected as the match. Thus 300000, the cosine similarity score was also the confidence in the match. Using this approach, we did not have to fix a number of mismatches, insertions, or deletions to be considered a match. This approach also had the advantage of correctly identifying barcodes even if there were errors in the published barcode sequences associated with the collection. Empirically, we determined a minimum cosine similarity score of 0.8 (with 1 being a perfect match) was a conservative threshold, and this included 96% of the extracted barcodes.

Differential enrichment of barcodes in the high vs low mNeonGreen/mCherry ratio groups was analyzed using the R package “DESeq2” (Love et al., 2014). Log fold change estimates were shrunk using the R package “apeglm” (Zhu et al., 2019).

Unless otherwise indicated, statistical modeling was performed using the functions “lm” from base R (R Core Team 2024) for fixed-effect models, “lmer” from the “lme4” package for mixed-effect models (Bates et al., 2015), or “coxph” from the “survival” package (Therneau and Grambsch 2000) for lifespan comparisons. The specific formulas used for each model are referenced in their corresponding sections below.

Point estimates, contrasts, and P-values (all adjusted for multiple hypothesis testing) were estimated using the R package “emmeans” (Lenth 2024), using ordinary marginal means and the default P-value adjustment method “tukey.” Correlations were calculated using the R function “cor.test” in base R. Datasets whose natural range was only positive values (e.g., sizes, fluorescence intensities, concentrations) were log-transformed before analyzing to improve normality and satisfy the assumptions of linear modeling (i.e., that data span the range of negative to positive infinity). Other datasets were assumed to be normally distributed, but this was not formally tested.

The YLMs were operated according to the freely available documentation (Thayer et al., 2022, Preprint). All experiments were performed at 30°C. Bright-field images were taken every 15 min. Remade mutants from the screen were run on a DMi8 (Leica) microscope, and fluorescent images were taken every 2 h using a SpectraX (Lumencore) light source, a “GFP” filter cube (470/40-nm excitation, 495-nm dichroic, 525/50-nm emission), and a “Texas Red” filter cube (560/40-nm excitation, 585-nm dichroic, 630/75-nm emission) for mNeonGreen and mCherry, respectively, a 40× air objective (Leica, 0.95 NA), and a pco.edge 4.2 CLHS sCMOS camera (Excelitas). Each gene deletion was represented by at least two clones, and each clone was analyzed in at least two microfluidic channels within one experiment. Some clone replicates were lost due to channel clogging or other technical issues. The WT strain was included in every experiment as a reference.

Subsequent lifespans were acquired on custom-made microscopes (Thayer et al., 2022, Preprint) equipped with 40× air objectives (Nikon, 0.95 NA) and pco.panda 4.2 sCMOS (Excelitas) cameras, and using microfluidics devices with “DetecDiv”-style catchers (Aspert et al., 2022). No fluorescence illumination was used. Every clone was analyzed in at least two channels within one experiment, and at least two experiments were performed. Experiments were run for 99 h. Channels that were observed to clog with cells before 66 h were excluded from further analysis.

Fluorescence data were extracted using custom software. Raw trajectory data were filtered to only include trajectories that started at time 0, stopped before time point 300 (to avoid false event calls at the end of the run), and had MMP_s and MMP_i signals >50. Raw trajectory data are available in Table S7.

MMP ratio comparisons with WT after 24 h were performed by extracting MMP ratio data from cells that were between 21.5 and 26.5 h old. The data were then fit to a statistical model with a term for the interaction between genotype and time, a random effect for the trajectory id, and a random effect for the experiment: “log10(MMP ratio) ∼ gene * time + (1|id_lifespan) + (1|experiment).” Individual clones and microfluidic channels were not explicitly modeled in order to average over their contributions.

The genomes of clones used in lifespan analysis were sequenced to ensure that rDNA copy number was not too low (≥100 copies [Hotz et al., 2022]). Bud counts were determined from a machine learning model (Thayer et al., 2022, Preprint) using publicly available code (https://github.com/calico/ylmcv). Lifespan comparisons and estimates of the median replicative lifespan were made with Cox proportional hazard (CPH) models using the "survival" package in R (Therneau and Grambsch 2000). The basic model used was “Surv(bud count, event) ∼ gene,” where “event” was a parameter describing whether the cell was observed to die in the catcher, or washed out at some point (i.e., censored from the data). The experiment and channel information was concatenated to make a unique identifier and used as the clustering parameter to account for within-channel correlations. Potassium concentration and nitrogen source type were added as interacting terms with genotype as appropriate.

Culture doubling time was calculated over the 24-h period of exponential growth preceding every experiment. These data were fit to a model of the form “log10(doubling time) ∼ gene + experiment.”

OCR was measured on a Seahorse XFe96 (Agilent) according to the manufacturer’s instructions. All temperature-sensitive steps were performed at 30°C.

Each cell sample was diluted to an OD₆₀₀ of 0.1, and 180 μl was added to the cell culture plate. The plate was then spun down for 3 min at 500 rpm. 20 μl antimycin A (AA) was added to port A at a concentration of 25 µM. Three measurements of the OCR were made. AA was injected (final concentration 2.5 µM in growth media) to inhibit complex III, and then three more measurements of OCR were made. The measurements before and after adding AA were averaged over time, replicate, and strain. The mitochondrial OCR was calculated as the difference between the averaged OCR before adding AA and the averaged OCR after adding AA.

Isolated mitochondria were added to a final concentration of 0.12 mg/ml in 180 μl MAS plus the indicated concentration of KCl. An injection solution of 5 µM AA was prepared in MAS buffer plus the appropriate concentration of KCl. After three baseline measurements, 20 μl of the injection solution was added to the well (0.5 µM final concentration) and three more measurements were made. Mitochondrial OCR was calculated as above. Data were analyzed across experiments using two linear models. The first provided point estimates at each potassium concentration by treating potassium concentration as a factor, while the second provided the intercept and slope of the line through the points by treating potassium concentration as a continuous variable. In both cases, the formula was “mitochondrial_OCR ∼ potassium + experiment,” where “potassium” was the potassium concentration, either treated as a factor or a continuous variable.

Cells were grown as described in “General media and growth conditions” in the indicated media. Cells were collected to a final OD₆₀₀ of ∼0.01, and TAF was added to stop metabolic function (Olmo and Grote 2010). SYTOX Blue (Thermo Fisher Scientific) was added for live/dead cell staining. For every sample, at least 10,000 cells were analyzed on BD LSRFortessa X-20 (BD Biosciences) using the FITC (488-nm excitation, 505 long-pass (LP), 525/50-nm emission) and PE-Texas Red (561-nm excitation, 595LP, 610/20-nm emission) filter sets to measure mNeonGreen and mCherry signals, respectively. SYTOX Blue was measured using the BV421 (408-nm excitation, 450/50-nm emission) filter set. At least three technical replicates were collected for each sample. At least two experiments were performed on different days to generate the data for each figure.

The mean values for each channel and the mean value for the ratio of the green to red channels were fit separately using the formula “log10(signal) ∼ gene + experiment,” where “experiment” was added as a factor to account for batch effects. The effects of concentration and nitrogen source were modeled as interacting terms with genotype, such that the most complicated model involving multiple genotypes, nitrogen sources, and potassium concentrations was “log10(signal) ∼ gene * concentration * nitrogen + experiment.”

TMRM staining was performed similar to (Hughes and Gottschling 2012), with some modifications. Cells were grown in the exponential phase for 24 h to an OD₆₀₀ < 0.1. Cells were sampled to a final total OD₆₀₀ of 0.02 (∼500,000 cells) into 1.5-ml tubes. To each tube, 5 μl of 50% polyethylene glycol 3,350 (PEG) was added to assist in forming a cell pellet upon spinning. Cells were spun down at 6,500 g for 1 min, the supernatant was removed, and cells were resuspended in 500 μl HEPES buffer containing 5% glucose (10 mM HEPES, 5% glucose, pH 7.6). Cells were spun at 6,500 g for 30 s, and the supernatant was removed. Cells were resuspended in 500 μl TMRM buffer (50 nM TMRM, 10 mM HEPES, 0.25% PEG) and incubated in the dark at room temperature for 15 min. Cells were spun down and resuspended in 600 μl HEPES buffer (without PEG) containing 6 μl SYTOX Blue for live/dead staining.

Stained cells were analyzed on BD LSRFortessa X-20 (BD Biosciences) using the PE channel (561-nm excitation, 586/15-nm emission). The TMRM signal decayed exponentially during the course of analysis, a result of dye export from the cell (Homolya et al., 1993); this decay rate varied depending on the mutant being analyzed. To correct for this decay, we measured each set of samples at least three times, analyzing 10,000 cells for each sample. We then calculated the exponential decay rate constant to correct the signal. Experiments were performed at least twice on different days. Results were analyzed using the corrected mean values of each sample using the formula “log10(TMRM) ∼ gene + experiment,” with experiment as a factor to account for batch effects.

For the comparison of TMRM with the MMP_s reporter in young cells (Fig. S3 F), strains containing the reporter system were grown to exponential phase, stained, and analyzed by flow cytometry as above. (TMRM is more than 10-fold brighter than the TIM50-mCherry signal, so the contribution of mCherry could be ignored.) The mean values of the corrected TMRM signal and the mNeonGreen signal (from the FITC channel) for each strain were analyzed using the model formula "log10(signal) ∼ gene + experiment." The resulting estimates of the MMP_s signal (on the original, linear scale) were adjusted in the following way. For each deletion mutant, the mitochondrial mass adjustment was estimated by taking the fractional change, d, of MMP_i compared with WT using data from the YLM experiments by averaging over cells 0–5 divisions old. Then, that mutant’s corresponding MMP_s value was multiplied by −(1 + d) to get the adjusted estimate. The Deming regression estimate was made using the R package “deming” (Therneau 2024).

Isolated mitochondria were resuspended to the indicated concentration (0.3–0.5 mg/ml) in MAS plus the indicated concentration of KCl. Each condition and replicate were paired, with TMRM added to one “signal” well to a final concentration of 200 nM, and the equivalent volume of DMSO added to the other “background” well. Unless otherwise indicated, 10 mM glutamate, 10 mM malate, 10 mM succinate, and 10 mM glycerol-3-phosphate were added as substrates. The final volume used in analysis was 100 μl. TMRM was measured ratiometrically by monitoring emission at 605 nm while exciting at 546 or 572 nm (Scaduto and Grotyohann 1999) in a CLARIOstar Plus plate reader (BMG Labtech). For each experiment, conditions were performed in triplicate, and five sequential measurements were made to improve estimates of the signal. If required, ADP was added to a final concentration of 0.5 mM, and FCCP was added to a final concentration of 15 µM. For experiments that used ADP and/or FCCP, the plate was first read without ADP or FCCP. Then, ADP (if required) was added to each well and the plate was read again. Finally, FCCP was added to each well and the plate read again. Background for each wavelength was calculated by averaging over the background wells for each condition. After subtracting the averaged background, the 572/546 ratio was calculated for each well and each time point, and then data from the individual time points were averaged to give the final result.

Analysis was performed using the model “TMRM ratio ∼ buffer * treatment + experiment,” where “buffer” indicated which buffer the mitochondria were incubated in, and “treatment” indicated which combination of ADP and FCCP had been added. For experiments where determining the slope was important, data were analyzed using two linear models. The first provided point estimates at each potassium concentration by treating the buffer as a factor, while the second provided the intercepts and slopes of the lines through the points by treating the buffer as a continuous variable. When the same well was measured multiple times in an experiment, a random effect for the well was added to account for the nonindependence of these measurements.

A strain (CGY62.43; Table S5) containing the MMP_s part of the construct was grown as described above (General media and growth conditions), and a total of 0.02 OD₆₀₀ of cells were stained with TMRM as described above (TMRM staining protocol) with the following changes. After washing with 1 ml HEPES +5% glucose buffer, cells were resuspended in the same buffer plus 5 µg/ml WGA647 and incubated in the dark at room temperature for 5 min. After another wash to remove unbound WGA647, cells were resuspended in the same buffer plus 50 nM TMRM and incubated in the dark at room temperature for 45 min. From there, 150 μl was added to a glass-bottom well (18-well µ-Slide, Ibidi USA) that had been pretreated with 5 mg/ml concanavalin A in PBS. The plate was spun at 100 g for 2 min to settle cells. The entire viewable portion of the well (∼15 × 15 fields of view) was imaged on a microscope (Leica) with a 63× oil objective (HC PL APO, 1.40 mm NA), and kept at 30°C using an Okolab incubation system. Each field of view was imaged over a 10-µm z-stack, with a 0.5-µm step size. The filters were 470-nm excitation/510-nm emission for mNeon, 550 ex./590 em. for TMRM, and 640 ex./666–724 em. for WGA. Leica Thunder background subtraction was applied to all images.

For segmentation, four-channel 3D image volumes were processed as CxZxYxX, with bright-field in channel 0, WGA in channel 1, TMRM in channel 2, and mNeon in channel 3. Voxel sizes in z, y, and x were taken from acquisition metadata and used for all anisotropy corrections and volumetric calculations. Cells were segmented in 2D from the medial bright-field z-slice using CellposeSAM on a GPU with inverted contrast and a batch size of 64 (Pachitariu et al., 2025, Preprint). Connected components with area <300 pixels were removed. Any label touching the image border was excluded. The resulting 2D label map defined cell IDs and was broadcast across z to constrain downstream operations. For WGA and mNeon, organelle structures were segmented in 3D using Nellie’s segmentation pipeline (Lefebvre et al., 2025). In brief, segmentation was performed independently within each cell footprint using a multiscale, anisotropy-aware vesselness filter, with spatial scales of 1.0 and 1.5 pixels and a z/y scale ratio set by the recorded voxel sizes. For each cell label, the distribution of positive vesselness values across z was used to compute a label-specific "Minotri" threshold (the minimum of the Otsu and Triangle thresholds (Lefebvre et al., 2025)); voxels meeting or exceeding this threshold were provisionally assigned to that label. The binary mask was cleaned with a single 3D binary closing using a spherical structuring element of radius 2 voxels and relabeled, and components smaller than 7 voxels were discarded. Surviving voxels were re-encoded by multiplying with the broadcast 2D cell label map so that every organelle voxel carried its parent cell ID. For validation, a three-plane stack per field was saved comprising the WGA mask, the mNeon mask, and the z-broadcast cell labels. Per-cell metrics were computed from the 2D bright-field labels and the 3D organelle masks. Cell planar area (pixels) was converted to an effective radius using the in-plane pixel size, and cell volume was approximated as a sphere with that radius. Median WGA intensities were measured from the WGA image at the WGA-mask voxels. Median TMRM and mNeon intensities were measured from their respective images at the mitochondrial-mask voxels. All per-cell measurements were written to CSV files, and raw volumes and label stacks were saved as TIFFs for auditing and quality control. Implementation details and exact parameter values are available in the public code repository (https://github.com/aelefebv/nellie).

Bud scars of 2,576 cells were counted by hand and matched to existing labels from the processed images. To increase the proportion of aged cells, the data were binned on the WGA signal, and more cells were chosen from higher WGA bins for bud scar counting (Fig. S2 A). After trimming the top and bottom 1 percentile of intensity values and filtering on cell size (between 39 and 235 µm³) to exclude developing cells still connected to their mother cells, the processed data from each field of view were normalized to the average value in that field to account for experiment-, position-, time-, and density-dependent differences in intensities. The resulting normalized intensities of the TMRM and mNeon signals were fit using the model formula “log10(intensity) ∼ bud_scar * fluorophore+experiment + (1|cell_id),” where “bud_scar” indicated the number of bud scars (treated as a factor), “fluorophore” indicated the TMRM or mNeon channel, and “cell_id” uniquely described each cell. A random effect for cell_id was necessary since the intensity information for TMRM and mNeon came from measurements of the same cell. Means and CIs were determined on the original (linear) scale using emmeans. Finally, each fluorophore was normalized to its maximum value across bud scars to facilitate direct comparison of signal as a function of age.

Cells were grown as above, and 1 ml of culture was sampled into a 1.5-ml tube. Di-8-ANEPPS (Invitrogen) was added to a final concentration of 1.5 µM, and 1 μl SYTOX Blue was added. Samples were incubated at room temperature in the dark for 7 min and then run on BD LSRFortessa X-20 (BD Biosciences) using the BV510 (408-nm excitation, 505LP, 525/50-nm emission) and PerCP-Cy5.5 (488-nm excitation, 635LP, 695/40-nm emission) channels. The ANEPPS ratio was calculated by dividing the PerCP-Cy5.5 signal by the BV510 signal. As with TMRM (see “TMRM assay by flow cytometry”), sample sets were set up in triplicate (or more) so that signal decay time constants could be estimated and corrected for.

Data were analyzed as described in TMRM assay by flow cytometry. The model formula used was “log10(ANEPPS ratio) ∼ gene + experiment.”

Relative vacuolar pH was measured using the vSEP ratiometric reporter construct (Okreglak et al., 2023). The plasmid containing the ratiometric vSEP construct (CGB8.15) was digested with NotI and integrated into the HO locus. After adding SYTOX Blue, samples were analyzed on BD LSRFortessa X-20 (BD Biosciences) using the FITC (488-nm excitation, 505LP, 525/50-nm emission) and PE-Texas Red (561-nm excitation, 595LP, 610/20-nm emission) filter sets to measure vSEP (“pH-sensitive”) and mCherry (“pH-insensitive”) signals, respectively. SYTOX Blue was measured using the BV421 (408-nm excitation, 450/50-nm emission) filter set.

Data were analyzed as described in TMRM assay by flow cytometry, with separate models being fit to the vSEP signal, the mCherry signal, and vSEP/mCherry ratio. The model used was “log10(vSEP ratio) ∼ gene * conc + experiment,” where “conc” was the concentration of KCl in the media.

For the young/old cell comparison, cells were grown in exponential culture or aged in MADs for 24 h as above (MMP aging screen) and harvested in RNAlater (Sigma-Aldrich). Bulk RNAseq was performed by Admera Health. Total RNA was extracted and purified using RNeasy Plant Mini Kit for RNA Extraction (Qiagen). RNA was quality-controlled and quantified using a TapeStation (Agilent) and only high-quality samples (RNA integrity number >6) were used. NEB Ultra II Directional RNA Library Prep Kit (New England Biolabs Inc.) was used to generate cDNA and sequencing libraries. Libraries were indexed using Admera’s in-house index primers and cleaned with AFTMag NGS DNA Clean Beads (ABclonal). The indexed libraries were pooled and sequenced on the Illumina NovaSeq X plus platform with a read length of 2 × 150 bp. Raw sequencing reads were demultiplexed with bcl2fastq.

For the comparison of sis2Δ and WT in standard (7.35 mM) or low (0.5 mM) KCl, strains were grown to saturation overnight in media containing standard potassium. From these overnight cultures, strains were diluted into media containing the appropriate potassium concentration in biological triplicate and grown for an additional 24 h, ensuring that the OD₆₀₀ did not go above ∼0.1. The OD₆₀₀ of each culture was measured just prior to sampling. Enough volume to obtain a total of 0.5 OD₆₀₀ of cells was applied to a vacuum filtration system onto a 0.45-µm nylon filter (HNWP02500, MilliporeSigma). Cell pellets were resuspended in 200 μl of lysis buffer (10 mM Tris [pH 8.0], 0.5% SDS, 10 mM EDTA). After adding 200 μl acid phenol (pH 4.3), samples were vortexed for 30 s. Samples were then incubated at 65°C for 30 min in a thermomixer with intermittent shaking (2,000 rpm for 1 min every 15 min). Finally, a total of 400 μl of ethanol was added and the RNA was purified using the Direct-zol RNA Miniprep Plus kit (Zymo Research) according to the manufacturer’s instructions, including the optional DNase digestion step. RNA integrity was confirmed using Agilent Bioanalyzer.

Differential gene expression was analyzed using the R package DESeq2 (Love et al., 2014). Strain and potassium condition information was concatenated and used as the covariate in the analysis. Log fold change shrinkage was performed with the “ashr” R package (Stephens, 2017). The R package “clusterProfiler” (Yu, 2024) was used to perform gene set enrichment. Enrichment was performed with the “enrichGO” function, and redundant GO terms were removed using the “simplify” function with a significance cutoff of 0.5.

Strains were inoculated in biological triplicate for each experiment. The OD₆₀₀ of each culture was measured just prior to sampling. Enough volume to obtain a total of 2.5 OD₆₀₀ of cells was applied to a vacuum filtration system through a 0.5-μm nylon filter (HNWP02500, MilliporeSigma). Using forceps, the filter was immediately removed and submerged into 3 ml of −20°C extraction solvent (80:20 methanol:water) in a 5-ml tube. Samples were stored at −20°C for at least 20 min. On wet ice, the extraction solvent was pipetted over each filter to get cells off of the membrane, and the membrane was removed and discarded. Samples were spun at maximum speed at 4°C in a microcentrifuge for 5 min. Samples were diluted an additional 10-fold into 40:40:20 acetonitrile:methanol:water after membrane removal and centrifugation. Undiluted and diluted samples were then transferred to an autosampler vial for analysis.

Targeted metabolomics were acquired on an Agilent Bio 1290 UPLC coupled to a Sciex 7500 QTRAP. MS acquisition occurred in scheduled MRM mode with two transitions per compound whenever possible. Dwell time was set to a minimum of 4 ms and a maximum of 250 ms, with a target cycle time of 1 s. Source conditions were as follows. Ion source gas 1: 60 psi; ion source gas 2: 60 psi; curtain gas: 46 psi; CAD gas: 8; source temperature: 350°C. Polarity switching was performed with a 1,600 V spray voltage for positive mode, 1,900 V for negative mode, and a pause time of 4 ms. Chromatographic separation for both methods was achieved on a SeQuant ZIC-pHILIC 5-µm 200 Å 150 × 2.1 mm HPLC column. Mobile phase A was 20 mM ammonium carbonate, pH 9.2, with 1 µM medronic acid. Mobile phase B was 95% acetonitrile, 5% mobile phase A. The gradient was as follows at 150 μl/min flowrate: 0 min: 84.2 %B; 20 min: 21.1 %B; 22 min: 15.8 %B; 22.5 min: 84.2 %B; 30 min: 84.2 %B.

Data files were converted to mzML format using ProteoWizard msConvert (Chambers et al., 2012), and all peaks were manually reviewed using MAVEN2 software (Seitzer et al., 2022). Targeted compounds were matched by retention time and a secondary transition whenever possible. Each compound was fit to its own model. If data at both concentrations were available, it was fit to the model “log₂(abundance) ∼ gene * concentration.” If only one concentration was available, it was fit to the model “log₂(abundance) ∼ gene.” The P-values were then adjusted to q values using the R package “q value” (Storey et al., 2025).

Cytochrome c oxidase activity was determined using Cytochrome c Oxidase Assay Kit (Sigma-Aldrich) according to the manufacturer’s instructions, modified to adapt the assay to a 96-well plate format. Isolated mitochondria (see “Mitochondrial isolation” above) were resuspended to 1 mg/ml in mitochondrial isolation buffer, then diluted 1:10 in enzyme dilution buffer with 1 mM n-dodecyl β-D-maltoside, and left on ice for 15 min before assaying. Reduced ferrocytochrome c substrate solution was prepared according to the manufacturer’s instructions. Each condition and replicate were paired, with 10 μl mitochondrial suspension added to one signal well, and the equivalent volume of 1× enzyme dilution buffer added to the other background well. 95 μl 1× assay buffer (10 mM Tris-HCl, pH 7.0, containing the appropriate concentration of KCl and NaCl) and 5 μl reduced ferrocytochrome c substrate solution were added to each well for a final volume of 110 μl. Absorbance at 550 nm was measured on a CLARIOstar Plus plate reader (BMG Labtech) immediately. The plate was read every 30 s for 20 cycles.

To calculate activity, the slopes describing absorbance at 550 nm vs time of each sample were determined using the “rlm” function from the R package “MASS” (Venables and Ripley 2002). For each experiment, the average slope of all the background conditions was subtracted from each sample and scaled to the amount of mitochondria added in milligrams to get the activity. Negative activities in the samples were set to 0. The model formula was “activity ∼ ionic_strength * buffer + experiment,” where “ionic_strength” was treated as a factor.

Cell sampling and preparation were based on Eide et al. (2005). Strains were grown to saturation overnight in media containing the appropriate nitrogen source (ammonium sulfate or arginine) and standard (7.35 mM) potassium. From these overnight cultures, strains were diluted into media containing the appropriate potassium concentration in biological triplicate and grown for an additional 24 h, ensuring that the OD₆₀₀ did not go above ∼0.1. The OD₆₀₀ of each culture was measured just prior to sampling. Samples of each media were also processed to estimate the amount of potassium collected by the filter. Enough volume to obtain a total of 1 OD₆₀₀ of cells was applied to a vacuum filtration system onto a 1.2-µm polycarbonate filter (RTTP02500, MilliporeSigma). This was washed 3 times with 5 ml of 1 µM EDTA disodium salt, followed by 3 washes of 5 ml water. Using forceps, the filter was placed into a screw-top microcentrifuge tube (HS10060, MilliporeSigma) containing 500 μl 35% HNO₃ such that it was completely submerged. During this process, ∼1 ml of the culture was placed into a 50-ml Falcon tube and placed on ice for cell count and size analysis. The sample in HNO₃ was placed in a shaking incubator set to 65°C and left to digest overnight. Cell counts and cell diameter were determined on a CoulterCounter, as above. The previously collected samples on ice were diluted into 10 ml Isoton II Diluent (C96980; Beckman Coulter) such that ∼5,000–10,000 cells were analyzed in a 500 μl sample.

The next day, each sample was processed by adding 500 μl water, vortexing, and removing the filter. The sample was centrifuged for 10 min at 12,000 g, and the supernatant was transferred to a 15-ml Falcon tube. To this, 7.75 ml water was added to achieve a final HNO₃ concentration of 2%.

Reference standards were prepared by pouring out 1 ml of a 1 g/l potassium ion reference solution (06335-100 Ml, MilliporeSigma) into a 1.5-ml tube. A 50-ml Falcon tube was filled with 50 ml blank solution (2% HNO₃). 200 μl of blank was removed and replaced with 200 μl of the potassium ion reference solution. This initial solution was diluted by serial transfer of 5 ml to four Falcon tubes each containing 20 ml blank solution. The final standard series contained 4,000, 800, 160, 32, and 6.4 ng/ml potassium ion.

Samples were run on a PerkinElmer 5300 DV optical emission ICP (OEICP) with an autosampler at UC Berkeley’s Rausser College of Natural Resources. Blank solution and reference standards were prepared fresh and run with every experiment. The OEICP signal was background-subtracted and converted into units of µg/l using the calibration curve generated from the reference standards. The amount of potassium collected on the filter in the blank samples was subtracted from all samples. Cell diameter was estimated as a function of genotype, potassium concentration, and nitrogen source using the model formula “mean_cell_diameter ∼ gene + nitrogen + log(concentration),” and then used to calculate internal potassium concentration for each sample. Estimates for internal concentration were then analyzed using the formula “internal_concentration ∼ genotype * external_concentration * nitrogen_source + (1|experiment).”

Triplicate cell cultures were aged for 48 h on MADs as described above. Separately, triplicate cell cultures were inoculated such that they spent 24 h in the exponential phase at the end of the 48-h aging period. All cultures were harvested on the same day and processed as above (Internal potassium of young cells), with the following modifications. For the exponential cultures, 1 OD₆₀₀ was collected from each culture. For the 48-h cultures, the OD₆₀₀ of a 1:50 dilution of each culture was measured, and the total OD₆₀₀ corresponding to the most dilute culture was collected from each sample. Cell counts and volumes of all cultures were determined using the CoulterCounter, as above.

To avoid altering potassium ion content, the cultures were not fixed during the wash steps. This resulted in significant daughter cell contamination in the 48-h cultures (Fig. S5 D, orange, left peak), and required us to estimate the proportion and volume of old mothers in the 48-h culture. To do this, we modeled each size distribution as a mixture of three Gaussians using the “GaussianMixture” class from the Python package “scikit-learn” (Pedregosa et al., 2011), and designating the distribution with the smallest mean as “daughters.” The two other distributions were summed to create an old cell size distribution. The potassium content of old cells was estimated in two ways. First (Fig. S5 E, purple), we assumed that the daughters of old mothers had the same potassium content as cells in exponential culture (Fig. S5 D, blue). Second, to get a maximum upper bound on the estimate (Fig. S5 E, brown), we made the (unlikely) assumption that the daughters of old cells contained no potassium (i.e., all of the extracted potassium came from old cells). In both cases, the concentration of potassium in old mothers was significantly lower than cells in an exponentially growing culture.

Fig. S1 shows reporter values for individual cells as a function of different aging parameters. Fig. S2 shows control experiments characterizing the MMP reporter system. Fig. S3 shows data associated with the screening platform for finding mutants that affect the age-associated decline of MMP. Fig. S4 shows other phenotypes associated with the sis2Δ mutant. Fig. S5 shows additional control experiments and a model relating potassium ion homeostasis to mitochondrial function through the activity of transporter Trk1p. Table S1 shows RNAseq data comparing young (exponentially growing) and 24-h old mother cells. Table S2 shows summary data for the deletion mutants and includes data from the initial screen, lifespan data, and the MMP ratio after 24 h compared with WT. Table S3 shows metabolomics data of the sis2Δ mutant compared with WT in standard (7.35 mM) KCl. Table S4 shows RNAseq data comparing WT and sis2Δ mutant in standard (7.35 mM) and low (0.5 mM) KCl. Table S5 shows a table of strains used in this study. Table S6 shows a table of oligos used to make the deletion mutants and check the deletions. Table S7 shows raw MMP and lifespan data of the remade deletion mutants and WT. Table S8 shows all other data used to generate the figures. Video 1 shows an example of a WT cell expressing the reporter system aging in the YLM, which is the same cell shown in Fig. 1 B.

The data underlying all figures are available in the published article and its online supplemental material. Raw RNAseq data underlying Table S1 and Table S4 are openly available in the NCBI Gene Expression Omnibus as GSE314946.

The authors wish to acknowledge Tina Mahatdejkul-Meadows for her initial development of the MMP reporter system, Jonathan Paw for operating the FACS equipment during the screen sorts, and Wenbo Yang (UC Berkeley College of Natural Resources) for operating the OEICP machine.

Author contributions: Adam James Waite: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, supervision, validation, visualization, and writing—original draft, review, and editing. Beiduo Rao: data curation, investigation, methodology, validation, and writing—review and editing. Elizabeth Schinski: data curation and software. Nathaniel H. Thayer: conceptualization, data curation, formal analysis, methodology, resources, and software. Manuel Hotz: conceptualization, investigation, and writing—review and editing. Austin E. Y. T. Lefebvre: data curation, formal analysis, methodology, resources, software, validation, and writing—review and editing. Celeste Sandoval: data curation, methodology, validation, and writing—original draft. Daniel E. Gottschling: conceptualization, supervision, and writing—review and editing.