A Competitive-Aging Method for Quantitative Genetic Analysis of the Chronological Lifespan of Saccharomyces cerevisiae

The chronological lifespan of budding yeast is a model of aging and age-related diseases. This paradigm has recently allowed genome-wide screening of genetic factors underlying post-mitotic viability in a simple unicellular system, which underscores its potential to describe the aging process in a systematic manner. However, results from different large-scale studies show little overlap and lack enough quantitative resolution to derive interactions of genetic aging factors with the environment. Here, we present a standardized, replicable, and parallelizable method to quantify the chronological-lifespan effects of gene deletions. We estimate the relative survival of stationary-phase cells by recurrently outgrowing co-cultures of wild-type and mutant strains expressing different fluorescent proteins. Importantly, we introduce a model to estimate the association between death rates and fluorescent signals, accounting for differences in growth rate and experimental batch effects. We describe the experimental procedure— from data acquisition to calculation of relative survivorship—for nine deletion strains and contrast replicability of our method with an established parallel approach. Furthermore, we apply our strategy to quantitatively characterize the gene-drug interactions of 76 deletion strains subjected to a lifespan-extending pharmacological treatment. Our competitive-aging approach with multiple-regression modeling provides a powerful screening platform to identify aging factors and their interactions with pharmacological interventions.


INTRODUCTION
A major challenge in aging research is to describe the way in which different genetic pathways and biochemical processes mediating aging are interconnected to one another and the environment 1,2 . Simple cellular models provide a starting point to grant a systems-level understanding of aging, in which the lifespan phenotype is addressed as a complex trait resulting from the action of multiple genes, cellular processes, environmental factors, and their interactions.
The chronological lifespan (CLS) of Saccharomyces cerevisiae is used to describe genetic, nutrimental, and pharmacological factors underlying survivorship of postmitotic, non-dividing cells 3 . The budding yeast's replicative-lifespan and CLS are simple experimental models that have been used to reveal the conserved lifespan-extending effects of reduced TOR and RAS/PKA signaling, as well as the anti-aging effect of rapamycin, spermidine, and caloric restriction [3][4][5][6] . Traditionally, the CLS of a yeast-cell population is measured by counting colony-forming units from samples of a long-term stationary-phase culture 7,8 . More recently, large-scale screening approaches have been implemented to screen for genetic aging factors in yeast. These studies provide unbiased catalogues of CLS mutant phenotypes 9-13 , mutants with diminished or enhanced response to dietary restriction or nutrient limitation 14,15 , and CLS phenotypes of collections of wild isolates and lines derived from biparental crosses 16 .
A current limitation of large-scale CLS phenotyping in yeast is that available screens have resulted in a large fraction of false positive hits when further confirmed by smallerscale approaches, ranging from 50% to 94% [9][10][11][12][13] . In addition, comparisons of different large-scale studies show that there is little overlap among the identified genetic factors.
While this may be explained in part by differences in genotypic background, media composition, and subtle environmental variations 17 , the large fraction of false positives and little overlap may also suggest that available large-scale CLS phenotyping approaches are still lacking enough technical replicability. Moreover, changes in specific controlled and uncontrolled environmental conditions are known to be important modifiers of CLS phenotypes and may be confounding causes of aging 12,[17][18][19][20] . Hence, a combination of high throughput and good resolution is needed to correctly determine not only genetic aging factors, but also to quantitatively derive their interactions with nutrimental, chemical, or pharmacological environments. This poses an important technical challenge, given that the number of experiments scales exponentially while describing genetic and environmental interactions based on phenotypic measurements.
In this study, we describe a novel multiple regression modeling approach to analyze measurements from a previously introduced competition-based method for quantitative large-scale genetic analysis of CLS in yeast 13 . Systematic analyses of the method's replicability and scripts to quantify CLS of strains aging in competition with a wild-type reference are also provided. Importantly, this previous study assumed that mutant strains had no effects in doubling time, which biased relative-survivorship measurements of slow-growing strains. For nine knockout strains, we compare the replicability of our results with those obtained with a useful parallelizable approach based on outgrowth kinetics 21,22 . Finally, we take advantage of our improved method to derive gene-environment interactions by measuring the relative effects on survival of metformin in 76 single-gene deletion strains, revealing some of the genes that mediate longevity by metformin in yeast. We discuss the potential of competitive-aging screening to describe not only interactions between thousands of genes and different environmental factors, but also large numbers of genetic interactions underlying aging and longevity in yeast.

MATERIALS AND METHODS
Strains and media. Nine single-gene deletion strains targeting ATG1, HAP3, MSN2, MSN4, RAS2, RIM15, RPS16A, STE12, and SWR1 were generated de novo by PCRbased gene replacement in the YEG01-RFP background 13 using the natMX4 module from pAG25 (Euroscarf). In addition, two isogenic reference strains were generated over the YEG01-RFP and YEG01-CFP backgrounds by deleting the neutral HO locus.
For gene-drug interactions, a collection of mCherry-tagged gene-deletion strains was generated by mating an array of 85 strains from the yeast deletion collection to the hoΔ YEG01-RFP SGA-starter strain, as previously described 13

Relative survivorship in stationary phase can be estimated from bulk fluorescence signal of two populations in co-culture
We sought to test and to improve a competition-based method to describe CLS phenotypes in budding yeast. To directly measure the lifespan effects of gene deletions, we tracked changes in the relative abundances of viable RFP-and CFP-tagged deletion (x RFP ) and wild-type strains (WT CFP ), respectively, as a function of time in stationary phase (T, days) in co-culture ( Figure 1A). To this end, we inoculated stationary-phase cells at different time points into fresh medium and monitored the outgrowth at multiple times (t, hours) by measuring absorbance at 600nm (OD 600 ), bulk RFP signal (RFP), and bulk CFP signal (CFP), until the outgrowth co-cultures reached saturation. First, we characterized the CLS of nine deletion strains that are known to show increased or decreased lifespan 5,12,13,15,[26][27][28] . Each RFP-tagged deletion strain was co-cultured with the WT CFP reference in up to seven replicates in a single deep-well plate, until saturation (see Materials and Methods). Competitive-aging cultures were monitored for ~15 days in stationary phase.
As expected, outgrowth kinetics measured by OD 600 showed a clear shift with time (days) in stationary phase; aging co-cultures gradually took a longer time in outgrowth (hours) to reach a given cell density ( Figure 1B). This overall shift in growth kinetics reflects the loss of viability with age, as previously described 21 . In terms of fluorescence-signal kinetics, we observed that the WT RFP or WT CFP monocultures mostly recapitulated OD 600 kinetics, suggesting that loss of viability can also be This data analysis procedure is a more exhaustive description of the actual competitiveaging setup, compared to our original report (Garay et al. 2014). Specifically, it takes into account relevant parameters fitting the measured data and has fewer assumptions, in particular we explicitly described the difference in growth rates between both strains in the outgrowth cultures using the parameter , so the quantification of relative survivorship is not biased by growth-rate phenotypes. We also consider systematic measurement errors by including the term. In the following sections, we show that competitive-aging experiments with multiple-regression modeling provide reliable quantifications of relative survivorship, which are useful to identify CLS phenotypes and to score their interactions with environmental factors.

Competitive-aging experiments provide replicable estimates of survivorship despite inter-batch variation
To systematically assess the technical replicability of the competitive-aging method, we measured the CLS of nine mutants and reference strains in 96-well plates with multiple independent replicate wells. Specifically, we measured CLS of up to seven replicate samples of each one of the nine deletion mutants together with 31 wild-type reference competitions in a 96-well plate (Figure 2A); the entire experiment was carried out twice (two experimental batches). We validated our results with another large-scale method that provides precise estimates of survivorship ( Figure 2B). In particular, we measured CLS of the same array of mutants and wild-type strains using an established highthroughput method that is based in the changes of outgrowth kinetics of aging monocultures 21,22 . Qualitative inspection showed that both methods scored mutants with known CLS effects 26,29 , such as rim15 and hap3 that resulted in reduced survivorship, and ras2 showing increased lifespan compared to wild type.
To quantitatively contrast the replicability of both experimental approaches, we fitted decay curves from the outgrowth kinetics experiments to an exponential model. We then compared the difference of adjusted exponential death rates of wild-type and mutant strains from monoculture aging to the parameter obtained from competitiveaging. We observed that both methods performed similarly when comparing the technical variation within each of the experimental batches; there was no significant difference in the typical standard error of the mean in outgrowth kinetics and competitive aging (Supplementary Figure S3). However, when looking at the correlation of quantitative data resulting from independent experiments, we found that the correlation was remarkably higher when experiments were done with the competitive-aging approach (Figure 2C; r 2 =0.56 and r 2 =0.94 for outgrowth kinetics and competitive aging, respectively, Pearson correlation). We note that one of the outgrowth-kinetics experiments was atypically noisy for ras2 (not shown); hence this strain was excluded in the analysis, as it would overestimate the intra-and inter-batch variability of the outgrowth-kinetics approach. Together, these results indicate that the competitionbased method is highly replicable, despite the inherent variation of different experimental batches.
We looked closer into the the diminished replicability of the outgrowth-kinetics method and compared the distribution of effects of all deletion mutants in both batches, as determined by the two experimental approaches (Figure 2D). Deletions rim15 and hap3 showed short-lived phenotypes with high statistical support in all experiments.
Conversely, the short-lived phenotypes of ste12, msn4, msn2, and atg1 were consistently detected only by competitive-aging but not by monocultures, in which short-lived phenotypes were only scored in one of two experimental batches. Long-lived phenotypes were identified in both approaches, but competitive-aging screening had better resolution, with better statistical support for the long-lifespan phenotype of the swr1 strain (p-val<10 -9 in both batches, t-test).  Figure S4). Thus, including 6-10 reference samples are enough to provide a robust description of relative-CLS phenotypes in large-scale genetic analyses of mutants, environments, and their interactions.

Competitive-quantification of CLS under different conditions successfully describes gene-drug interactions
Lifespan is a complex trait determined by different cellular pathways, hundreds of genes, and environmental variables 2,31 . A current challenge in the field is to understand how different factors are integrated with one another to control cell survivorship. Our competitive-aging method provides high-resolution and replicable data, which enables a quantitative description of CLS-phenotype interactions. As a proof of principle, we screened for gene-environment interactions in an array of knockout mutants aged with and without the lifespan-extending drug metformin.
We confirmed that the CLS of WT reference samples of yeast increased significantly from a half life of 10.7 to 15.5 days when treated with 40mM metformin (Figure 3A; p<10 -9 , Wilcoxon rank sum test). Next, we used our competitive-aging approach to measure the CLS of an array of 76 knockout strains aged with or without metformin. In both conditions, we observed high quantitative correlation of the phenotypes between replicate plates in the same experiment and between two independent experiments (Supplementary Figure S5). A direct quantitative comparison of CLS phenotypes under both conditions is shown in Figure 3B (see Supplementary Figure S6 for data rescaling). Most samples were found to fall close to the diagonal; namely the phenotypic effect of the knockout relative to WT was similar under both conditions. However, the phenotypes of 17 of the 76 knockouts (22%) were significantly different when treated with metformin (t-test, p<0.05; Figure 3C); these are potential gene-drug interactions.
For the most extreme differential phenotypes, we show the modeled change in relative survivorship as a function of age ( Figure 3D). In many cases, we observed that metformin alleviated or even reverted the short-lifespan effects of the gene deletion, with pep1, die2, and alg3 being the most extreme instances. In only one mutant, lsc2, the drug significantly aggravated the short-lived gene-deletion phenotype. On the opposite scenario, the nominal long-lived phenotype of certain mutants was rendered neutral or closer to neutral with the metformin treatment (eg. hsv2, ubp13, and ubp5). The identified gene-drug interactions, with specific quantitative information on the magnitude and sign of the effects, constitute a powerful means to pinpoint the underlying mechanisms of longevity by metformin in yeast.
Finally, we used the data set of 76 knockout strains to evaluate the influence of mutant's growth rate in the context of our competitive-aging method and multiple-regression model, which includes parameter . Importantly, we observed that the model correctly identified mutants with growth defects, as reported in the literature (Supplementary Figure S7); hence, this parameter may enhance a more accurate quantification of relative survivorship of deletion strains. The data was fitted to assuming =0; we observed that the difference in the estimation of when was included was modest, but mostly explainable by . As expected, the relative survivorship of slow-growth mutants was usually underestimated when differences in growth rate were not taken into account (Supplementary Figure S7). Together, results in this section show that competitive-aging yields accurate and replicable large-scale CLS data, providing valuable information to shed light on the mechanisms of pharmacological interventions that extend lifespan.

DISCUSSION
Genetic analysis of the CLS of budding yeast has led to the genome-wide identification of genes involved in aging; recent efforts have sought to describe interactions between genetic and environmental modulators of the phenotype 13,[15][16][17] . A previous report from our group has shown some advantages of using a competitive-aging approach, in which fluorescently-labeled strains in co-culture provide high resolution in parallel setups 13 .
Competitive-aging has also allowed scoring gene-environment interactions at the genomewide level, specifically interactions with dietary restriction 15 . Here, we have introduced a new model to calculate the relative survivorship of deletion strains, taking into account the possible confounding effects coming from growth-rate differences and systematic batch effects. In addition, we directly quantitatively compared the performance of a competitive-aging setup to an established approach 21,22 . CLS phenotypes were successfully recapitulated with both approaches, but competitiveaging provided higher replicability and resolution. With this enhanced method and data analysis, we were able to unravel significant gene-drug interactions in an array of 76 deletions strains subjected to the lifespan-extending drug metformin.
Early CLS genome-wide screens were based on large pools of gene deletions followed by molecular-barcode hybridization or sequencing 10,11,14 . These studies provided important insight into which genetic factors mediate stationary phase survival, such as autophagy, vacuolar protein sorting, and regulation of translation. However, the high rates of false positives-specially in the cases of long-lived phenotypes-and low overlap among the sets of genes from different studies 17 suggest that systematic errors in barcode detection or major experimental batch effects result in poor experimental replicability. On the other side of the spectrum, an ingenious outgrowth-kinetics approach of yeast monocultures increases the feasibility of percent-survivorship estimates, compared to the conventional colony-forming units method 21 ; but throughput is still limited with this approach. To overcome this limitation, Jung and co-workers scaled-up this strategy using monocultures in multi-well plates, whereby more strains can be tested in parallel 16,22 . There is still the issue that mild environmental variation can affect separate cultures differently, that could lead to low reproducibility 12 ; for instance, strains may reach stationary phase at different times after inoculation. In this regards, competitive-aging provides a direct phenotypic comparison and, arguably, more consistent results, given that the mutant population of interest is aged with an internal reference strain under the exact same microenvironment. Importantly, competitive-aging can also be carried out in multi-well plates, enabling high-throughput experimental setups.
Results herein presented confirmed that, in our hands, CLS phenotypes replicated better when obtained by competitive-aging and multiple-regression modeling using several predictor variables such as relative survivorship, relative growth-rate, initial frequency, and systematic biases in batch measurements. While the degree of agreement between measurements conducted on replicate samples in different laboratories remains to be addressed, it is likely that competitive-aging could provide higher reproducibility for the field of yeast CLS genetics. One of the inherent drawbacks of the method is that no absolute death-rate information is provided, which can be easily able to re-enter the cell cycle, which can only be distinguished from actual death using outgrowth-independent methods, such as live/dead staining.
We have illustrated the potential of competitive-aging screening by characterizing an array of yeast deletion strains exposed to metformin. A number of lifespan-extending pharmacological interventions are already being tested for age-related diseases in humans, even when the mechanisms underlying their beneficial effects are frequently not fully understood 32,33 . Given that lifespan is a complex phenotype, identifying showing that metformin alters glycation, protein transport, and protein degradation in yeast 24,34 . Our results also uncovered interactions between metformin and genes involved in mitochondrial function (erp6Δ, lsc2Δ, ema35Δ, sap155Δ, and ylh47Δ), which is a known player in the cellular response to metformin 24,25 . It remains to be addressed how these proteins are specifically related to the known response involving the mitochondrial electron transport chain and homeostasis of copper and iron.
Competitive-aging can readily be adapted to screen double mutants at large scale and to score genetic interactions underlying CLS phenotypes. A proof-of-principle study suggests that positive (alleviating) genetic interactions are common between shortlifespan autophagy mutants 13

Conflict of interest
The authors declare that they have no competing interests.

LIST OF SUPPLEMENTARY ITEMS
Note S1. Supplementary Note 1. Development and implementation of a linear model to calculate relative survival of mutant strains in competitive-aging cultures. Table S1. Recipes for media used in this study. Figure S1. The change in the RFP and CFP events detected by flow cytometry in outgrowth cultures reflects differential death rates.      Figure S7. The G parameter corrects effects of differential growth rates that otherwise underestimate relative survivorship.
Data S1. Array of mutants used for gene-drug interactions, with their relative survivorship, rescaled survivorship, and relative growth rates.