Studies using the budding yeast, Saccharomyces cerevisiae, have been at the forefront of recent efforts to understand the molecular mechanism of action of CR. Yeast lifespan can be measured in two ways. Replicative lifespan is defined as the number of buds produced by an individual yeast mother cell and is independent of calendar time (Mortimer & Johnston, 1959; Muller et al., 1980); in contrast to the alternative measure of the duration of viability in stationary phase, termed chronological lifespan (MacLean et al., 2001). Reducing the concentration of glucose in yeast growth media from the standard 2% to 0.5% or below increases both replicative and chronological lifespan irrespective of genetic background and has been suggested to be a model of CR (Jiang et al., 2000; Lin et al., 2000; Fabrizio et al., 2005; Smith et al., 2007).

Although the effect of glucose limitation in extending yeast replicative lifespan is not disputed, its mechanism of action is the subject of considerable controversy and heated debate. Three main models have been suggested, two of which centre on the evolutionarily conserved NAD-dependent histone deacetylase, Sir2 (Fig. 1). Initially, it was proposed that CR causes a metabolic shift away from fermentation and towards respiration, resulting in an increased NAD:NADH ratio, and thus increased Sir2 activity (Lin et al., 2000, 2002, 2004) (model 1). Alternatively (model 2), it was suggested that CR does not alter the NAD:NADH ratio, but rather acts by increasing expression of Pnc1, which degrades the endogenous Sir2 inhibitor nicotinamide, thus increasing Sir2 activity (Anderson et al., 2003). In both models, activation of Sir2 leads to increased rDNA silencing and a consequent reduction in recombination between rDNA repeats; this reduces the formation of extrachromosomal rDNA circles (Sinclair & Guarente, 1997) and so retards aging.

In contrast, the observation that CR could extend lifespan in some sir2 deletion strains (Jiang et al., 2002; Kaeberlein et al., 2004) led to the proposal of a third mechanism (Fig. 1, model 3). This postulates that CR extends lifespan in a Sir2-independent manner via inhibition of the Tor and Sch9 kinase signalling pathways (Kaeberlein et al., 2004, 2005c). In this model, the downstream molecular mechanism effecting longevity is not entirely clear, but may be related to reduced ribosomal protein biogenesis (Kaeberlein et al., 2005c; Steffen et al., 2008). Controversially, it has recently been claimed that such Sir2-independent effects of CR are in fact due to activation of Sir2 homologues and so simply result in the same increase in rDNA silencing invoked in models 1 and 2 (Lamming et al., 2005; Medvedik et al., 2007) [but see (Kaeberlein et al., 2006)].

Although CR had no obvious effect on telomeric silencing in spot tests, subtle effects could be missed in this simple visual assay. We therefore performed quantitative silencing assays by counting the number of individual colonies formed on selective FOA plates relative to the total number of colonies on nonselective media, as previously described (Kaeberlein et al., 2005c). Increased silencing should result in increased viability in this assay, but over a series of 6 independent experiments totalling more than 60 000 colonies counted, we found no significant difference between 2%, 0.5% and 0.05% glucose (Fig. 2B). Thus, these quantitative silencing assays confirmed the findings from spot test assays: neither moderate nor severe CR affects telomeric silencing.

Although CR had no effect on telomeric silencing, it nevertheless remained possible that rDNA silencing was indeed increased. We therefore utilized the widely used rDNA reporter strain, JS128, which has a URA3 marker integrated into the nontranscribed spacer region 1 of the rDNA (Smith & Boeke, 1997). Again, partial silencing of this reporter gene enables growth on media lacking uracil and media containing FOA. Deletion of the SIR2 gene inhibits rDNA silencing, resulting in increased expression of the URA3 reporter gene, which is routinely visualized in such rDNA reporter strains as increased growth on −ura plates (Smith & Boeke, 1997; Smith et al., 2007) (Fig. 3A). If CR increases rDNA silencing, this should be seen as decreased growth on −ura plates relative to nonselective media. However, no gross effects were seen in spot tests using 0.5% and 0.05% glucose (Fig. 3A). To quantify these effects, a similar approach was taken to that used for the telomeric reporter assays described above, where the number of viable colonies on −ura plates relative to nonselective media was calculated. Over a series of three independent experiments with over 50 000 colonies counted, no significant difference in rDNA silencing was seen between 2% and 0.5% glucose (Fig. 3B). However, a partial reduction in viability on −ura plates was noted at 0.05% glucose. Such reduced growth on −ura plates could either be due to increased rDNA silencing or to a synthetic growth phenotype caused by the simultaneous imposition of the double stress of limiting glucose and uracil levels. To control for the latter, duplicate assays were performed in parallel using a different selective media, FOA, where increased silencing could not correlate with decreased colony growth (Fig. 3B, grey bars). This revealed a similar growth inhibition at 0.05% glucose (but not at 0.5% glucose), indicating that the partial reduction in viability on −ura media observed is not due to increased rDNA silencing. Confirmation of the artefactual nature of this reduced growth at 0.05% glucose was evident by comparison with a positive control sir2Δ strain (Fig. 3B), which exhibited the expected reciprocal relationship between growth on the two media (increased on −ura, decreased on FOA). Thus, both spot test and quantitative silencing assays indicate that moderate and severe CR have no effect on rDNA silencing.

The correlation between increased rDNA silencing and longevity is integral to Sirtuin-dependent models of yeast lifespan extension. As we found no effect of CR on rDNA silencing, we investigated whether deletion mutations that reproducibly extend lifespan in multiple yeast background strains alter rDNA silencing. Deletion of HXK2, one of three hexokinases in yeast, limits the entry of glucose into glycolysis and hence is widely used as a genetic mimic of CR acting upstream of Sir2/Tor/Sch9. Indeed, like CR, HXK2 deletion has been shown to increase lifespan in multiple yeast genetic backgrounds (Lin et al., 2000; Kaeberlein et al., 2005b). Deletion of FOB1 is thought to increase lifespan by reducing rDNA recombination [and hence extrachromosomal rDNA circle (ERC) generation] downstream of Sir2 (Defossez et al., 1999; Kaeberlein et al., 2005b). In Fig. 4A, it can be seen that the hxk2Δ strain behaves similarly to the JS128 parent strain on all media, indicating no major effect on rDNA silencing. However, the fob1Δ strain exhibited enhanced growth on −ura plates, and decreased growth on FOA plates, similar to the sir2Δ strain, indicating a loss of rDNA silencing. In contrast, deletion of RPD3, which is known to increase rDNA silencing in a Sir2-dependent manner (Sun & Hampsey, 1999), gave the expected decrease in growth on −ura plates, thus confirming that this assay is able to detect such increases in rDNA silencing. To check if these effects were rDNA-specific, we deleted FOB1 and HXK2 in a control strain (JS122) where the URA3 reporter gene is integrated in a nonsilenced genomic locus. Here, there was no difference in growth on selective media between the parent strain and its isogenic mutants: all grew similarly on −ura plates, none grew on FOA (Fig. 4B). Finally, we investigated whether telomeric silencing was similarly affected by creating deletion mutants in the AEY1017 strain. In spot test assays, the hxk2Δ strain was similar to its isogenic parent strain and to a fob1Δ strain (Fig. 4C). Therefore, the hxk2Δ genetic mimic of CR, like CR itself, has little effect on silencing at rDNA or telomeric loci, whereas deletion of FOB1 specifically inhibits rDNA silencing.

Despite these different effects on silencing, both CR and deletion of FOB1 have been reported to extend lifespan via inhibition of rDNA recombination. We therefore assayed mitotic rDNA recombination by monitoring the frequency of URA3 marker loss after serial culturing for approximately 120 generations (Dror & Winston, 2004). Over a series of experiments, the mean frequency of marker loss in the parent JS128 strain at 2% glucose was 1.3 × 10⁻³ per generation, similar to previously reported values for other strains (Merker & Klein, 2002; Lamming et al., 2005). Deletion of FOB1 reduced the frequency of URA3 marker loss by around 90%, whereas deletion of SIR2 caused an approximately threefold increase in marker loss (Fig. 5A), consistent with the known pro- and anti-recombinase functions of Fob1 and Sir2, respectively, at the rDNA (Merker & Klein, 2002). These effects on URA3 marker loss were specific for the rDNA locus, as no marker loss was observed over 120 generations in the JS122 parent strain where URA3 is integrated outside the rDNA, or in its isogenic JS122 fob1Δ and sir2Δ strains (Fig. 5A). These data confirmed the validity of assaying the frequency of URA3 marker loss in the JS128 strain as a readout of rDNA recombination rate, thus enabling the effects of CR to be determined. Both moderate (0.5% glucose) and severe (0.05% glucose) CR produced a significant reduction in rDNA recombination of around 20% (Fig. 5B). Therefore, both CR and FOB1 deletion act via a silencing-independent mechanism to reduce rDNA recombination.

Finally, we investigated whether the Sir2-independent Sch9/Tor1-mediated longevity pathway of model 3 might also impact on rDNA silencing and recombination. Deletion of SCH9 appeared to slightly improve growth on −ura media relative to nonselective media in qualitative spot test assays (Fig. 6A), similar to the general CR mimic hxk2Δ (Fig. 4A, ​,6A);6A); whereas, deletion of TOR1 appeared to slightly decrease growth on −ura plates (Fig. 6A). Although these small effects were also apparent in quantitative silencing assays (Fig. 6B), the differences in viability on −ura plates between the various deletion mutants and the isogenic parent strain were not statistically significant. Therefore, like CR itself, genetic mimics of CR do not cause significant increases in rDNA silencing. In contrast, the hxk2Δ, sch9Δ and tor1Δ strains all exhibited significantly reduced rDNA recombination (Fig. 6C), reinforcing the notion that various lifespan-extending interventions can reduce rDNA recombination independently of rDNA silencing.

How can we reconcile our data with earlier work that established models 1 and 2? While there is no doubt that Sir2 histone deacetylase activity is of primary importance in establishing rDNA silencing, a survey of the literature reveals surprisingly little data directly demonstrating that CR affects rDNA silencing. Indeed, to our knowledge, the only such data is Fig. 1d in (Lin et al., 2002), where the result of a qualitative assay using MET15 as a reporter gene inserted in the rDNA was presented. The increased brown colouration of this strain on 0.5% glucose compared with 2% glucose was consistent with an effect of moderate CR increasing rDNA silencing. However, such assays are not quantifiable and are subject to interpretations over ‘shades of brownness’. Furthermore, similar effects of CR on MET15 reporter gene expression are observed when the marker is integrated into nonsilenced regions of the genome, suggesting that the increased pigmentation observed is a metabolic effect unrelated to silencing (Smith et al., 2009). The data presented here, using both qualitative and quantitative assays of URA3 reporter gene expression, strongly argue that CR has no effect on rDNA silencing. As CR has no effect on telomeric silencing either (this study and Kaeberlein et al., 2005c), it seems unlikely that activation of Sir2 (or its homologues) occurs in response to CR. Furthermore, rDNA silencing is not significantly affected in the CR genetic mimics, hxk2Δ, sch9Δ and tor1Δ; whereas fob1Δ strains actually exhibit greatly reduced rDNA silencing. As CR and all these deletion mutants have been shown to increase lifespan in various genetic backgrounds, these observations further emphasize the lack of correlation between rDNA silencing and lifespan, and so make it unlikely that Sirtuin-mediated rDNA silencing is crucial for longevity.

What implications do our data have for the competing theories of CR outlined in Fig. 1? It is important to note that although our data clearly argue against the essential role for Sir2 activation and increased rDNA silencing implicit in models 1 and 2, they are nevertheless consistent with the proposal that CR extends lifespan by reducing rDNA recombination, albeit via a silencing-independent mechanism. Indeed, the observation that FOB1 deletion inhibits rDNA silencing (this study and Huang & Moazed, 2003), clearly demonstrates that these two processes can be uncoupled; as FOB1 deletion strongly decreases rDNA recombination (this study and Defossez et al., 1999; Kobayashi et al., 1998). As Fob1 is a component of the RENT complex that targets Sir2 to the nucleolus (Huang & Moazed, 2003), the consequent loss of Sir2 from the rDNA explains the reduced silencing in FOB1 deletion strains, but this in turn inevitably suggests that the decrease in rDNA recombination is Sir2- and silencing-independent. It has previously been shown that rDNA recombination is reduced by moderate CR, rapamycin and by the putative CR genetic mimic cdc25-10 (Lin et al., 2000; Lamming et al., 2005; Medvedik et al., 2007). We report here that both moderate and severe CR, as well as deletion of the HXK2, SCH9 and TOR1 genes all reduce rDNA recombination. Thus, it may be that models 1 and 2 are essentially correct in that CR ultimately extends lifespan, at least in part, by decreasing rDNA recombination, but that this occurs by an (unknown) Sirtuin-independent process. In both models, the longevity associated with reduced rDNA recombination is attributed to reduced ERC generation. While this remains a plausible mechanism, it is interesting to note that in fob1/sir2 double mutants, which have extremely low ERC levels, CR increases lifespan and reduces rDNA recombination (Kaeberlein et al., 1999, 2004; Lamming et al., 2005). In addition, lifespan reduction is associated with increased rDNA recombination in the face of unchanged ERC levels in sgs1, hpr1 and dna2 mutants (Heo et al., 1999; McVey et al., 2001; Hoopes et al., 2002; Mankouri et al., 2002; Merker & Klein, 2002). Further work is required to determine if the CR-induced reduction in rDNA recombination impacts on longevity via ERCs or a different mechanism.