Epigenetic alteration has been implicated in aging. However, the mechanism by which epigenetic change impacts aging remains to be understood. H3K27me3, a highly conserved histone modification signifying transcriptional repression, is marked and maintained by Polycomb Repressive Complexes (PRCs). Here, we explore the mechanism by which age-modulated increase of H3K27me3 impacts adult lifespan. Using Drosophila, we reveal that aging leads to loss of fidelity in epigenetic marking and drift of H3K27me3 and consequential reduction in the expression of glycolytic genes with negative effects on energy production and redox state. We show that a reduction of H3K27me3 by PRCs-deficiency promotes glycolysis and healthy lifespan. While perturbing glycolysis diminishes the pro-lifespan benefits mediated by PRCs-deficiency, transgenic increase of glycolytic genes in wild-type animals extends longevity. Together, we propose that epigenetic drift of H3K27me3 is one of the molecular mechanisms that contribute to aging and that stimulation of glycolysis promotes metabolic health and longevity.https://doi.org/10.7554/eLife.35368.001
Aging is characterized by the progressive decline in cellular and organismal functions that lead to reduction of fitness and increased risks to diseases and death. Epigenetic alterations in histone modification represent a prominent hallmark of aging (Sen et al., 2016). Chromatin and epigenetic complexes that modify histones regulate the accessibility of DNA to transcriptional machinery, thereby permitting direct control of gene expression (Kouzarides, 2007). Previous studies have established that changes in histone markings, such as acetylation on histone H4 at lysine 16 (H4K16ac) in Saccharomyces cerevisiae (Dang et al., 2009), trimethylation on histone H3 at lysine 4 (H3K4me3) in Caenorhabditis elegans (Han et al., 2017), and trimethylation on histone H3 at lysine 36 (H3K36me3) in S. cerevisiae, C. elegans and Drosophila melanogaster substantially impact adult lifespan (Pu et al., 2015; Sen et al., 2015). Polycomb repressive complexes (PRCs), including PRC1 (Shao et al., 1999) and PRC2 (Kuzmichev et al., 2002), are important histone-modifying enzymes (Czermin et al., 2002; Müller et al., 2002; Saurin et al., 2001) that are evolutionally conserved from the worm to Drosophila and mammals (Whitcomb et al., 2007). Tri-methylated histone H3 at lysine 27 (H3K27me3) denotes transcriptional silencing, which is produced by PRC2 (Czermin et al., 2002; Müller et al., 2002) and functionally maintained by PRC1 (Cao et al., 2002). Although mutation in the H3K27me3 demethylase couples an increase of H3K27me3 with longevity in C.elegans (Jin et al., 2011; Maures et al., 2011), reducing H3K27me3 by heterozygous mutation in the H3K27me3 methyltransferase results in lifespan extension in both C. elegans and D. melanogaster (Ni et al., 2012; Siebold et al., 2010). Hence, it remains unclear whether the life-benefit phenotype is due to manipulation of specific genes or modulatory effect of H3K27me3. Importantly, H3K27me3 levels are known to increase with age (Liu et al., 2013; Sun et al., 2014); however, the manner by which H3K27me3 epigenome changes over the course of lifespan and its consequent impact on the progression of aging have not been addressed.
Metabolic homeostasis is intimately connected with aging and lifespan regulation (López-Otín et al., 2016). Glucose is the dominant provider of energy for cells. Glycolysis, the first step in the breakdown of glucose to extract energy in the form of ATP for cellular metabolism, is present in all cellular organisms and critical for life. The anti-aging effect of caloric restriction correlates with a metabolic feature of limiting glucose metabolism including a reduction of glycolysis (Feuers et al., 1989). On the other hand, glycolysis is essentially involved in a wide-range of biological processes (Chang et al., 2013; Okamoto et al., 2001; Volkenhoff et al., 2015). It is unclear, however, how the beneficial outcome of restricting glucose metabolism and glycolysis as a result of caloric restriction could be reconciled with the essential role of glycolysis during normal adult lifespan. Strikingly, decreased brain glucose metabolism is present prior to the onset of clinic symptoms in individuals at risk of Alzheimer's disease (Cunnane et al., 2011). Since aging is the most significant factor for human neurodegenerative diseases, it is important to better understand how glycolysis is modulated with age and whether modulation of glycolytic process could impact adult lifespan.
Mounting evidence associates aging with epigenetic, transcriptional, and metabolic alterations (López-Otín et al., 2013), yet how these processes are integrated into the regulation of adult physiology is only beginning to be revealed. Although epigenetic effects could potentially elicit broad regulatory mechanisms, it remains to be determined whether epigenetic modulations are simply associated with aging or directly contribute to transcriptional and metabolic modulations that profoundly impact adult lifespan. Here, using Drosophila, we discover that adult-onset fidelity loss results in epigenetic drift of H3K27me3, which in turn impinges on the transcription and metabolism of the glycolytic pathway. Our data presented in this study are consistent with the model that epigenetic drift of H3K27me3 is one of the molecular mechanisms that contribute to at least a subset of age-associated phenotypes and limit lifespan in Drosophila. Moreover, we reveal that a reduction of H3K27me3 by PRC2-deficiency stimulates glycolysis and healthy lifespan. Our study proposes that, during natural aging, stimulation of glycolysis promotes metabolic health and longevity.
H3K27me3 is a signature histone modification, but its dynamics at the epigenome level during aging has not been studied. To characterize H3K27me3 during aging, we examined the level of H3K27me3 in muscle by western blot and revealed an age-associated increase in wild-type (WT) flies (Figure 1—figure supplement 1A). To quantitatively profile H3K27me3, we adapted the chromatin immunoprecipitation (ChIP) followed by high-throughput DNA sequencing with reference exogenous genome method (ChIP-Rx) (Bonhoure et al., 2014; Orlando et al., 2014a) in fly muscle tissues (Figure 1—figure supplement 1B). We generated H3K27me3 epigenome profiles in adult flies of 3, 15, and 30d. In addition, we also sampled embryo, larvae, and pupae to examine H3K27me3 epigenome profiles during development. Surprisingly, despite the fact that H3K27me3 levels increased with age, H3K27me3 peak profiles in the aging adult remained largely unchanged (Figure 1A). In a sharp contrast, H3K27me3 signals during developmental stages showed dramatic gain or loss (Figure 1B). This evidence highlights a distinct, yet unknown mechanism of H3K27me3 modification that occurs during aging.
To define the landscape of aging epigenome, we divided distribution of H3K27me3 into peak regions (IP/input ≥ 2 and signal spanning ≥3 kb) and inter-peak regions. Analysis in 3d WT revealed 222 peak regions reproducibly identified from a total of four replicates, which covered approximately 9.73 Mbp of the genome (Figure 1—figure supplement 1C, Supplementary file 1). Analysis of global occupancy of H3K27me3 between young and aged WT exhibited comparable patterns (pair-wise Pearson correlation coefficients ≥ 0.70) (Figure 1—figure supplement 1D), demonstrating that the genome-wide distributions of H3K27me3, particularly the peak regions, are maintained in aged WT.
We next analyzed signal intensity of individual genes using reference-adjusted read counts. Scatter plots demonstrated that H3K27me3 signals in WT clearly had an adult-onset, progressive increase (Figure 1C and D). To further characterize H3K27me3 dynamics, we sorted individual genes into peak genes (1141 genes) according to the peak regions and genes therein covered (see Supplementary file 1 for peak regions and included genes) or inter-peak genes (16370 genes). In aging WT flies, despite a genome-wide increase, we noted that inter-peak genes gained relatively more signals than those of peak genes (Figure 1E). To quantify the H3K27me3 marks gained specifically during aging, we subtracted the signals of 3d from those of older age points. Genome browser views illustrated that subtracted signals propagated along the chromosome, showing no direct correlation with pre-existing peaks (Figure 1F). We then extended this analysis to the entire epigenome. Modification in flies that were 3d old was predominantly biased at the peaks, with approximately 323% more signals as compared to the inter-peaks (Figure 1G). In contrast, the signals acquired during aging were only about 9% more at the peaks than those at the inter-peaks (Figure 1G), suggesting that there was a dramatic shift in the pattern of H3K27me3 modification in aging. Altogether, analysis of H3K27me3 aging dynamics implicates a strong selection on peak regions compared to inter-peak regions when animals are young; however, adult-onset modification of H3K27me3 becomes less biased at the peak regions.
Given that above study indicated a new feature of H3K27me3 dynamics at the epigenome level during adult lifespan, we further explored the roles of epigenetic pathways in the aging process via an unbiased deficiency-based genetic screen. Using CRISPR/Cas9 mutagenesis (Ren et al., 2013), we systematically generated site-specific deletion mutants of 24 key regulators involved in distinct epigenetic and chromatin modifications in the same homogenous genetic background (Supplementary file 2). Since the effects of epigenetic genes are usually dose-dependent, we analyzed adult survival using heterozygous mutants. Lifespan analysis demonstrated that whereas majority of mutants showed mild or no effect, animals deficient in PRCs, including esc, E(z), Pcl, Su(z)12 of PRC2 and Psc and Su(z)2 of PRC1, lived substantially longer (Figure 2A, Figure 2—figure supplement 1A, Supplementary file 2). Roles of select PRC2 factors have been previously implicated in Drosophila lifespan regulation (Siebold et al., 2010), although the underlying mechanism remains to be addressed. In addition, it was still possible that life-extension might be due to manipulations of individual genes, but not linked to the epigenetic change of H3K27me3. We further examined pairwise combinations of PRC2 double mutants in trans-heterozygosity. Remarkably, double mutants demonstrated more striking effects on H3K27me3-reduction and life-extension (Figure 2B, Figure 2—figure supplement 1B). Identification of multiple components of PRC2 and PRC1 in the screen is interesting, suggesting the involvement of the epigenetic changes regulated by these two complexes in animal lifespan.
We then investigated whether the age-delaying effect of PRCs-deficiency might be due to its ability to modulate H3K27me3 dynamics with age. Using muscle tissues, we examined the modification of H3K27me3 in young (3d) and aged (30d) mutants with their age-matched WT controls. For this and subsequent experiments, we studied the PRC2 deficiency using Pclc421 Su(z)12c253 trans-heterozygote double mutants, which gave rise to the strongest effect in H3K27me3-reduction and life-extension. Of note, western blot analysis showed that only H3K27me2/3 were selectively reduced in PRC2 mutants, while other histone markings remained unchanged (Figure 3—figure supplement 1A). Interestingly, global occupancy of H3K27me3 between WT and age-matched PRC2 mutants was preserved, as demonstrated by the comparable patterns of peak profiles and genes known to be H3K27me3-modified (pair-wise Pearson correlation coefficients ≥ 0.77) (Figure 3A, Figure 3—figure supplement 1B and C), although we cannot rule out the possibility that certain H3K27me3 targets might be altered. H3K27me3 signals in WT were generally higher than those in mutants (Figure 3B, see Figure 3—figure supplement 1D for escc289 E(z)c239). In PRC2 mutants at 3d where H3K27me3 bulk levels were substantially decreased, interestingly, the signals became more likely to coalesce into the peak regions (Figure 3C, see Figure 3—figure supplement 1E for escc289 E(z)c239). Given limited H3K27me3 repertoire in the mutants, this shift might be critical for the establishment of the peaks. In aging PRC2 mutants, adult-onset modification was less biased or even decreased at the peaks (Figure 3D, see Figure 3—figure supplement 1F for escc289 E(z)c239). More strikingly, age-associated shift in H3K27me3 modification that occurred in WT was now dampened in mutants (Figure 3E, see Figure 3—figure supplement 1G for escc289 E(z)c239). This evidence potentially indicates that PRC2-mutation mitigates the drifting of H3K27me3 that occurs during natural aging.
To extend this analysis beyond muscle tissues, we further profiled H3K27me3 using dissected heads of Pclc421 Su(z)12c253 mutants and their age-matched WT controls. Similarly, our data shows that H3K27me3 in heads has age-associated increase and drift during normal aging in WT animals, and that this shift can be partially diminished by PRC2-mutation (Figure 4, Figure 4—figure supplement 1).
Since biological activities of histone modification often link chromatin accessibility to gene expression (Kouzarides, 2007), we speculated that alterations in the levels of H3K27me3, a repressive epigenetic mark in transcription, could impact the transcriptome. We asked whether transcriptional change of particular genes might account for PRC2-dependent life-extension. Using muscle tissues, we generated RNA-seq datasets for polyA-selected mRNAs. Analysis of individual PRC2 mutants resulted in several hundreds of upregulated genes, most of which were those at the inter-peak regions (Figure 5—figure supplement 1A). On the other hand, analysis of 63 genes with known effects on aging, including genes in insulin/IGF-1, mTOR pathways, etc., revealed no consistent changes in the expression in PRC2 mutants (Supplementary file 3), thus excluding them from being the major contributors in mediating PRC2-dependent longevity. Interestingly, comparative transcriptomics converged on a common set of genes (Figure 5A, Supplementary file 3). Gene ontology (GO) analysis revealed that the ‘glycolytic process’ and closely related pathways were highlighted for genes upregulated (Figure 5A), while the ‘oxidation-reduction process’ was enriched for genes downregulated (Figure 5A). Using weighted gene co-expression network analysis (WGCNA) (Langfelder and Horvath, 2008), we determined the significance of highly correlated changes in the expression of glycolytic genes (Figure 5B). We then extended RNA-seq analysis to the heads, and further narrowed down to two glycolytic genes, Tpi and Pgi, whose expressions were upregulated in both tissue types across individual long-lived PRC2 mutants (Figure 5—figure supplement 1B and C, Supplementary file 4). As noted, glucose is metabolized through sequential reactions of glycolysis and the citric acid cycle. However, we found no consistent changes in the expression of genes mediating pentose phosphate pathway (PPP), citric acid cycle, and oxidative phosphorylation (Figure 5—figure supplement 1C, Supplementary file 4), suggesting modulatory effects on the glycolytic genes.
Moreover, RNA-seq analyses of WT animals revealed that, although aging might exert a much broader effect on the changes of the transcriptome, genes downregulated with age were mainly enriched in carbohydrate metabolism and energy production (Figure 5C). In particular, we noted a transcriptional decline of genes related to glycolysis/gluconeogenesis in old animals (Figure 5D). To verify these results, we subsequently performed single-gene analysis on Tpi and Pgi with respect to their H3K27me3 states and mRNA levels using ChIP-qPCR (Figure 5E, Figure 5—figure supplement 1D) and qRT-PCR (Figure 5F and G), respectively. In addition to PRC2 mutants, we attested the transcriptional increase of these two glycolytic genes in Su(z)2c433, a long-lived PRC1 mutant (Figure 5H).
Given the fact that H3K27me3 naturally increased with age, we investigated whether further increase of H3K27me3 could impact lifespan and gene expression. To address this, we examined flies deficient in histone demethylation. The Drosophila Utx gene has been previously implicated in catalyzing the removal of methyl groups from H3K27me3 (Smith et al., 2008); a null mutation, utxc499 (Supplementary file 2), displayed a further increase of H3K27me3 compared to age-matched WT (Figure 5E and I, Figure 5—figure supplement 1D). Flies lacking utx were viable, but detailed characterization indicated a significantly shortened lifespan, to about 70% that of WT animals (Figure 5J). Concomitant with an increase in H3K27me3 modification, expression of both Tpi and Pgi was further decreased (Figure 5K). Finally, control studies using short-lived miR-34 mutants (Liu et al., 2012) and a long-lived piwi deficiency that was identified from our unbiased genetic screen (Supplementary file 2) demonstrated that transcriptional changes of the glycolytic genes had no simple correlation with lifespan alterations (Figure 5—figure supplement 1E and F).
Altogether, these data reveal that H3K27me3 dynamics with natural aging couples with a downregulation of glycolytic genes, and this reduction is enhanced in utx mutants upon further increase of H3K27me3. Long-lived PRC2 mutants partially reverse the decline in the expression of glycolytic genes that are otherwise significantly decreased with age.
To further investigate the mechanism by which epigenetic changes regulate aging, we conducted another unbiased study by profiling cellular metabolites using LC-MS-based untargeted metabolomics (Patti et al., 2012). In total, 151 metabolites were identified. This experiment demonstrated changes in the levels of metabolites during aging of WT flies, and between WT and mutants (Figure 6A and C, Figure 6—figure supplement 1A). Pathway analysis revealed that glycolysis among additional metabolic processes was significantly enriched (Figure 6B and D). In particular, lactate, a specific indicator for anaerobic glycolysis (Schurr and Payne, 2007), was significantly decreased during normal aging in WT animals but became elevated in PRC2 mutants (Figure 6E). Of note, metabolites related to citric acid cycle remained unchanged between WT and mutants (Figure 6—figure supplement 1B). Combined, this evidence reveals a metabolic profile reflective of modulated glycolysis, being decreased during normal aging but increased in long-lived PRC2 mutants. Thus, this unbiased study using metabolomic approach showed modulation in the level of lactate and glycolysis during aging and in PRC2 mutants, in good congruence with the conclusion from the studies of epigenomic and transcriptional alterations (Figure 5—figure supplement 1C and D).
To provide quantitative insights into the changes in glucose metabolism with age and in PRC2 mutants, we devised a metabolic flux assay using 13C-labeled glucose as a tracer in adult aging flies. For young (3d) and aged (25d) animals, fly diet was switched from 12C-glucose to 13C-glucose, such that the glucose metabolism can be traced by measuring the 13C-labeled metabolites (Figure 6F, Figure 6—figure supplement 2A). After 5 days of feeding, 13C-labeled glucose accounted for more than 93% of the total glucose pool (Figure 6—figure supplement 2B), suggesting a near-complete replacement. This experiment demonstrated that while the level of labeled glucose was similar (Figure 6—figure supplement 2C), specific metabolites (Figure 6G) and the glycolytic pathway as a whole (Figure 6H) exhibited an age-associated decline; strikingly, the extent of decline was partially reversed by PRC2-deficiency. This finding, together with above results obtained by untargeted metabolomics, provides definitive evidence that glycolysis is modulated by aging. Since metabolites of the citric acid cycle are mix lineages from glycolysis, glutamate, and fatty acids, each metabolite may contain multiple 13C-labeled isotopologues via glycolysis-derived reactions as well as recurrent cycling (Figure 6F). Thus, we analyzed all isotopologues pertaining to each metabolites (Figure 6—figure supplement 2D) and their summed intensity (Figure 6—figure supplement 2E). Our data indicated no consistent change of the citric acid cycle with age and in PRC2-deficiency, which was in line with constant levels of metabolites as above measured by untargeted metabolomics (see Figure 6—figure supplement 1B).
These data combined establish a regulatory event from epigenetic alterations to transcriptional and metabolic modulations impacting the manner by which glucose can be metabolized during aging and in PRC2 mutants. While natural aging couples with to a metabolic decline including a reduction of glycolysis, PRC2 deficiency relatively maintains the level of glycolysis as compared to age-matched WT animals.
We next characterized specific metabolites in the glycolytic pathway. While anaerobic glycolysis yields ATP only, glycolysis at the step of pyruvate formation has a net outcome of both ATP and NADH, a reduced form of nicotinamide adenine dinucleotide (NAD+) (see Figure 5—figure supplement 1C). While the overall glucose levels between WT and mutants had no significant difference (Figure 7A), pyruvate was increased in mutants (Figure 7A). On the other hand, the ratio of NADPH/NADP+, an indicator of the PPP, foliate metabolism, and malic enzyme (Fan et al., 2014) was even slightly decreased in mutants (Figure 7B). In parallel, we also examined the glutathione, an important indicator of the cellular redox state (Jones, 2006). Our data showed that ATP, NADH/NAD+, and GSH/(GSH +GSSG) ratio were decreased with age but increased in mutants (Figure 7C and D). Combined, analysis of specific metabolites demonstrates a metabolic decline associated with normal aging, and that PRC2 deficiency promotes ATP and cellular redox potential as compared to age-matched WT controls.
We then asked how this ability of enhanced glycolysis in PRC2 mutants was translated into fitness and stress resistance in aging. To illuminate this, we characterized age-related phenotypes. Analysis of climbing of young (3d) and aged (30d) animals demonstrated that although WT and mutants at 3d behaved similarly, PRC2 mutants at 30d of age had better locomotion (Figure 7E). To examine the sensitivity to oxidative stress, we utilized hydrogen peroxide (H2O2), a source of reactive oxygen species (ROS). We observed substantially enhanced resistance to oxidation in mutants (Figure 7F). Elevated energy and NADH/NAD+ ratio in PRC2 mutants might underlie improved motility and resistance to ROS, respectively. These data implicate that PRC2-deficiency couples longevity and metabolism with enhanced ability to handle stress, a property reflecting improved adult fitness.
We subsequently determined whether perturbing glycolysis could diminish the lifespan benefits of PRC2 mutants. In PRC2 mutants, Tpi and Pgi were two genes exhibiting a concerted decrease in H3K27me3 modification (see Figure 5E, Figure 5—figure supplement 1D) and a corresponding increase in transcription (see Figure 5G, Figure 5—figure supplement 1C); thus Tpi and Pgi might be important targets of PRC2. In light of this, we generated a null mutation of the Tpi gene by CRISPR/Cas9, Tpic511 (Supplementary file 2). Tpi gene, encoding triosephosphate isomerase, is a central glycolytic enzyme (Bar-Even et al., 2012). Given homozygous Tpic511 had a pre-adult lethality, we combined Tpic511 heterozygote with Pclc421 Su(z)12c253, and evaluated the adult phenotypes by using the triple mutants. Control studies indicated that the Tpic511 heterozygote alone retained normal lifespan (Figure 8—figure supplement 1A), oxidative stress (Figure 8—figure supplement 1B), locomotion (Figure 8—figure supplement 1C), and comparable levels of ATP and NADH/NAD+ ratio (Figure 8—figure supplement 1D). Lowering Tpi in PRC2 mutants slightly decreased the expression of Tpi gene (Figure 8—figure supplement 1E) and diminished the extent by which glycolysis could be stimulated as having relatively reduced ATP and NADH/NAD+ ratio (Figure 8D). Analysis of lifespan exhibited that Tpi deficiency partially mitigated the longevity phenotype in PRC2 mutants (Figure 8A). Traits related to adult fitness, including resistance to ROS and climbing mobility, were also attenuated by Tpi deficiency (Figure 8B and C). To consolidate this result, we generated a null mutation for Pgi by CRISPR/Cas9, Pgic392 (Supplementary file 2). Consistently, while Pgic392 heterozygote alone had minimal effect on aging (Figure 8—figure supplement 1F–1I), triple mutants combining Pgic392 heterozygote with Pclc421 Su(z)12c253 diminished the pro-lifespan benefits mediated by PRC2-deficiency (Figure 8E–8H, Figure 8—figure supplement 1J). These data implicate that perturbing glycolysis by genetic mutation diminishes longevity traits in PRC2 Mutants. Given the effect is partial, we cannot rule out the possibility that other mechanisms might be involved.
Our data thus far have shown the effect of glycolysis in PRC2-dependent life-benefits. Furthermore, we investigated whether upregulating glycolytic genes in WT animals could be sufficient to extend lifespan. To address this, we increased the gene dosage of Tpi and Pgi via genomic transgenes, which expressed these genes under their endogenous regulatory elements (Figure 9—figure supplement 1A and B). Western blot analysis manifested that both Tpi and Pgi proteins had a modest decrease with age, but became upregulated upon PRC2-deficiency, an outcome consistent with the alteration in the level of H3K27me3 (Figure 9A–9D). We then interrogated the effects of Tpi and Pgi transgenes on WT flies. While single transgenes had modest phenotypes, animals combining both Tpi and Pgi transgenes stimulated glycolysis, as shown by elevated pyruvate, ATP, and NADH/NAD+ ratio as compared to age-matched WT (Figure 9E). Accordingly, adult aging phenotypes, including lifespan (Figure 9F), locomotion (Figure 9G), and resistance to oxidative stress (Figure 9H), were improved.
Taken together, these data suggest that upregulation of glycolytic genes alone recapitulates antiaging features of PRC2 mutants. This finding, combined with data obtained from H3K27me3 dynamics and consequently its effects on gene expression and metabolism, underscores the mechanistic link between epigenetic, transcriptional, and metabolic processes in aging, further heightening the role of glycolysis in promotion of metabolic health and longevity.
How changes in epigenomics, transcription, and metabolomics are integrated into the regulation of adult lifespan is a long-standing question in the biology of aging. Our study provides a new paradigm by which epigenetic drift of H3K27me3, a highly conserved histone mark, links glycolysis to healthy longevity (Figure 9I). Though H3K27me3 levels have been shown to increase with age in mammals (Liu et al., 2013; Sun et al., 2014) and flies (current study), the epigenomic feature of such changes is unknown. Our quantitative assessment shows that aging couples with a decline of fidelity in H3K27me3 modification, as demonstrated by a genome-wide drift of this repressive mark. Concomitant to this drift, glycolytic genes are downregulated, which in turn causes a reduction of glycolysis and glucose metabolism. Importantly, the age-delaying effect of PRCs-mutation likely arises from its ability to slow down or reverse this trend. Therefore, we suggest that epigenetic drift of H3K27me3, enhanced by its age-modulated increase, is one of the molecular mechanisms that drive the progression of aging. Expression of Tpi and Pgi of glycolytic genes is inherently regulated by the alterations of H3K27me3 levels, as shown by their corresponding changes during natural aging, in utx mutants as well as in PRCs-deficiency. We propose that H3K27me3 dynamics may elicit changes in local chromatin environment—especially inter-peak regions that are potentially more sensitive to the gain or loss of the epigenetic mark—thus enabling fine-tuning of target gene expression.
It is unclear how aging might promote the drift in H3K27me3 modification. One possible reason is age-associated DNA damage that reduces the fidelity of epigenetic markings. It has been well-established that aging is associated with the accumulation of DNA damage (Lu et al., 2004). In Bombyx mori, UV-C irradiation can induce the increase of H3K27me3 mediated by PRC2 (Li et al., 2014). In Neurospora crassa, on the other hand, the increase and redistribution of H3K27me3 can be induced by the loss of H3K9 methyltransferase complex (Basenko et al., 2015). Thus, dysregulation of H3K27me3 may be due to feed-forward interactions between DNA damage and changes in the levels of complexes that control epigenetic markings during aging.
Metabolic control has emerged as a critical player to determine the aging outcome. Previous data implicates that impairing glycolysis can extend lifespan perhaps by triggering compensatory processes (Schulz et al., 2007). Quantitative proteomic analysis in mouse heart shows that aging correlates with an increased abundance of proteins in glycolysis, and that caloric restriction and rapamycin treatment known to extend longevity can reverse this trend (Dai et al., 2014). D-Glucosamine supplementation that impairs glucose metabolism can extend lifespan in C. elegans and mouse (Weimer et al., 2014). However, rat models supplemented with 2-deoxyglucose, a known anti-glycolytic compound, are instead short-lived (Minor et al., 2010). Moreover, glucose-enriched diet can prevent pathophysiological decline and early death of telomere dysfunctional mice by stimulating glucose metabolism including glycolysis (Missios et al., 2014). Importantly, recent evidence from studies on a wide range of species suggests that a balanced diet from protein to carbohydrate and their interactive effects are key elements for healthy longevity. Mouse dietary studies have demonstrated that animals fed on diets that are low in protein and high in carbohydrate enjoy longer life with lower blood pressure, improved glucose tolerance and lipid profiles (Solon-Biet et al., 2014). This is also supported by human data on the deleterious effects of diets with high protein and low-carbohydrate (Floegel and Pischon, 2012; Lagiou et al., 2012). Hence, the impact of glucose metabolism including glycolysis on natural aging might be poised to integrate age-associated physiology; the precise outcome must be rather complicated, which could be influenced not only by diverse spatiotemporal need in normal or pathological conditions, but also upon newly evolved biological activities correlated with the increased complexity from nematode to Drosophila and mammals. It is important to note that since our study is based on Drosophila somatic tissues that are primarily post-mitotic cells in adult animals, it is possible that the effects are more dominant in the fly model than others. Our findings in Drosophila propose that, during natural aging, the ability to maintain glycolysis is an alternative avenue to ensure a healthy level of physical capability by delaying the bioenergetic decline with advanced age as which otherwise might form a vicious cycle from decrease in physical activity to hypometabolism, and perhaps disease susceptibility.
Aging is a complex process that can be regulated by a network of multiple mechanisms. It has been well-established that enhancing NAD+ biogenesis promotes healthy lifespan (Anderson et al., 2002; Balan et al., 2008; Mills et al., 2016). As noted, supplementation of NAD+ precursors profoundly elevates energy metabolism by increasing the expression of genes in the TCA cycle as well as glycolysis in C. elegans (Mouchiroud et al., 2013), and promotes glucose metabolism with increased flux through pentose phosphate and glycolytic pathways in mice on a high-fat diet (Mitchell et al., 2018). Therefore, it would be interesting to test whether the life-benefit effects of NAD+ might be through at least in part by the activation of glycolysis. Intriguingly, while declining intracellular NAD+ and thus increased NADH/NAD+ ratio correlate with aging (Zhu et al., 2015), our experiments in Drosophila demonstrate that increased ratios of GSH/(GSH +GSSG) and NADH/NAD+ due to enhanced glycolytic activities may provide a simple but effective way to retard aging. The oxidative phosphorylation, although produces more ATP than glycolysis, can yield intracellular ROS. The accumulation of ROS is the leading proposed cause of decline in cellular function and integrity in aging (Balaban et al., 2005). Thus, modulating H3K27me3 may reprogram bioenergetic decline during aging, which in effect reduces cellular damage and deterioration. Importantly, mammalian glycolytic genes have also been shown as PRCs targets (Brookes et al., 2012). Future investigations, including in-depth comparative analysis of PRCs and glycolytic pathway in the aging process in both flies and humans, may harness common operative mechanisms that modulate metabolic homeostasis and healthy longevity. Given the reversible nature of epigenetic pathways, this study proffers a tempting strategy against age-associated physiological decline and disease.
Flies were cultured in standard media at 25°C with 60% humidity in a 12 hr light and 12 hr dark cycle unless otherwise specified. Briefly, standard Drosophila food contains sucrose (36 g/L), maltose (38 g/L), yeast (22.5 g/L), agar (5.4 g/L), maizena (60 g/L), soybean flour (8.25 g/L), sodium benzoate (0.9 g/L), methyl-p-hydroxybenzoate: (0.225 g/L), propionic acid (6.18 ml/L), and ddH2O to make up 1 L of the food. The WT line used was 5905 (FlyBase ID FBst0005905, w1118). All fly lines used in this study have been backcrossed with 5905 for five consecutive generations for a uniform genetic background, to assure that phenotypes were not associated with any variation in background.
Fifty heads or 30 muscles were harvested and immediately frozen in liquid nitrogen. Tissues were homogenized by Dounce Homogenizer. For total protein extract, tissues were lysed by RIPA buffer and then treated with five cycles of sonication with 30 s on and 30 s off (Bioruptor Pico, Belgium). Resulted histone extractions were loaded onto a NuPAGE 12% Bis-Tris Gel (Thermo Fisher Scientific, Carlsbad, CA, USA), and then transferred by electrophoresis to a polyvinylidene fluoride membrane (Millipore, Germany). The corresponding primary antibodies were diluted 1:1000 and incubated with the membranes at 4°C overnight, the HRP-labeled secondary antibodies were diluted 1:40000 and incubated for 1 hr at room temperature. The signals were detected using ECL (Thermo Fisher Scientific) by Amersham Imager 600 (GE Healthcare, Sweden). The signal intensity was quantified by ImageQuant TL (GE Healthcare). The relative normalization of histone modifications was normalized to the histone H3 signal. Three biological replicates were done for each experiment.
CRISPR/Cas9 mutagenesis was performed as previously described (Ren et al., 2013). Two sgRNA plasmids for target gene were injected into fly embryo. Single-fly-PCR assays were used to screen mutants. To do this, single fly was homogenized in 50 μL squashing buffer (10 mM Tris buffer (pH 8.5), 25 mM NaCl, 1 mM EDTA, 200 μg/mL Proteinase K), then incubated at 37°C for 30 min, followed by 95°C, 10 min for inactivation. Screen primers for each target genes were listed in Supplementary file 2. The virgin females carrying the deletion were backcrossed into WT (5905) male flies for five consecutive generations for a uniform genetic background, to mitigate background effects.
Genomic DNA covering the Tpi and Pgi genes was prepared by PCR-amplification and fusion PCR. DNA was clone into the pBID vector (Wang et al., 2012). Fly microinjection and transgenesis followed standard procedures.
160 male flies for each genotype at 3d or 25d of age were used, with 20 flies per vial. Prior to the test, flies were pre-treated with 1% Agar media for 6 hr before transferred to the vials containing a small piece of Kimwipe filter paper pre-soaked in 1 mL of 10% U-13C6-glucose. Flies were treated for 3 days, and then transferred to new vials with fresh U-13C6-glucose for additional 2 days. Fly heads were dissected for subsequent metabolic analysis. For each genotype, eight biological repeats were conducted, with 20 heads for each repeats.
Adult male flies were collected at the day of eclosion and maintained at 20 flies per vial at 25°C with 60% humidity and a 12 hr light/12 hr dark cycle. Flies were transferred to new vials every other day and scored for survival. The Prism software (Graphpad) was used to draw the survival curve and statistical analysis.
Ten male flies were transferred to an empty vial for 30 min adaption in dark. Flies were tapped to the bottom of the vial; the percentage of flies failed to climb up to a mark at 2 cm from the bottom within 5 s was scored. Ten biological replicates were done for each genotype at given age.
One hundred male flies for each genotype at given age were used, with 20 flies per vial. Prior to the test, flies were pre-treated with 1% Agar media for 6 hr before transferred to the media containing oxidative radical species. For H2O2 treatment, flies were transferred to vials containing a small piece of Kimwipe filter paper pre-soaked in 1.5 ml of 10% glucose and 2% H2O2 (Sinopharm Chemical Reagent). Dead flies were scored every 12 hr, and the Prism software (Graphpad) was used to generate the survival curve and statistical analysis.
For developmental stages, embryo (30 min-1 hr), larva (96 hr), and pupae (7d) were analyzed. For adult stage, 100 fly muscles and 200 fly heads at each time point were used. Samples were collected into a 1.5-mL microcentrifuge tube and immediately frozen in liquid nitrogen. Tissues were pulverized to fine powder using a liquid nitrogen cooled mortar and pestle. Resulted fine power was resuspended in 1.2 mL PBS; 32.4 μL of 37% formaldehyde was added for cross-linking at room temperature for 10 min. To quench formaldehyde, 60 μL of 2.5 M glycine was added. Centrifugation was using 5000 g for 5 min at 4°C. After removing the supernatant, pellet was washed twice with 1 ml PBS containing proteinase inhibitor (Roche). This lysis step was performed at 4°C for more than 1 hr in RIPA buffer (Sigma) supplemented with proteinase inhibitor. The lysate was sonicated using Bioruptor Pico (Diagenode), for 15 cycles with 30 s on and 30 s off. Centrifugation was using 10,000 g for 10 min at 4°C. Supernatant was carefully transferred to a new tube without disturbing the debris. Pre-clear was performed using ChIP-Grade Protein A/G Plus Agarose (Thermo Fisher Scientific) at 4°C for 1 hr. For spike-in epigenome, mouse neuro-2a cells were prepared with the same procedures. Prior to mixing mouse epigenome with that of fly’s, 5% from each preparations was transferred for quantification of DNA mass. To do this, RIPA buffer was added to a final volume of 120 μl, followed by 1 μl RNase A to remove the RNA at 37°C for 30 min. Samples were treated with 0.2 M NaCl (final concentration) at 65°C for 4 hr and 200 ng/μl proteinase K (final concentration) at 55°C for 2 hr. DNA was purified by PCR purification kit (Qiagen), and DNA mass was quantified by Qubit dsDNA HS assay kit (Thermo Fisher Scientific). According to the DNA mass, 5% of the mouse epigenome was added to the fly sample. 10% of sample was transferred to a new tube as ChIP input control. Immunoprecipitation was performed by incubating 3 μg antibody at 4°C for 5 hr with rotation. ChIP-Grade Protein A/G Plus Agarose was added. The sample was incubated overnight at 4°C with rotation. Beads were washed twice with ChIP Wash Buffer (150 mM NaCl, 2 mM EDTA (pH8.0), 20 mM Tris- HCl (pH8.0), 1% Triton X-100, and 0.1% SDS), and another two washes using ChIP Final Wash Buffer (500 mM NaCl, 2 mM EDTA (pH8.0), 20 mM Tris-HCl (pH8.0), 1% Triton X-100, and 0.1% SDS). The elution step was using 120 μL of ChIP Elution Buffer (100 mM NaHCO3 and 1% SDS) by incubating at 65°C for 30 min. Resulted DNA was purified as above described, followed by RNase A treatment. 5–10 ng of DNA was used to generate sequencing library using NEB DNA library prep kit. The quality of libraries was checked by Bioanalyzer 2100 (Agilent). The quantification was performed by qRT-PCR with a reference to a standard library. The libraries were pooled together in equimolar amounts to a final 2 nM concentration. The normalized libraries were denatured with 0.1 M NaOH (Sigma). Pooled libraries were sequenced on the illumina Miseq/Next-seq platform with single end 100 bps.
Thirty muscles and fifty heads were dissected and homogenized in a 1.5-mL tube containing 1 mL of Trizol Reagent (Thermo Fisher Scientific). RNA isolation was followed in accordance with manufacturers instruction. RNA was resuspended in DEPC-treated RNase-free water (Thermo Fisher Scientific). TURBO DNA-free kit was used to remove residual DNA contamination according to manufacturers instruction (Thermo Fisher Scientific). 1 μg of total RNA was used for sequencing library preparation. PolyA-tailed RNAs were selected by NEBNext Poly(A) mRNA Magnetic Isolation Module (NEB), followed by the library prep using NEBNext Ultra RNA library Prep Kit for Illumina according to manufacturer’s instruction (NEB). The libraries were checked and pooled as described above. Pooled denature libraries were sequenced on the illumina NextSeq 550 or Hiseq 2500 platforms with single end 100 bps.
ChIP experiments were performed as described above. The same volume of input and IP DNA were used as template for quantitative PCR experiment. Analysis was performed using the QuantStudio 6 Flex real-time PCR system with SYBR selected master mix (Thermo Fisher Scientific). The percentage of input method was used for data analysis. Primers for target genes were listed in Supplementary file 5.
RNA extractions were done as described above. 1 μg of total RNA was used for reverse transcription by random primers using SuperScript III First-strand synthesis system for RT-PCR (Thermo Fisher Scientific). Analysis was performed using the QuantStudio 6 Flex real-time PCR system with SYBR selected master mix (Thermo Fisher Scientific). The 2-ΔΔCT method was used for quantification upon normalization to the RP49 gene as internal control. Primers for target genes were listed in Supplementary file 5.
Twenty heads for each replicates were collected into a microcentrifuge tube and immediately frozen into liquid nitrogen. For each experiment, 10 biological replicates were prepared. Fly heads were homogenized with 200 μl of H2O and 20 ceramic beads (0.1 mm) using the homogenizer (Precellys 24, Bertin Technologies). 800 μl of ACN:MeOH (1:1, v/v) was added for metabolite extraction. Vortex was for 30 s, followed by incubation in liquid nitrogen for 1 min. This freeze-thaw cycle was repeated three times. To precipitate protein, samples were incubated for 1 hr at −20°C, followed by 15 min centrifugation using 15,000 g at 4°C. The supernatant was removed and evaporated to dryness in a vacuum concentrator (Labconco, German). Dry extracts were then reconstituted in 100 µl of ACN:H2O (1:1, v/v), followed by 10 min sonication (50 Hz, 4°C) and 15 min centrifugation using 13,000 rpm at 4°C to remove insoluble debris. Supernatants were transferred to HPLC glass vials and stored at −80°C prior to LC/MS analysis.
LC-MS was performed using a UHPLC system (1290 series, Agilent Technologies) coupled to a quadruple time-of-flight mass spectrometer (AB Sciex TripleTOF 6600). Waters ACQUITY UPLC BEH Amide columns (particle size, 1.7 μm; 100 mm (length) ×2.1 mm (i.d.)) were used. Mobile phases A = 25 mM ammonium acetate and 25 mM ammonium hydroxide in 100% water, and B = 100% acetonitrile, were used for both ESI positive and negative modes. And the linear gradient eluted from 95% B (0.0–1.0 min), 95% B to 65% B (1.0–14.0 min), 65% B to 40% B (14.0–16.0 min), 40% B (16.0–18.0 min), 40% B to 95% B (18.0–18.1 min), then stayed at 95% B for 4.9 min. The flow rate was 0.3 mL/min and the sample injection volume was 2 μL.
ESI source parameters on TripleTOF 6600 were set as followings: ion source gas 1 (GS1), 60 psi; ion source Gas 2 (GS2), 60 psi; curtain gas (CUR), 30 psi; temperature (TEM), 600°C; ion spray Voltage floating (ISVF), 5000 V or −4000 V, in positive or negative modes, respectively; declustering potential (DP), 60 V or −60 V in positive or negative modes, and collision energy for Product Ion, 30 eV or −30 eV, in positive or negative modes, respectively; collision energy spread (CES), 0 eV. LC-MS/MS data acquisition was operated under information-dependent acquisition (IDA) mode. The instrument was set to acquire over the m/z range 60–1200 Da for TOF MS scan and the m/z range 25–1200 for product ion scan. The accumulation time for TOF MS scan was set at 0.20 s/spectra and product ion scan at 0.05 s/spectra. The unit resolution was selected for precursor ion selection, and the collision energy (CE) was fixed at 30 V and −30 V for positive or negative modes, respectively. Collision energy spread (CES) was set as 0. IDA settings which were set as followings: charge state 1 to 1, intensity 100 cps, exclude isotopes within 4 Da, mass tolerance 10 ppm and maximum number of candidate ions 6. The ‘exclude former target ions’ was set as 4 s after two occurrences. In IDA Advanced tab, ‘dynamic background subtract’ was also chosen.
Metabolites were extracted as described above for untargeted metabolomics.
LC-MS analysis was performed using a UHPLC system (1290 series, Agilent Technologies) coupled to a quadruple time-of-flight mass spectrometer (6550 series, Agilent Technologies). Merck SeQuant ZIC-pHILIC column (particle size, 5 μm; 100 mm (length) ×2.1 mm (i.d.)) was used. Mobile phases A = 25 mM ammonium acetate and 25 mM ammonium hydroxide in 100% water, and B = 100% acetonitrile, were used for both ESI positive and negative modes. And the linear gradient eluted from 80% B (0.0–2.0 min, 0.2 mL/min), 80% B to 20% B (2.0–17.0 min, 0.2 mL/min), 20% B to 80% B (17.0–17.1 min, 0.2 mL/min), 80% B (17.1–22.1 min, 0.4 mL/min), 80% B to 80% B (22.1–22.2 min, 0.2 mL/min). The sample injection volume was 2 μL.
ESI source parameters on QTOF 6550 were set as followings: sheath gas temperature, 300°C; dry gas temperature, 250°C; sheath gas flow, 12 L/min; dry gas flow, 16 L/min; capillary voltage, 2500 V or −2500 V in positive or negative modes, respectively; nozzle voltage, 0 V; and nebulizer pressure, 20 psi. The MS1 data acquisition frequency was set as 4 Hz, and the TOF scan range was set as m/z 60–1200 Da.
Data processing for qualitative metabolic flux analysis was performed using VistaFlux software suite from Agilent Technologies. First, Pathways to PCDL software (version B.07.00, Agilent Technologies) and PCDL Manager software (version B.07.00, Agilent Technologies) was used to build a metabolite library for metabolites in both glycolysis and citric acid cycle. Then, acquired LC-MS raw data files (.d) were loaded into Profinder (version B.08.00, Agilent Technologies) for the extraction of metabolite isotopologues using the constructed metabolite library. The ion abundance criterion for data extraction was set as ‘use peak core area 20% of peak height’. The qualification parameters were set as: mass tolerance: ±15 ppm +2.00 mDa; retention time tolerance :±0.20 min; anchor ion height threshold: 250 counts; sum of ion heights threshold: 1000 counts; correlation coefficient threshold: 0.5. The final results were exported into a csv file for the subsequent comparative and statistical analyses.
Ten fly muscles were harvested and immediately frozen in liquid nitrogen. Tissues were homogenized using the homogenizer (Precellys 24, Bertin Technologies) with 300 μl ddH2O or extraction buffer provided by the kit. For glucose, ATP, GSH and pyruvate assays, tissues were homogenized with ddH2O, then frozen at −80°C for 30 min and thawed at room temperature, followed by 5 cycles of 30 s on and 30 s off sonication. For NADH and NADPH assays, tissues were homogenized with extraction buffer provided by the kit. Lysates were centrifuged using 10,000 g at 4°C for 5 min and the supernatant was transferred to 10 kDa molecular weight cut-off spin filter (Sigma), followed by centrifugation using 10,000 g at 4°C for 30 min for deproteinization. Assays were assembled according to the instruction of manufacturers (Glucose [Sigma, MAHK20], ATP [Promega, FF2000], GSH [Abcam, ab13881], pyruvate [Sigma, MAK071], NADH [Sigma, MAK037], NADPH [Sigma, MAK038]). Signals were detected by the EnSpire Multimode Plate Reader (PerKin Elmer).
Sequencing reads were mapped to the reference genome dm6 or mm10, respectively, with Bowtie2-2.2.9 (Langmead and Salzberg, 2012) using default parameters. Samtools-1.3.1 (Li et al., 2009) was used for sam to bam format conversion. Peak regions were identified by homer-v4.8.3 (Heinz et al., 2010) function findPeaks with parameter ‘-style histone -F 2 -size 3000 -minDist 5000’. The overlapped peaks of replicate datasets were found using homer function mergePeaks using default parameter. Peak annotation was performed by homer function annotatePeaks with default parameter. The intersecting peak regions between different genotypes and aging stages were identified by BEDTools-2.26.0 (Quinlan and Hall, 2010) using default parameter.
For quantitative comparison, the dm6 mapped reads were normalized to a scale factor using deeptools-2.2.4 (Ramírez et al., 2014) function bamCoverage with 10 bp bin size. The scale factor for each sample was calculated as reported in Orlando et al., 2014a with a small modification by using the percentage of mapped spike-in mouse genome (genome build mm10) in input from which we derived the scale factor, a, as follows:
Let us define a as the normalized factor, b as the reference signal from the mouse cells, c as the total number of reads (in millions) from a sample aligned to mm10 genome, and d as the percentage of mapped spike-in mouse genome (genome build mm10) in input.
Based on our experimental design, the fraction of the spike-in mouse genome over the total fly genome is a constant. Mathematically, b can be expressed as follows:
The epigenome of the reference sample does not vary appreciably, such that the reference signal from mouse cell should be the same across all sample being examined. Taking equation (1) and the fact that b is constant across all the samples, it is easy to show that:
Without loss of generality, we can set the value of b to one and the scale factor a can be estimated by:
The ChIP intensity for the gene bodies (transcriptional start site to transcriptional termination sites as annotated in dm6) of all protein-coding genes or peak region was calculated using the bwtool (Pohl and Beato, 2014) function summary with default parameters. The input bigwig file, normalized by a scale factor, was generated by bamCoverage function. Bootstrapping was used for calculating the mean reference-adjusted ChIP intensity of peak genes or inter-peak genes (10000 draws with replacement of n = 500) and the confidence intervals. Violin plots represent the bootstrapped mean H3K27me3 level of inter-peak genes and peak genes at different ages. The scatter plot and violin plot was generated by R package ggplot2.
For visualization of the selected genome regions, we used IGV-2.3.31 (Robinson et al., 2011) with the bigwig files generated by bamCoverage. Circos plot was generated by J-circos-V1 (An et al., 2015) with the bigwig files. Cistrome correlation tools (Liu et al., 2011) were used to generate the heatmap with hierarchical clustering (hclust function in R) according to pair-wise correlation coefficients. Pair-wise Pearson correlations of genome-wide H3K27me3 levels were calculated at windows of 1Kbps on the genome scale using multiple bigwig file. The deeptools function bamCompare was used to generated bigwig file based on ChIP and input bam files that are compared to each other with ratio while being simultaneously normalized using RPKM.
Sequencing reads were mapped to the reference genome dm6 with STAR2.3.0e (Dobin et al., 2013) by default parameter. The read counts for each gene were calculated by HTSeq-0.5.4e (Anders et al., 2015) htseq-count with parameters ‘-m intersection-strict -s no’ with STAR generated SAM files. The count files were used as input to R package DESeq (Anders and Huber, 2010) for normalization and the differential expression genes were differential expression genes were computed based on normalized count from three biological replicates (p<0.05). The venn diagrams of overlapped differential genes were generated by VENNY-2.1 (http://bioinfogp.cnb.csic.es/tools/venny/index.html). GO term analysis was performed by David (Huang et al., 2009a, 2009b) with default parameters. WGCNA (1.51) was used for network analysis of co-regulated genes across PRC2 mutant samples with parameters “Soft thresholding power β = 8; Scale-free topology fit index R2 = 0.897). For selected module, Cytoscape (3.5.1) was used for visualizing interaction network.
The raw MS data (.wiff) were converted to mzML files using ProteoWizard MSConvert (version 3.0.6526) and processed using XCMS (Smith et al., 2006) for feature detection, retention time correction and alignment. Part of XCMS processing parameters were optimized and set as followings: mass accuracy in peak detection = 25 ppm; peak width c = (5, 20); snthresh = 6; bw = 10; minfrac = 0.5. R package CAMERA (Kuhl et al., 2012) was used for peak annotation after XCMS data processing. A customized R script was used to preprocess and extract the MS/MS data for database match with in-house library. The resulting raw data was then preprocessed for data normalization using SVR algorithm (Shen et al., 2016). Pathway enrichment analysis was executed using MetaboAnalyst 3.0 (Xia et al., 2015). In MetaboAnalyst 3.0, metabolic pathways for Drosophila melanogaster were selected. The pathway analysis algorithms were hypergeometric test for over representation analysis and relative-betweeness centrality for pathway topology analysis. For normalized intensity calculation, the preprocessed metabolomic data was further scaled (see below).
σ is the standard deviation of the population.
Specifically, in each sample group, the pathway intensity was determined by calculating the sum of detected metabolites' normalized intensities in each metabolic pathway.
The raw data files of sequencing experiments have been deposited in the NCBI Gene Expression Omnibus, as well as the normalized read density profiles of ChIP-seq and differential expression results from DESeq of RNA-seq reported in this paper. The accession number is GEO:GSE96654.
All statistical details of experiments are included in figure legends. Sample numbers and experiment replicates are stated in figure legends.
Manipulation of a nuclear NAD+ salvage pathway delays aging without altering steady-state NAD+ levelsJournal of Biological Chemistry 277:18881–18890.https://doi.org/10.1074/jbc.M111773200
Life span extension and neuronal cell protection by Drosophila nicotinamidaseJournal of Biological Chemistry 283:27810–27819.https://doi.org/10.1074/jbc.M804681200
Rethinking glycolysis: on the biochemical logic of metabolic pathwaysNature Chemical Biology 8:509–517.https://doi.org/10.1038/nchembio.971
Effect of chronic caloric restriction on hepatic enzymes of intermediary metabolism in the male Fischer 344 ratMechanisms of Ageing and Development 48:179–189.https://doi.org/10.1016/0047-6374(89)90049-3
Jupyter and galaxy: easing entry barriers into complex data analyses for biomedical researchersPLOS Computational Biology 13:e1005425.https://doi.org/10.1371/journal.pcbi.1005425
Extracellular redox state: refining the definition of oxidative stress in agingRejuvenation Research 9:169–181.https://doi.org/10.1089/rej.2006.9.169
Dynamics of polycomb proteins-mediated histone modifications during UV irradiation-induced DNA damageInsect Biochemistry and Molecular Biology 55:9–18.https://doi.org/10.1016/j.ibmb.2014.10.001
Chronic ingestion of 2-deoxy-D-glucose induces cardiac vacuolization and increases mortality in ratsToxicology and Applied Pharmacology 243:332–339.https://doi.org/10.1016/j.taap.2009.11.025
ATP from glycolysis is required for normal sodium homeostasis in resting fast-twitch rodent skeletal muscleAmerican Journal of Physiology-Endocrinology and Metabolism 281:E479–E488.https://doi.org/10.1152/ajpendo.2001.281.3.E479
Venny: An interactive tool for comparing lists with Venn's diagramsVenny: An interactive tool for comparing lists with Venn's diagrams, http://bioinfogp.cnb.csic.es/tools/venny/index.html.
Innovation: Metabolomics: the apogee of the omics trilogyNature Reviews Molecular Cell Biology 13:263–269.https://doi.org/10.1038/nrm3314
R: A language and environment for statistical computingR Foundation for Statistical Computing, Vienna, Austria.
deepTools: a flexible platform for exploring deep-sequencing dataNucleic Acids Research 42:W187–W191.https://doi.org/10.1093/nar/gku365
H3K36 methylation promotes longevity by enhancing transcriptional fidelityGenes & Development 29:1362–1376.https://doi.org/10.1101/gad.263707.115
Drosophila UTX is a histone H3 Lys27 demethylase that colocalizes with the elongating form of RNA polymerase IIMolecular and Cellular Biology 28:1041–1046.https://doi.org/10.1128/MCB.01504-07
Polycomb Group proteins: an evolutionary perspectiveTrends in Genetics 23:494–502.https://doi.org/10.1016/j.tig.2007.08.006
MetaboAnalyst 3.0--making metabolomics more meaningfulNucleic Acids Research 43:W251–W257.https://doi.org/10.1093/nar/gkv380
Matt KaeberleinReviewing Editor; University of Washington, United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article "Epigenetic Drift of H3K27me3 in Aging Links Glycolysis to Healthy Longevity" for consideration by eLife. Your article has been favorably evaluated by Kevin Struhl (Senior Editor) and three reviewers, one of whom, Matt Kaeberlein (Reviewer #1), is a member of our Board of Reviewing Editors. The following individual involved in review of your submission has agreed to reveal their identity: Adrienne Wang (Reviewer #3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
Overall, the referees appreciated the novelty and importance of the work and the provocative new hypotheses suggested. Several major concerns were noted, however, that should be addressed before publication. Many of these concerns involve more thorough integration of these studies within the existing literature and a discussion of how the results fit, or in some cases run counter to, prevailing models for aging. In particular, this study is limited in that it may be unique to Drosophila. While the reviewers found the data to be credible, the proposed model does contradict data in other organisms regarding the importance of glycolysis in aging (large body of CR literature, work from Ristow in worms and mice https://www.nature.com/articles/ncomms4563) and is directly contradicting the current thinking about NAD/NADH ratio and aging. The authors at least need to acknowledge these contradictions and the possibility that this is a private mechanism of aging in Drosophila that will not be conserved in other organisms. Unless they have data to suggest that this is a conserved mechanism of aging, in which case it is strongly recommended that they add it to the paper.
There is also a striking absence of discussion of prior literature related to epigenetic marks modulating aging and lifespan in yeast and worms. The studies from Shelley Berger's lab showing that histone acetylation impact aging in yeast should be discussed, as should the work from Anne Brunet and others in C. elegans and in stem cells, some of which specifically deals with H3K27me3. The authors need to put their work into context with the prior literature rather than ignoring it. To claim that this study shows that H3K27me3 is a "new mechanism that drives the progression of aging" would seem an overstatement in light of the prior literature in this area. As another example of missing literature, the authors completely ignore a study by the Rabinovitch group in which quantitative proteomics was used to characterize age-related changes in the heart proteome and in which glycolytic enzymes were shown to increase with age, while calorie restriction was shown to decrease the abundance of these enzymes, promoting health of this tissue (Dai et al., 2014).
The referees also felt that in many cases, the data were over-interpreted. For example, the subsection “Long-lived PRC2 Mutants Diminish the Epigenetic Drift of H3K27me3 During Aging”, ends with 'Combined, our study implicates that the life-extension effect of PRC2-mutation likely arises from its ability to mitigate the drifting of H3K27me3 during aging.' At this point in the paper, it has only been shown that PRC2 mutants live longer and that they have reduced H3K27me3 levels. A functional or causal connection between these phenotypes has not been established, and the conclusion is thus not valid. There are many other instances of such over-interpretation, and the authors need to edit the text significantly to more conservatively and accurately interpret the results.
In addition, there are a few experimental studies, additional statistical analyses, and additional textual changes that were considered to be essential for further consideration at eLife.
1) There is the lack of detail on the collection, processing, and statistical analysis of the large datasets. It is therefore difficult to evaluate how rigorous the analysis of these data was. Raw datasets (for example FPKM tables for the RNAseq results, as well as cohort data for the lifespan studies – for which statistical analysis is wholly missing) should be made available to evaluate the data. Further consultation with a statistician before resubmission is highly recommended.
2) A majority of the results are based on muscle samples, but the authors occasionally use head-specific samples throughout the manuscript. Though both tissues have high energy demand, their regulation of metabolism is very different. This is something that should be acknowledged in the manuscript. In particular, since epigenetic marks are very cell-type dependent, a clearer characterization and description of epigenetic changes specifically in muscle and brain tissue is warranted. Based on the Materials and methods section, the data was acquired from muscle, but later RNAseq studies also include the brain. Tissue-specificity needs to be clearly described and if transcriptome or metabolomic data from brain is used, then the ChIPseq data should be obtained from brain as well.
3) The metabolomics data raises important unresolved questions that could be addressed. The authors measured a total of 151 metabolites. This likely represents far less than the total metabolome found in Drosophila. The size of their dataset is critical in drawing inference about the key pathways that are associated with lifespan. The authors do look specifically for changes in the TCA cycle, but might there be other equally predictive pathways associated with metabolism that are not represented here given the small number of metabolites measured overall? What happens to TCA intermediates? (ATP levels are down, suggesting glycolysis is the main source of ATP, but this isn't actually measured. What happens to oxphos activity/efficiency? What is happening to the PPP pathway?
4) How does the global loss of PRC2 affect the developmental establishment of H3K27me3 patterns (i.e. do certain marks never get established in the first place? It is mentioned that 'global occupancy of H3K27me3 between wt and age-matched PRC2 mutants was highly preserved', but the correlation coefficient is only 0.77, suggest that there are significant changes in certain loci that are being ignored). A more acute depletion of PRC2 in the adult, for example by RNAi would be useful to analyze its effects specifically on aging-related pattern changes. It would also be important to acutely deplete PRC1 to assess the specificity of the PRC2 effects. And to perturb both PRC1 and PRC2 to establish whether the establishment and/or maintenance of epigenetic marks is affected by age. It is shown that PRC1 mutants have extended lifespan but no changes in H3K27me3, and this discrepancy with PRC2 mutants is ignored – if PRC1 mutants are also long-lived, it seems likely that the longevity effects are not caused by changes in H3K27me3.
5) As this paper is reliant upon accurate metabolic analysis, and the levels of protein and sugar in Drosophila food have documented effects on lifespan and metabolism, the authors should specify the "standard" food their flies are reared on.
6) Given that the authors are studying changes that occur with aging, age is an important factor for the interpretation of most of the figures. Throughout the manuscript, there are many places where age of samples used is not explicitly stated, or is unclear – for example, the samples used for the transcriptomic analysis.
7) The authors show that down-regulation of these PRC2-mediated glycolytic genes blocks the lifespan extension seen in PRC2 mutants, thus establishing that the mechanism of lifespan extension by PRC2 is at least partially reliant upon glycolysis. However, the extent to which downregulation of these genes blocks the PRC2-mediated lifespan extension is unclear from the text and without referring to supplementary figures. As there was only partial reduction of lifespan by knocking down glycolytic genes in comparison to WT, it indicates that other mechanisms might be in play mediating lifespan in this model. This should be explicitly addressed.
8) The authors should rearrange the figures such that the actual data are shown in Figure 2 and the summary table is moved to supplemental.
9) The observations related to NADH/NAD ratio run counter to a growing body of literature suggesting that NADH/NAD ratio increases during aging and that restoring this imbalance (for example through NAD precursors) can increase lifespan and/or healthspan. The authors should at least comment on this apparent discrepancy and offer some speculation about why they are getting an opposite result.
10) The statement "drives the progression of aging" is perhaps overly reaching. It would be better to say that the data presented in this study are consistent with the model that epigenetic drift of H3K27me3 is one of the molecular mechanisms that contributes to at least a subset of age-associated phenotypes and limits lifespan in fruit flies.
11) Typo in the subsection “A CRISPR/Cas9 Deficiency Screen Identifies the Role of PRCs in Aging”: Should read "Given that the above study indicated…"
12) Figure 1C and D: it is unclear why red highlighted area extends below diagonal line, if the authors claim that signals in aged animals are higher than in younger? (same goes for some of the other scatter plots…)
13) Figure 4C: what is the age of analyzed animals?
14) Locomotion: usually longer distances than 2cm are recorded (e.g. PMID:25850672 uses 8cm in 20 seconds). Any specific reason for short distance?https://doi.org/10.7554/eLife.35368.029
- Feng Liu
- Zhandong Liu
- Zhandong Liu
- Zhandong Liu
- Zhandong Liu
- Zhandong Liu
- Yaoyang Zhang
- Yaoyang Zhang
- Yaoyang Zhang
- Yaoyang Zhang
- Nan Liu
- Yaoyang Zhang
- Zheng-Jiang Zhu
- Nan Liu
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank Drs. Jilong Liu, Yun Zhao, Meng-Qiu Dong, Cong Liu, Chao Tong, Xing Guo, Lanfeng Wang, and Ye Tian for advice on the manuscript; Wei Wu and the Core Facility of Drosophila Resource and Technology, Shanghai Institute of Biological Sciences, Chinese Academy of Sciences, Shanghai, China for fly microinjections. Prof. Junying Yuan provided considerable support, advice on the experiments, and critical suggestions on the manuscript. NL and ZZ are Junior Scholar of the 1000 Plan of China. YZ is supported by 100 Talents Program of the Chinese Academy of Sciences. This work was supported by grants from the National Program on Key Basic Research Project of China to NL and YZ (2016YFA0501900); the National Natural Science Foundation of China to NL (31371326), YZ (31671428, 31500665, 31530041), ZZ (21575151), and FL (81770143); a National Institute of Health grant (GM120033) to ZL; a National Science Foundation grant (DMS-1263932) to ZL; a Cancer Prevention Research Institute of Texas grant (RP170387) to ZL; the Robert A and Renée E Belfer Family Foundation to ZL, and the Chao Family Foundation to ZL.
- Matt Kaeberlein, Reviewing Editor, University of Washington, United States
© 2018, Ma et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.