|Home | About | Journals | Submit | Contact Us | Français|
Research into the genetic mechanisms of aging has expanded rapidly over the past two decades. This has in part been the result of the use of model organisms (particularly yeast, worms and flies) and high-throughput technologies, combined with a growing interest in aging research. Despite this progress, widespread consensus regarding the pathways that are fundamental to the modulation of cellular aging and lifespan for all organisms has been limited due to discrepancies between different studies. We have compared results from published genome-wide, chronological lifespan (CLS) screens of individual gene deletion strains in S. cerevisiae in order to identify gene deletion strains with consistent influences on longevity as possible indicators of fundamental aging processes from this single-celled, eukaryotic model organism.
Three previous reports have described genetic modifiers of chronological aging in the budding yeast (S. cerevisiae) using the yeast gene deletion strain collection. We performed a comparison among the data sets using correlation and decile distribution analysis to describe concordance between screens and identify strains that consistently increased or decreased CLS. We used gene enrichment analysis in an effort to understand the biology underlying genes identified in multiple studies. We attempted to replicate the different experimental conditions employed by the screens to identify potential sources of variability in CLS worth further investigating.
Among 3209 strains present in all three screens, nine (2.80%) deletions strains were in common in the longest-lived decile and thirteen (4.05%) were in common in the shortest-lived decile for all three screens. Similarly, pairwise overlap between screens was low. When the same comparison was extended to three deciles to include more mutants studied in common between the three screens, enrichment of cellular processes based on gene ontology analysis in the long-lived strains remained very limited. To test the hypothesis that different parental strain auxotrophic requirements or media formulations employed by the respective genome-wide screens might contribute to the lack of concordance, different CLS assay conditions were assessed in combination with strains having different ploidy and auxotrophic requirements (all relevant to differences in the way the three genome-wide CLS screens were performed). This limited but systematic analysis of CLS with respect to auxotrophy, ploidy, and media revealed several instances of gene × nutrient interaction.
There is surprisingly little overlap between the results of three independently performed genome-wide screens of CLS in S. cerevisiae. However, differences in strain genetic background (ploidy and specific auxotrophic requirements) were present, as well as different media and experimental conditions (e.g., aeration and pooled vs. individual culturing), which, along with stochastic effects such as genetic drift or selection of secondary mutations that suppress the loss of function from gene deletion, could in theory account for some of the lack of consensus between results. Considering the lack of overlap in CLS phenotypes among the set of genes reported by all three screens, and the results of a CLS experiment that systematically tested (incorporating extensive controls) for interactions between variables existing between the screens, we propose that discrepancies can be reconciled through deeper understanding of the influence of cell intrinsic factors such as auxotrophic requirements ploidy status, extrinsic factors such as media composition and aeration, as well as interactions that may occur between them, for example as a result of different pooling vs. individually aging cultures. Such factors may have a more significant impact on CLS outcomes than previously realized. Future studies that systematically account for these contextual factors, and can thus clarify the interactions between genetic and nutrient factors that alter CLS phenotypes, should aid more complete understanding of the underlying biology so that genetic principles of CLS in yeast can be extrapolated to differential cellular aging observed in animal models.
Knowledge about aging-related cellular mechanisms has advanced rapidly in recent decades, yet an integrated understanding of age-related deterioration lacks consensus. While aging is multi-factorial and organism-specific in some respects, genetic and nutritional interventions can influence the onset and rate of age-related decline similarly across different species, indicating fundamental processes of cellular aging are evolutionarily conserved. Thus leveraging the powerful genetic analysis of the single-celled organism, Saccharomyces cerevisiae, to dissect the complexity of gene-nutrient interaction effects on cellular aging could aid derivation of biological principles that apply to higher eukaryotes (Curran and Ruvkun 2007; Kennedy 2008; Smith and others 2007).
There are two types of aging commonly studied in budding yeast (Longo and others 2012). Replicative lifespan (RLS) is the number of times a mother cell produces daughters prior to senescence (mothers and daughters are easily distinguished due to asymmetric budding). RLS is seen as a useful model of cellular aging for mitotically active cells, such as stem cells. The second model for aging in budding yeast, which is the topic of this article, is chronological lifespan (CLS), referring to the maintenance of viability of non-mitotic cells. The percentage of colony forming units versus time is considered to be the gold standard measure of CLS. Assuming a cessation of cell division in ‘stationary’ culture and thus a constant number of total cells, CLS reflects the duration of cell survival after transitioning from a dividing to non-dividing state, which is quantified by shifting cells back to fresh media after increasing intervals of time in stationary phase. In contrast to animal cells yeast do not enter a post-mitotic state independently of nutrient depletion, however the nutrient milieu during logarithmic growth (non-starvation state) ultimately affects yeast CLS, which could therefore be indicative of factors generally affecting eukaryotic cellular health and post-mitotic survival (Fabrizio and Wei 2011; Ocampo and others 2012). In summary, the yeast CLS model offers advantages for large-scale genetic analysis, yet much remains to be discovered regarding its relevance, alongside other experimental models, for understanding human healthspan.
RLS and CLS are considered to have complementary relevance for understanding aging in multi-cellular organisms; RLS being relevant to mitotic tissues with regenerative reserve capacity such as epithelial linings and hematopoetic tissue, whereas CLS has greater relevance to cells comprising non-dividing tissues such as brain and cardiac or skeletal muscle. Although the biological relationship between yeast CLS and RLS and its relevance to aging across the broader eukaryotic kingdom remains to be further demonstrated, overlapping mechanisms have been suggested (Burtner and others 2011; Delaney and others 2013; Mirisola and Longo 2012; Polymenis and Kennedy 2012; Smith and others 2008). Similarly, nutrient signaling pathways and dietary interventions exhibit analogous effects on RLS, CLS, and aging of multicellular organisms (Fabrizio and others 2001; Jiang and others 2000; Kaeberlein and others 2005).
For technical reasons, CLS has been more amenable to high throughput studies, as a genome-wide RLS study was only recently completed (McCormick and others 2015). Though CLS (plating and colony counting) is higher throughput than RLS (dissection and counting of daughter cells), neither accommodate the throughput for routine genome-wide assessment of the entire YKO/KD collection. Other approaches have been developed to estimate CLS, (Fabrizio and others 2010; Matecic and others 2010; Powers and others 2006), which we discuss below. High throughput CLS techniques sample large cell populations in stationary culture over days to weeks to estimate the decline in viable cells vs. time, based on the percentage of cells re-entering proliferative growth when exposed to fresh media (Figure 1).
Two different techniques for high-throughput CLS estimation include: (1) outgrowth in liquid media in a 96-well format (Murakami and Kaeberlein 2009; Powers and others 2006) and (2) outgrowth of pooled cultures on agar followed by DNA extraction and relative hybridization to microarray chips (Fabrizio and others 2010; Matecic and others 2010). Each gene deletion cassette has been designed with unique flanking oligonucleotides that anneal to complementary oligonucleotides synthesized on hybridization chips (Giaever and others 2002; Winzeler and others 1999), which are used to estimate changes in relative abundance of particular deletion strains over time (Irizarry and others 2003; Ooi and others 2003; Yuan and others 2005). There have been two genome-wide CLS screens performed by the pooled library approach (Fabrizio and others 2010; Matecic and others 2010) and one by the individual cultures grown in micro-well plates (Powers and others 2006). The main focus of this article is on the comparison of these threes screens (Table 1), including whether genes and biological processes known to influence cellular aging are revealed by their overlap, and consideration of possible reasons for any lack of consensus (Table 2, Figures 2 and and33).
We found remarkable lack of agreement between the three genome-wide screens and hypothesize that gene-environment interaction is a major cause of discordance because different media were used (Table 2). It is notable that large-scale aging studies performed in other organisms, in particular C. elegans, have also resulted in low concordance, which could also be explained by nutritional differences (Curran and Ruvkun 2007; Hamilton and others 2005; Hansen and others 2005; Kennedy 2008; Smith and others 2007). Consistent with this possibility, nutrient interventions alone such as methionine restriction exert influence on aging across multiple species (Johnson and Johnson 2014; Orentreich and others 1993; Troen and others 2007). Furthermore, in mouse or rat models with otherwise tightly controlled experimental conditions, variable longevity results obtained for interventions performed at different sites (e.g. National Institute on Aging’s Interventions Testing Program) or at the same site across multiple years appear to be in part due to unexpected effects of different rodent chow impacting lifespan (Ghirardi and others 1995; Harrison and others 2014; Strong and others 2008). Although there are other plausible explanations for the discordant CLS results observed for the yeast gene deletion strain collection, we conclude that yeast genetic analysis of CLS (and RLS) provides a powerful platform to investigate gene-environment and gene-gene interaction from a comprehensive perspective to gain additional clarity regarding their effects on cellular health and aging. To investigate gene-nutrient interaction with respect to CLS, we performed a controlled experiment using quantitative high throughput cell array phenotyping to systematically compare the effect of different media on CLS, as well as the possible interaction of media with auxotrophic requirements and/or ploidy. The results indicate broad dependency of the CLS phenotype on media, auxotrophy, and ploidy (i.e., gene × nutrient interaction) in the three genome-wide CLS screens published to date.
CLS data were obtained from supplemental data files and compared by correlation and decile distribution analysis to describe concordance between genome-wide screens performed by three independent laboratories (Fabrizio and others 2010; Matecic and others 2010; Powers and others 2006). Differences between the screen methodologies are illustrated in Figure 1, and summarized in Table 1. Differences in media composition are summarized in Table 2.
The total number of Yeast Gene Deletion Strains (YGDS) reported on by each of the three screens was variable, and less than the total number of strains in the respective collections, due to lost detection of strains in the respective screens (Table 1, Supplemental Table 1). The Matecic et al study reported 3209 unique strains identified by the “UP” tags, and 2538 by the “DOWN” tags, thus the more inclusive UP tag set was used for comparisons with the other two screens. The three studies were compared pair-wise by Pearson correlation, revealing little agreement between them considering the studies shared essentially the same objective (Figure 2). In contrast, a within-study analysis of Matecic et al. showed reproducibility of viability estimates by comparing barcode abundance of the UP tags and DOWN tags for each ORF deletion construct (Matecic and others 2010) (Table 1, Supplemental Table 1). The correlation coefficient (R) for the up and down tags from Matecic et al was 0.79 (Figure 2A), while that for Matecic et al (UP tags) with Powers et al was R=0.11 (Figure 2B). The correlation coefficient for Matecic et al and Fabrizio et al was R=0.09 (Figure 2C), while the lowest correlation was between Powers et al and Fabrizio et al, with R=0.018 (Figure 2D).
We compared decile distributions to look with finer detail for possible correlation in extension or shortening of CLS between the different screens (Figure 3, Supplemental Tables 1–5). In the within-study comparison of Matecic et al (UP and DOWN tags), there was smooth overlap in the distribution across all CLS deciles (Figure 3A, Supplemental Figure 1A). Among the across study comparisons, there was discernable correlation for short-lived and long-lived deletion strains, albeit very limited (Figure 3B, ,3C,3C, ,3D3D and Supplemental Figure 1B, ,1C,1C, ,1D).1D). There appeared to be slightly greater agreement between the Matecic et al and Fabrizio et al screens (Figure 3C and Supplemental Figure 1C), particularly for short-lived strains, however the correlation was low for every between-screen comparison.
Unlike other genetic models where screens are non-saturating or differences in genetic backgrounds exist, thus providing potential explanations for discrepancies among studies, the results with the YGDS suggest factors in addition to gene deletions of interest are contributing to the longevity effects. Discrepancy may be best explained by differences in experimental conditions used such as media recipe and/or subtle differences in the auxotrophic requirements or ploidy of the YGDS employed in the respective studies.
We next investigated whether the overlap in genes identified between screens could be related to common aging mechanisms, as each published report implicated specific classes of genes in longevity regulation, such as nutrient sensing pathways including TOR1 or SCH9, purine biosynthesis, and autophagy (Fabrizio and others 2001; Kennedy and others 2005; Matecic and others 2010; Powers and others 2006). Consistent with the low correlation of results overall, the small overlapping set of CLS extending mutants, common to all three screens, did not over represent any particular pathway (Table 3).
Although there was no statistically significant enrichment based on gene ontology assessments, it was notable that deletion strains related to autophagy (e.g. ATG14, VPS27) and nutrient sensing/stationary phase transition (RIM8, RIM15, RIM101) were identified among the overlapping short-lived deletion strains (Table 4). Expanding the analysis to include overlap in the top two deciles from all three screens revealed 52/642 long-lived and 102/642 short-lived strains in common between the three screens (Figure 4, Supplemental Tables 2–5), yet when compared to the background of genes (Supplemental Table 1) provided no enrichment for biological processes, molecular functions or cellular components associated with long-lived or short-lived phenotypes. However, expansion of the analysis to inclusion of the top three deciles (Figure 4) revealed functional enrichment for chromatin accessibility complex among long-lived mutants and enrichment of autophagy-related genes among short-lived mutants (Supplemental Table 6).
Although there are relatively few overlapping genes between the screens, and the genes that do overlap for longevity extension show little significant enrichment in pathways or functions known to regulate aging, they may nevertheless suggest clues to consensus CLS pathways. For example, the identification of long-lived deletion strains with connections to gene expression regulation (ITC1), mRNA turnover (PUF3) and transcriptional regulation (SSN2) could emphasize the importance of a coordinated response for entry into and maintenance of stationary phase for successful cell survival (Table 3). Similarly, the handful of overlapping short-lived deletion strains might suggest the proper establishment of stationary phase transcriptional and metabolism state (RIM15, RIM8, RIM101, UBP8); maintenance of redox balance through glutathione metabolism (GCV3, DUG2); protein production, stability and turnover (VPS27, MON2, ATG14); and possibly DNA integrity exemplified by telomere maintenance (MTC7, NMD2, ECM29) are important for the long-term survival of cells in non-dividing states (Table 4). Perhaps, more importantly, the consistency between the technical replicates within the Matecic et al study suggests that reproducibility is possible (Figures 2 and and3)3) and thus experimental factors such as media and genetic background (Tables 1 and and2)2) may explain differences in CLS results obtained between the three YGDS screens. Genetic background differences can occur due to the presence of different auxotrophic markers, mating type, and ploidy of the respective deletion strain collections as well as from genetic drift and/or selection of secondary mutations resulting from suppression of the engineered gene deletions (known to occur in some deletion strains). From this perspective, the network of genes that regulates aging could exhibit plasticity in response to variations in dietary, nutritional, and/or environmental context. Although variable CLS poses a challenge, analogous challenges exist for understanding aging in multi-cellular organisms, and thus genome-wide CLS in the YGDS provides a relatively tractable genetic model to address the complexity of aging resulting from genes interacting with the environment and other genes.
Cellular health and longevity depend in part on the availability and sensing of nutrients available for growth, reproduction and maintenance, and coordinating cellular activities accordingly. For yeast, nutrients can be freely manipulated via liquid or agar-based media, allowing highly controlled experimentation about nutrient effects. The most common media used in yeast research consist of “rich” media, “YPD”, comprised of three ingredients: 1) Yeast Extract, 2) Peptone and 3) Dextrose (glucose). Yeast extract is derived from cultures of yeast, while peptone is generally derived from enzymatic digestion of animal protein, which are ‘undefined’ due to variability between proportions of molecular species between preparations. In contrast, ‘defined’ media has been commonly used in genome-wide CLS assays, which is interchangeably called “SC” (Synthetic Complete) or “SD” (Synthetic Defined) media. The specific makeup of synthetic media differs between laboratories, as was the case for the three CLS studies (Table 2). Most laboratories use a component of defined media, called yeast nitrogen base (YNB), for which the standard formula derives from the work of Wickerham (Wickerham 1946). Ammonium sulfate is often added in a standard amount (5 gm/L) as a nitrogen source, and amino acids are added separately in varying combinations and amounts (e.g., Table 2). With YPD, only the carbon source is typically manipulated, whereas the amino acids and additional nitrogen sources can also be individually adjusted in composition or amount with defined media. Media constitution likely has significant effects on CLS, yet media differences have received little attention in the study design relative to genetic differences. It is also worth considering that genetic factors and media components can interact, for example auxotrophic markers and corresponding nutrient requirements. Leucine, histidine, uracil, methionine and lysine are frequently used auxotrophic markers in the genetic background of the YGDS library, and thus the amino acid complements are provided in excess, but often in varying amounts (Table 2). Notably, amino acid levels (e.g., leucine and glutamine) can impact nutrient signaling pathways shown to influence cellular aging, such as TOR (target of rapamycin) (Ocampo and others 2012; Pan and others 2011). These and related considerations are further complicated with screens employing pooled cultures, where strain-specific metabolite production, secretion and response are not readily tracked, but co-occur and potentially interact. CLS in isolated monocultures provides greater control in experimental design with regard to detection of main effects of auxotrophy and media composition as well as gene-gene and gene-media interactions, but in principle reduces throughput for genome wide CLS analysis.
As shown in Table 2, the three studies all used 2% dextrose as carbon source and 5 gm/L ammonium sulfate for excess nitrogen. Matecic et al. also screened with media containing 0.5% dextrose, but it was the only study to do so and thus the 2% condition was used for comparative analysis. The amino acids component of the recipes varied considerably. The Matecic et al. and Longo et al. studies used the MATa haploid genetic background, which has the his3, leu2, ura3, and met17 auxotrophies, while the Powers et al. study used the homozygous diploid background with complete auxotrophy for his3, leu2, and ura3 and heterozygous for met17/MET17 and lys2/LYS2. Additionally, the Longo media contains a phosphate buffer (Fabrizio and others 2010; Wei and others 2011), and their protocol calls for removal of the media and assay of CLS in water. Whether media differences account for the low concordance between different genome-wide CLS screens remains to be formally tested.
Yeast transition between metabolic states in response to available nutrients and oxygen, engaging to variable extents in fermentative and respiration-based processes. Yeast repress respiration in a high glucose, nutrient rich environment, undergoing glycolysis and fermentation until glucose or other nutrient limitation de-represses respiration, leading to further extraction of energy by conversion of ethanol to organic acids, in particular acetic acid, which reduces CLS. Respiration increases with reduction of glucose concentration to 0.5% or lower (“calorie-restricted”) from 2% (“non-restricted”) (Ocampo and others 2012; Pan and others 2011), but the significance of this metabolic change with regard to the many genetic effects on the CLS phenotype is not well understood, for example, there is less organic acid production by cells growing in media with lower glucose (Burtner and others 2009a). Thus, aerobic glycolysis and fermentation of glucose in the log phase of growth, organic acid production and secretion into the media, and its consequent effects on CLS represent an important mechanism of CLS, which can be greatly affected by media conditions and aeration status (Burtner and others 2009a; Murakami and others 2011; Ocampo and others 2012; Wierman and others 2015). Buffering the media (e.g., in the Longo lab media), or washing away the acetic acid in conditioned media, and aging in water can also potentially increase CLS. However, there may also be CLS-extending factors that are altered by these same interventions, and the production of either CLS- extending or shortening factors may themselves depend upon nutrient factors, genetic background, and gene interaction. The “acetic acid mechanism” appears to be fairly complex, but it may further be only one of many related or independent mechanisms of CLS. The number of factors that influence CLS is only beginning to be characterized (Matecic and others 2010; Ocampo and others 2012; Wierman and others 2015).
The S. cerevisiae YGDS has contributed greatly to understanding that gene interaction is ubiquitous; i.e., the knowledge that networks of genes contribute to every phenotype is a direct result of widespread use of the YGDS library (Costanzo and others 2010; Giaever and Nislow 2014; Hartman and others 2015). Gene interaction in the YGDS is usually reported in terms of surprising changes in cell proliferation of deletions mutants relative to reference strain, in response to genetic (e.g., double mutants or expression of heterologous gene) or chemical perturbation (Mani and others 2008). One way of quantifying gene interaction involves a dose response to perturbations such as concentrations of a drug, levels of gene expression, or both (Hartman IV 2007; Hartman IV and Tippery 2004; Louie and others 2012), which can be seen as analogous to CLS, where the perturbation is time (Figure 5). Gene interaction effects are observed extensively with respect to different yeast media (Hartman and others 2015), which is likely to also be true for CLS - a more complicated phenotype that scores cell proliferation as a function of age. Thus media recipes appear to have a greater effect on CLS than the simple logarithmic growth phenotype (Burtner and others 2009a; Murakami and Kaeberlein 2009; Murakami and others 2008). In summary, the complexities of high throughput growth phenotyping are amplified in CLS by serial assay of aging cultures. Based on these observations, it is plausible that lack of consensus between different laboratories could be explained by differences in strains, media and related biology rather than failed replication or stochastic effects. Carefully controlled experiments, for example involving paired analysis of the MATa/met17/LYS2, MATα/MET17/lys2 and or homozygous diploid libraries (also controlling for media components) would help to further explore gene × nutrient interaction in CLS.
Another potential source of discrepancy between large-scale CLS studies involves specific phenotypic screening methodologies employed, which can influence outcome and interpretation. The gold standard for measuring of CLS is to plate cells from an aerated stationary phase liquid culture to fresh agar media to determine the percentage change in CFU vs time (Figure 1). The high throughput screens used different strategies to estimate CLS; alternatively, outgrowth measured by re-inoculation in liquid media and measurement of growth curves by OD, or plating of pooled YGDS cultures (aged together in one stationary phase culture) onto agar followed by bulk harvest of colonies, genomic extraction of DNA and fluorometric labeling and microarray hybridization. Each methodology can provide accurate estimates, however contributions to total cell proliferation other than the number of colony forming cells potentially exist, such as doubling time (exponential growth rate) or carrying capacity (final density in nutrient-exhausted media), which in principle should not impact CLS by the gold standard CFU-estimating assays. Additionally, cell-extrinsic factors are secreted differentially by YGDS, which could affect CLS (Matecic and others 2010; Wierman and others 2015). The production of such factors and the dependence on genetic context of their CLS influence (i.e., interaction between gene deletion, CLS-factor production, and CLS-factor effect) could be detected differently by CLS of pooled vs. individually cultured YGDS. Finally, the aeration status influences CLS as evidenced by the different durations of analysis between the non-pooled (lower aeration of cultures in individual wells of 96 micro-well plates) approach and the pooled/barcode strategy (higher aeration in flask cultures). Delineation of these different contributors to the CLS phenotype could also provide greater ability to reconcile discordant results between different genome-wide screens for CLS (Fig. 5).
To clarify the potential for media and genetic factors to influence CLS, we investigated the different experimental conditions employed by the three previous genome-wide CLS screens using a single platform for measuring CLS (Louie and others 2012; Rodgers and others 2014; Shah and others 2007). Quantitative high-throughput cell array phenotyping (Q-HTCP) was used to measure CLS in six different media conditions and three strains with differing ploidy and auxotrophic requirements. Replicate cultures (n=192 for BY4741 and n=122 for BY4743) were assayed in the two different YGDS parental backgrounds (Brachmann and others 1998) used for the prior genome-wide CLS screens. Additionally, we analyzed BY4742 (n=96), the haploid strain that yields BY4743 if mated to BY4741. Many replicates were used in order to assess the variance in CLS under the different defined conditions. The media analyzed included the three used in prior genome-wide screens (Table 2). Additionally, low-aeration was assessed for the Powers et al media, since that screen was carried out with reduced aeration. In our experiment, 384 well plates were inverted to achieve aeration by cells settling at the air-liquid interface (surface tension holds the media in place); the non-aerated condition being obtained by housing plates in the standard upright position. A recently described ‘human-like’ (HL) media (Supplemental Table 7) was tested with and without 0.1M MES (2-(N-morpholino)ethanesulfonic acid) to buffer extracellular pH, which has been shown to extend CLS (Burtner and others 2009b; Burtner and others 2011; Fabrizio and Wei 2011; Mirisola and Longo 2012). CLS was estimated approximately once per week, assaying the aging, 384-culture arrays by Q-HTCP, consisting of time series imaging of agar cell arrays, image analysis, growth curve fitting and use of resulting parameters to estimate CLS (Figure 5 and Supplemental Figure 2). Some cultures appeared to lose viability completely (no growth curve detected). Occasional cultures appeared to regain viable cells, a phenomenon previously described as ‘gasping’ (Fabrizio and others 2004). Gasping occurred at different ages and to different extents across the strains and media conditions (Figure 5). Several observations from this experiment indicate that genetic factors that distinguish the BY reference strain backgrounds (BY4741, BY4742, or BY4743), media factors and interactions between them can exert influence on CLS. CLS differences among strains could result from different mating type, ploidy or auxotrophic requirements (Table 1), while the possible explanations for media effects on CLS are more numerous (Table 2). Examples of the genotype- and media-specific differences in CLS observed include (Figure 5): (1) In the Powers et al media, aeration shortens lifespan to greater extent for BY4742, than BY4741 or BY4743; (2) In the Powers et al media, we found CLS to be longer for BY4743 than either haploid strain, and observed BY4741 to have longer CLS than BY4742 between days 10 and 24; (3) In the HL media buffered with 0.1M MES, as expected, all strains exhibit increased CLS. The effect of media buffering on increasing CLS was particularly notable for the BY4742 strain between days 31 and 46; (4) The pattern of CLS for the three strains aged in buffered HL media most resembled that of the Longo media, having BY4742 with the longest CLS, followed by BY4743, and with BY4741 having the shortest CLS. This is likely due to the Fabrizio et al media also employing pH buffering (see Table 2); however, we note our assay omitted a step in the Longo CLS protocol of washing the cells with water after 3 days (growth saturation) to perform the CLS assessment in water. (5) In the Matecic et al media, CLS was much shorter for BY4743. (6) In the Powers et al media, reducing the aeration increased CLS to different extents for the three parental strains, but the relative CLS (BY4743 BY4742 >BY4741) was similar.
In summary, these results support the hypothesis that interactions between genetic factors such as auxotrophy, mating type and ploidy in addition to media markedly influence CLS and thus could potentially explain some discrepancies between the results of the different genome-wide CLS screens. Toward the goal of extrapolating observations about yeast chronological aging to human cells, and given the potential influence of media on CLS, it is conceivable that the use of yeast media that more closely resembles human tissue culture media could help to reduce gene × media interaction as a confounding factor thwarting the validation of yeast CLS findings for human cells (Hartman and others 2015).
Though beyond the intended scope of this review, other contributors to variability in YGDS phenotypes exist and probably affect CLS. Replacing a genomic sequence with a constitutively expressed, selectable marker can affect expression of its neighboring genes such that the resulting phenotype of the deletion strain is incorrectly attributed to the gene deletion (Giaever and Nislow 2014; Winzeler and others 1999). Mechanisms of neighboring gene effects (NGEs) include alteration of local chromatin structure and transcription factor binding. The importance of NGEs for interpretation of YGDS phenotypes has been described (Ben-Shitrit and others 2012), and should also be considered in the CLS phenotype.
Another phenomenon that potentially clouds the interpretation of CLS is gasping, which is the regrowth of viable cells after a stationary phase culture begins to lose viability (Fabrizio and others 2004). Gasping appears to be stochastic, which can make it difficult to detect, and depending on its timing with respect to the CLS assay points, gasping can contribute to differences between studies by masquerading as extended CLS. It is also possible that genetic or nutrient perturbations could influence gasping (Figure 5).
As mentioned above, gene deletion strain cultures can accumulate secondary mutations due to drift or to selection for suppression of growth defects resulting from the original deletion. Secondary mutations, like neighboring gene effects of gene deletion, can confound YDGS phenotypes leading to either false positive (phenotype independent of the gene deletion) or false negative (secondary mutation masks the phenotype of gene deletion) results (Hughes and others 2000). Assuming that secondary mutations are not fixed in the population, replicating phenotypic analysis on sub-clones of YGDS cultures can be used to assess the likelihood of stochastic or indirect/secondary effects. The confounding effects of secondary mutations can be reduced by minimizing passaging of the library and assessed by inclusion of replicate cultures and defining phenotypic variance in the parental strain (Figure 5).
In summary, neighboring gene effects, gasping, and secondary mutations are examples of phenomena that can cloud the genome-wide study of CLS in the YGDS. These confounders compound the complexity of gene-media interaction. Though challenging, systematically investigating gene × nutrient interaction in CLS requires high throughput phenotyping to achieve adequate replication for characterization of variance. Addressing these issues should provide clarity regarding the core set of genes directly responsible for aging phenotypes of interest. Advancing systems level modeling of yeast aging will improve our ability to more accurately model more complex aging processes in multi-cellular organisms, including extending human health and longevity.
It appears that multiple genetic and environmental interactions influence CLS in yeast, and thus there is a need for consistent, standardized, and well-documented procedures when performing CLS assays. We expect genetic and environmental interaction is also important for understanding CLS in multi-cellular organisms such as worms, flies, mice, monkeys. Factors affecting CLS have also been shown to influence RLS (Delaney and others 2013; Mirisola and Longo 2012; Murakami and others 2012; Polymenis and Kennedy 2012), perhaps broadening the relevance of gene-nutrient interaction to the RLS model of aging. Moreover, genetic × environmental interactions are likely to differentially affect cellular aging in different cell types, harnessing the power of simple model systems like yeast to derive applicable principles of gene-nutrient interaction and aging should generate hypotheses useful for investigating the cellular complexities of human aging.
Data were obtained from the supplemental files provided with the three publications reporting genome-wide analysis of CLS in the yeast gene deletion strain collection (Fabrizio and others 2010; Matecic and others 2010; Powers and others 2006). The time points used from comparison (see Table 1) were 21 days (Matecic et al), 20 days (Fabrizio et al), and 7 weeks (Powers et al). Pearson Product-Moment Correlations were calculated using SAS Software (v.9.3). Screen hit analyses were performed using YeastMine, provided through a collaboration of the Saccharomyces Genome Database and the InterMine project (Cherry and others 2012). Lists of overlapping genes were obtained for the indicated comparisons (Tables 3, ,4,4, Supplemental Tables 2–5) by YeastMine Intersection analysis of the corresponding gene lists (Supplemental Table 1). The Widget for Gene Ontology Enrichment analysis was used to assess significant enrichment (using the Holm-Bonferonni Test Correction, and p=0.05 significance cut off) for biological process, cellular component or molecular function, setting the background population for comparison to the total overlapping ORF set for all genes shared among the three CLS screens (Supplemental Table 1, Matecic upstream tag only).
Q-HTCP was performed as previously described (Hartman IV and Tippery 2004; Rodgers and others 2014; Shah and others 2007). Briefly, a 384-well happy array containing the 3 reference strains was grown in YPD for 2 days and used to inoculate the six different media (see Fig. 5, Table 2, and Supplemental Table 7). The aging arrays were covered with adhesive gas permeable membranes and placed with well opening side down at 30 °C in a humidified incubator. At the indicated times, the aging arrays were printed to fresh HL agar media (Supplemental Table 7, and Supplemental Figure 2). At each time point, a custom robotic system was used to collect cell array images every two hours (Supplemental Figure 2), and custom image analysis software used to quantify the images, fit each numerical time series to a logistic growth function, and calculate the area under the growth curve (Figure 5) at 96 hrs.
This research was supported by grants to JLH (Research Scholar Grant 10-066-01-TBE American Cancer Society, Physician-Scientist Early Career Award 57005927 Howard Hughes Medical Institute, and NIH grant R01 AG043076) and by T32 DK062710 and P30 DK056336. The opinions expressed herein are those of the authors and not necessarily those of the NIH or any other organization with which the authors are affiliated.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
John L. Hartman IV has a financial interest in Spectrum PhenomX, LLC, which aims to commercialize Q-HTCP technology. All other authors declare no competing interests.