SEARCH

SEARCH BY CITATION

Keywords:

  • branched-chain α-keto acids;
  • clk-1 ;
  • isp-1 ;
  • mev-1 ;
  • nuo-6 ;
  • tpk-1 ;
  • ucr-2.3

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

Mit mutations that disrupt function of the mitochondrial electron transport chain can, inexplicably, prolong Caenorhabditis elegans lifespan. In this study we use a metabolomics approach to identify an ensemble of mitochondrial-derived α-ketoacids and α-hydroxyacids that are produced by long-lived Mit mutants but not by other long-lived mutants or by short-lived mitochondrial mutants. We show that accumulation of these compounds is dependent on concerted inhibition of three α-ketoacid dehydrogenases that share dihydrolipoamide dehydrogenase (DLD) as a common subunit, a protein previously linked in humans with increased risk of Alzheimer's disease. When the expression of DLD in wild-type animals was reduced using RNA interference we observed an unprecedented effect on lifespan – as RNAi dosage was increased lifespan was significantly shortened, but, at higher doses, it was significantly lengthened, suggesting that DLD plays a unique role in modulating length of life. Our findings provide novel insight into the origin of the Mit phenotype.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

The Mit mutants of Caenorhabditis elegans have impaired mitochondrial electron transport chain (ETC) activity, yet are long lived (Rea & Johnson, 2003). Several proteins have been identified that modulate the longevity response in some or all of the Mit mutants, including HIF-1, p53, CEH-23, CRTC-1, CREB, and AMP kinase (Apfeld et al., 2004; Ventura et al., 2009; Lee et al., 2010; Mair et al., 2011; Walter et al., 2011). In addition, reactive oxygen species (ROS), the mitochondrial unfolded protein response, and autophagy have also been shown to modulate the severity of the Mit phenotype when their level of production or operation is altered (Haynes et al., 2007; Lee et al., 2010; Yang & Hekimi, 2010; Nargund et al., 2012). Despite all of these findings, the underlying events that establish the Mit phenotype are not known. Moreover, why Mit mutations extend lifespan in worms, whereas other ETC mutations that also disrupt the ETC chain shorten lifespan, remains unknown.

Recent studies using tissue-specific RNAi knockdown in C. elegans have suggested that restricting mitochondrial dysfunction to neuronal or intestinal cells may be sufficient to extend the lifespan of otherwise wild-type animals (Durieux et al., 2011). These studies have led to the suggestion that a mitokine may emanate from dysfunctional mitochondria to systemically reprogram unaffected cells and increase animal lifespan. The nature of such a signal remains unknown. In this study we identify a set of small metabolites that are uniquely generated by Mit mutants. We identify their enzymatic source and discover the unifying principle behind the Mit phenotype.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

GC-MS footprinting defines a Mit mutant-specific metabolic signature

We reasoned that dysfunctional mitochondria may accumulate novel metabolites, or accumulate normal metabolites to abnormal levels, which in turn may serve as systemic signaling molecules. Such molecules might also be expected to become enriched in the external environment. Synchronous populations of the long-lived strain isp-1(qm150), the short-lived strain mev-1(kn1), and wild-type animals were generated as previously described (Butler et al., 2010), and metabolite excretion was monitored over a 20-h period using GC-MS footprinting (Butler et al., 2012). All three strains generated ∼100 compounds that we could reliably detect, allowing us to identify compounds whose production over time differed significantly between strains (Fig. 1A, Supplementary Figure 1). Metabolites which covaried over time within each strain were also identified (Supplementary Figures 2–4). A unique feature of isp-1 Mit mutants was significantly increased production of branched-chain α-ketoacids and their corresponding α-hydroxyacid reduction products (Fig. 1A, red asterisks). Increased production of these compounds was also observed in two other long-lived Mit mutants, clk-1(qm30) (Wong et al., 1995) and nuo-6(qm200) (Yang & Hekimi, 2010), at levels that distinguished them significantly from wild-type animals (P < 0.005 and P < 0.0042, respectively) (Fig. 1B, Supplementary Figures 5–7). In contrast, short-lived mev-1(kn1) and ucr-2.3(pk732) mutants generated their own unique metabolic signature that was enriched in amino acids and tricarboxylic acid (TCA) cycle intermediates (Fig. 1C); a profile that may reflect enhanced flux into the TCA cycle. We next examined whether other long-lived (Age) mutants also generated products similar to those made by Mit mutants. Production of branched-chain α-ketoacids and their corresponding reduction products appears unique to Mit mutants as they were not elevated in the exometabolomes of four other long-lived mutants: daf-2(e1370), clk-2(qm37), eat-2(ad465), and slcf-1(tm2258) (Benard et al., 2001; Mouchiroud et al., 2011) (Fig. 1B and Supplementary Figures 5–7). These findings are consistent with genetic evidence suggesting daf-2, eat-2 and clk-2 mutants extend lifespan in a manner distinct from, or at best partially overlapping with, that of Mit mutants (Lakowski & Hekimi, 1996, 1998).

image

Figure 1. A metabolic signature for long life in the Caenorhabditis elegans Mit mutants. (A) Exometabolome analysis of wild-type worms (N2), isp-1(qm150) and mev-1(kn1) mutants: 120 000 worms were transferred to minimal media and their exometabolome sampled over a 20-h period (at +0, 0.5, 1, 2, 5, 20 h). GC-MS was used to identify and quantify metabolites within each sample. Data are presented using hierarchical clustering (Pearson's correlation coefficient): left panel, metabolite variation across row; right panel, metabolite variation relative to the entire array. Asterisks mark several α-ketoacids and α-hydroxyacids that are significantly overproduced by isp-1 mutants (statistical analyses summarized in Supplementary Figure 1). (B, C) Exometabolome analysis of various Mit, Age, and short-lived mutants. Data were collected as described in (A) following an 18-hr metabolite capture period. Results are presented as in the left panel of A. In (C) only metabolites that differed significantly (P < 0.0062) between long-lived and short-lived ETC mutants are shown (full results are provided in Supplementary Figures 6 and 7). Strains are marked at the top of each panel. (D) Many of the compounds detected in the exometabolome of worms strains are related by redox reactions. Metabolites that are specifically enriched in the exometabolome of Mit mutants are highlighted. (E) Quantification of select α-ketoacids and α-hydroxyacids in the +18-h exometabolome of wild-type (N2), isp-1(qm150), and nuo-6(qm200) animals. Ordinate is plotted on a log2 scale.

Download figure to PowerPoint

The Mit mutant metabolic profile is distinct from that of anaerobic worms

Several observations have led to the proposition that Mit mutants ectopically activate metabolic pathways normally reserved for survival under low or no oxygen. These observations include enhanced tolerance of Mit mutants to acute anoxia (Butler et al., 2010), and the fact that loss of hypoxia-inducible factor-1 (HIF-1) in isp-1(qm150) and clk-1(qm30) Mit mutants mitigates their increased longevity (Lee et al., 2010). We found that Mit mutants do not generate metabolic end products characteristic of wild-type worms exposed to anoxic conditions (Supplementary Figures 8 and 9). In particular, excretion of volatile fatty acids, signature molecules of anaerobic worms (Butler et al., 2012), is not observed in Mit mutants grown under normoxic conditions (Supplementary Figure 8A). All Mit mutants nonetheless remain capable of robust volatile fatty acid production when placed under anoxia (Supplementary Figure 8A). Collectively, these data indicate that the mechanisms functioning in Mit mutants to counteract their ETC deficit and extend their health- and lifespan are distinct from canonical C. elegans anaerobic survival processes, and imply use of a potentially novel metabolism.

Dihydrolipoamide dehydrogenase (DLD) inhibition phenocopies the metabolic profile of Mit mutants

Many of the compounds that differentially accumulate in the exometabolome of Mit mutants are related by redox chemistry (Fig. 1D). A second feature that unites many of these compounds is the connection of their parent α-ketoacid (Fig. 1D) with the enzyme dihydrolipoamide dehydrogenase. DLD (or E3) is a shared subunit of three evolutionarily related enzyme supercomplexes – branched-chain α-keto acid dehydrogenase (BCKADH), pyruvate dehydrogenase (PDH), and α-ketoglutarate dehydrogenase (α-KGDH) (Kochi et al., 1986; Matuda et al., 1991). All three enzymes are central components of intermediary metabolism and each acts to oxidatively decarboxylate one or more α-ketoacids. The striking metabolic profile of Mit mutants suggested to us that DLD may be inhibited in these animals. When a feeding RNAi was used to incrementally reduce DLD expression (Bhaskaran et al., 2011) in wild-type worms (Fig. 2A–C), we observed that the excreted metabolic profile that followed the most potent knockdown of DLD (~85%) correlated most closely with that of nuo-6(qm200) Mit mutants (rAv = 0.87, = 7) (Figs 2D, 3B and Supplementary Figures 10–16).

image

Figure 2. Inhibition of dihydrolipoamide dehydrogenase activity leads to metabolic and phenotypic recapitulation of the Mit phenotype. (A, B) Cross-reactivity of α-DLD polyclonal antibody with human (HEK293T) and mouse (3T3-L1 fibroblast) whole-cell extracts, and with whole-worm extracts of N2, isp-1(qm150), and mev-1(kn1). Also tested in (B) were whole-worm extracts from a feeding RNAi dilution series (Rea et al., 2007) targeting DLD (RNAi to empty vector ratio – 0:1, 1:200, 1:20, 1:5, and 1:0). Animals were fed DLD RNAi from the time of hatching. All lanes contain 25 μg of protein; β-actin served as loading control in (B). (C) DLD activity in whole-worm extracts from two independently collected DLD feeding RNAi dilution series. One unit of DLD activity is defined as the rate of production of 1 μmol of NADH in 1 min at 25°C (Bhaskaran et al., 2011) (error bars: ± SD). (D) Temporal changes in the exometabolome of N2 worms following treatment with increasing amounts of DLD RNAi (RNAi to empty vector ratio – 0:1, 1:200, 1:20, 1:5, and 1:0). Excreted metabolites were collected at +0, 0.5, 2, 5, and 18 h. Data were analyzed by GC-MS and are presented using hierarchical clustering – left panel, metabolite variation across row; right panel, metabolite variation relative to the entire array. (See also Supplementary Figures 10–15). (E, F) Increasing doses of RNAi targeting DLD (RNAi to empty vector ratio – 0:1, 1:1000, 1:500, 1:200, 1:100, 1:50, 1:20, 1:10, 1:5, 1:2, and 1:0) were fed to wild-type Caenorhabditis elegans from the time of hatching and the effects on both adult size (E, Scale bar: 200 μm) and lifespan (F) measured. Lifespan data are the mean of four replicates (±) SEM. (n = 60 worms/condition/replicate; asterisks indicates significantly different from vector, P < 0.005 (Bonferroni corrected), summary statistics are tabulated in Supplementary Figure 17). (G) DLD activity in whole-worm extracts from N2 and isp-1(qm150) animals (n = 3 replicates, error bars: ± SD).

Download figure to PowerPoint

image

Figure 3. Exometabolome analysis of tpk-1(qm162) mutants reveals concerted inhibition of α-ketoacid dehydrogenase activity is sufficient to recapitulate the Mit metabolic phenotype. (A) The +18-h exometabolome of the following strains was collected and analyzed by GC-MS: wild-type worms (N2), isp-1(qm150), and nuo-6(qm200) Mit mutants, short-lived mev-1(kn1) mutants, N2 exposed to DLD RNAi from the time of hatching (RNAi to vector ratios of 0:1, 1:200, 1:20, 1:5, and 1:0), and tpk-1(qm162) mutants exposed for 18 h to both normoxia or anoxia. Columns represent independent experimental replicates. Data are presented using hierarchical clustering – left panel, metabolite variation across row; right panel, metabolite variation relative to the entire array. Statistical analyses are summarized in Supplementary Figures 5. See also Supplementary Figure 16 for global correlation analysis. (B) Correlation matrix showing metabolic similarity between +18-h exometabolome of long-lived Mit mutants, long-lived dld-1 disrupted animals, and long-lived tpk-1(qm162) mutants. Details of distance measure calculations are described in Supplementary Figure 16. (C) Model for the genesis of the Mit phenotype (BCAA, branched-chain amino acids; Mito; ETC, mitochondrial electron transport chain).

Download figure to PowerPoint

Graded inhibition of DLD causes a novel effect on lifespan

We have previously shown that RNAi directed against Mit genes in wild-type worms reproduces many aspects of the Mit phenotype (Rea et al., 2007). We showed that as RNAi dosage was increased, postembryonic development slowed, adult size became smaller, and adult lifespan was extended. At a critical dose of RNAi these life-enhancing properties either plateaued or, for some target genes such as atp-3, lifespan reached a pathological turning point beyond which lifespan began to shorten. We tested whether incrementally reducing DLD activity by RNAi also reproduced other aspects of the Mit phenotype. RNAi-mediated inhibition of DLD in wild-type worms resulted in slowed development (not quantified) and decreased final adult size (Fig. 2E). Unexpectedly, we observed that intermediate doses of DLD RNAi significantly (P < 0.005) shortened adult lifespan, whereas more potent doses of DLD RNAi significantly (P < 0.005) extended adult lifespan (Fig. 2F, Supplementary Figure 17). This biphasic effect of gene dosage on longevity has, to the best of our knowledge, never before been reported for any gene and supports our hypothesis that DLD plays a central role in the regulation of aging in Mit mutants.

DLD has a fourth function in cells

DLD not only functions to channel electrons to NAD+ in the final step of the three α-ketoacid dehydrogenase supercomplexes, it also performs a similar role in the glycine cleavage system (GCS) in which 5,10-methylene-tetrahydrofolate is produced from glycine (Kikuchi et al., 2008). Unlike for the α-keto acid dehydrogenases, the ensemble of reactions that comprise the GCS is fully reversible (Kikuchi et al., 2008). We noted for wild-type worms treated with increasing amounts of RNAi targeting DLD that their exometabolome accumulated increasing amounts of glycine (Fig. 2D). Curiously, for clk-1(qm30) and isp-1(qm150) Mit mutants, glycine was not a compound that accumulated significantly (Fig. 1A, Supplementary Figures 6, 7). For nuo-6(qm200) mutants, glycine was elevated significantly (Supplementary Figure 7), but this level paled when compared with that in DLD RNAi-treated animals (Fig. 3A). One interpretation of this observation is that DLD activity is not directly inhibited in Mit mutants, leading to the alternate hypothesis that α-keto acid dehydrogenase supercomplexes are inactivated via DLD-independent mechanisms, possibly at the level of their E1 or E2 subunits. Consistent with this notion, when we assayed DLD enzymatic function in isp-1(qm150) Mit mutants, after disrupting the α-keto acid supercomplexes and freeing DLD from bound E1 and E2, we observed no decrease in its activity relative to that of wild-type animals (Fig. 2G).

Disruption of a positive regulator of α-keto acid dehydrogenases mimics the Mit mutant phenotype

E1 proteins catalyze the irreversible decarboxylation of their α-ketoacid substrate and they are the primary site of α-ketoacid dehydrogenase regulation in mammalian cells. We tested if either individual or concerted inhibition of these enzymes in wild-type worms could phenocopy the longevity of Mit mutants. When E1 proteins were separately targeted for disruption by RNAi, no increase in lifespan was detected (Supplementary Figure 18 A–C). We tested a range of concentrations for each RNAi and observed that, in all cases, lifespan remained either unchanged relative to vector control, or was significantly shortened (Supplementary Figure 18B). Of the three series, RNAi targeting E1 of α-KGDH had some semblance to DLD RNAi, but no condition extended lifespan significantly. To disrupt all three E1 proteins simultaneously we chose a genetic approach. Thiamine phosphate is an essential cofactor of E1 enzymes (Harris et al., 1997), and thiamine phosphorylase is required to phosphorylate and retain thiamine inside cells. tpk-1(qm162) mutants contain a hypomorphic disruption in thiamine phosphorylase (de Jong et al., 2004). Remarkably, when we analyzed the exometabolome of these mutants (Fig. 3A,B; Supplementary Figure 16), we observed a marked similarity with the exometabolome of nuo-6(qm200) Mit mutants (correlation coefficient rAv = 0.81, n = 7). The tpk-1(qm162) profile was also very closely related to the exometabolome profile of long-lived worms treated with undiluted DLD RNAi (rAv = 0.96, = 3). isp-1(qm150) Mit mutants were also clearly related to tpk-1(qm162) mutants by this metric because their correlation coefficient (rAv = 0.59, = 3) was still double the next nearest strain's. Across all four strains, we saw marked enrichment of a common set of α-ketoacids and α-hydroxyacids (Fig. 1D,E). Remarkably, tpk-1(qm162) animals are known to exhibit many Mit-like traits including slowed development, reduced adult size, anoxia tolerance, and extended lifespan (de Jong et al., 2004; Butler et al., 2010). These data suggest that concerted inhibition of the α-ketoacid dehydrogenases is necessary and sufficient to establish the Mit phenotype.

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

In this study we identified an ensemble of α-ketoacids and α-hydroxyacids that are differentially overproduced by Mit mutants. On the basis of this unique metabolite profile we predicted and showed that DLD, within the context of the α-ketoacid dehydrogenases, was a pivotal control point in the production of these compounds. When we disrupted DLD in wild-type worms using RNAi, we made the surprising finding that lifespan was significantly shortened at low RNAi doses, but at high RNAi doses it was significantly lengthened. We subsequently discovered that simultaneous disruption of the α-ketoacid dehydrogenases at the level of E1 was necessary and sufficient to recapitulate the Mit phenotype.

Our present findings provide new insight into the genesis of the Mit phenotype. In vitro studies show that electrons can be forced to flow backwards into DLD when matrix NADH levels are raised (Starkov et al., 2004). In fact, under these conditions, electrons can flow through DLD into E2 to reduce its bound lipoamide cofactor. The reaction catalyzed by E1 is irreversible and so electrons that reside at the level of E2, as dihydrolipoamide, become prone to massive oxidation. When this occurs superoxide is formed, leaving an E2-bound thiyl radical (Starkov et al., 2004) that is capable of inactivating the α-keto acid dehydrogenase complex (Bunik & Fernie, 2009). Indeed, the superoxide that forms by this process can be on a scale comparable with that produced by the mitochondrial ETC (Starkov et al., 2004). Within this framework a rationale for why some mitochondrial ETC mutations in C. elegans increase lifespan while others cause premature death can be proposed: Because both short-lived and long-lived worms containing mitochondrial mutations have elevated ROS levels (Senoo-Matsuda et al., 2001; Lee et al., 2010), we believe ROS formation is not the discriminating factor. Instead, we posit that ETC mutations which increase lifespan are simply ones that allow NADH levels to be raised enough so that electrons can flow backwards into DLD and inhibit the α-keto acid dehydrogenases (Fig. 3C). For short-lived ETC mutants, we suggest that their specific mutations either uncouple NADH consumption from ATP production – permitting electrons to leak from the mitochondrial electron transport chain and preventing NADH levels from ever becoming elevated enough to inhibit DLD – or that sufficient amounts of complex I activity remain in these mutants to keep NADH relatively low. Consistent with both ideas, ectopic production of ROS has been recorded in mev-1 mutants at the level of complex II (Senoo-Matsuda et al., 2001); whereas supercomplexes containing complex I were shown to be specifically decreased in isp-1(qm150) mutants, but the spontaneously derived isp-1 suppressor mutation ctb-1(qm189) functioned to recover these levels (Suthammarak et al., 2010). The Mit phenotype is specified before adulthood (Dillin et al., 2002; Rea et al., 2007). This period of development is a time when even wild-type worms are likely to compete with bacteria for local oxygen (Peters et al., 1987). We presume matrix NADH levels become critically elevated in Mit mutants during this period.

It is becoming increasingly evident that regular compounds of intermediary metabolism can moonlight as signaling factors to affect cell fate. For example, accrual of succinate following mutation of succinate dehydrogenase (Cervera et al., 2009), or of (R)-2-hydroxyglutarate following disruption of isocitrate dehydrogenase (Xu et al., 2011), both result in oncogenesis. These so-called oncometabolites are thought to act by disrupting hypoxia-inducible factor 1 (HIF-1) and/or the epigenetic landscape of cells; specifically by altering the activity of EGL-9/PHD1-3 prolyl hydroxylases (Koivunen et al., 2012), jmjC-type histone demethylases, and/or TET family 5-methylcytosine (5mC) hydroxylases (McCarthy, 2012). New studies have uncovered a more nuanced level of control by 2-hydroxyglutarate: (R)-2-HG activates EGL-9/PHD proteins by functioning as a rogue substrate; (S)-2-HG instead blocks these proteins by acting as a competitive inhibitor (Koivunen et al., 2012). One intriguing possibility is that the α-ketoacids and α-hydroxyacids that accumulate in Mit mutants may act in a similar signaling manner. To this end we note that at least one of the accumulated α-ketoacids, pyruvate, can extend lifespan when supplemented to the diet of wild-type worms (Mouchiroud et al., 2011). Moreover, several of the compounds that accumulate in Mit mutants are already known to inhibit the same enzymes targeted by the abovementioned oncometabolites (Fig. 1D), including EGL-9 which regulates HIF-1 and is required for Mit mutant life extension (Lu et al., 2005; Lee et al., 2010). Interestingly, the most potent dose of DLD RNAi that extended lifespan in wild-type worms was also the dose associated with the greatest amounts of α-ketoacid and α-hydroxyacid accumulation. Future studies will be aimed at exploring further this hypothesis.

We observed an unprecedented effect of DLD RNAi on nematode lifespan. At this point we can only speculate why lifespan was decreased at intermediate DLD RNAi doses, but was lengthened by more potent RNAi doses. Excluding off-target RNAi effects, one idea pertains to the fact that α-ketoacid dehydrogenases are multimeric assemblages of E1, E2, and DLD subunits (Zhou et al., 2001). Perhaps DLD protein levels have to reach a critically low concentration before complexes disassemble fully. If so, electrons fed from the direction of E1 may inadvertently turn these semistable complexes into ROS generators as DLD levels become progressively rate limiting. With full complex disassembly, ROS production would not be possible, allowing any potential prolongevity signal (α-ketoacids and α-hydroxyacids) to dominate the signaling landscape.

Four noncoding SNPs at the DLD locus of humans have been associated with increased risk of late onset Alzheimer's disease in both Caucasian males and in an Ashkenazi Jewish population (Brown et al., 2004). Whether these SNPs are associated with reduced DLD activity remains unknown, but the biphasic effects of dld-1 RNAi on lifespan in C. elegans, coupled with its impact on exometabolite composition, suggest that mild DLD dysfunction in humans might alter the penetrance of age-related disorders such as Alzheimer's disease, or perhaps even drive mutagenic events that result in cancer.

Experimental procedures

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

Caenorhabditis elegans maintenance

The following C. elegans strains were used for this study: N2 Bristol, CB1370 [daf-2(e1370)III], DA465 [eat-2(ad465)II)], MQ125 [clk-2(qm37)III], MQ130 [clk-1(qm30)III], MQ770 [tpk-1(qm162)III], MQ887 [isp-1(qm150)IV], MQ1333 [nuo-6(qm200)I], NL1832 [ucr-2.3(pk732)III], SLR0032 [mev-1(kn1)III], SP506 [rad-5(mn159)III], TJ564Bx1[isp-1(qm150)IV;(gst-4::gfp)III], TJ5032 [clk-1(qm30)III;(gst-4::gfp)III], TK22 [mev-1(kn1)III], and TM2285 [slcf-1(tm2285)X]. All strains were maintained at 20°C, on NGM agar plates containing lawns of Escherichia coli (OP50), using standard worm culture techniques (Wood, 1988).

Feeding RNAi

All feeding RNAi constructs were obtained from the Ahringer RNAi library (Kamath & Ahringer, 2003). Targeted genes included: DLD, BCKADH E1α, PDH E1 α, and α-KGDH E1 corresponding to RNAi clones JA:LLC3.1, JA:Y39E4A.3, JA:T05H10.6, and JA:T22B11.5, respectively. All constructs were sequence confirmed. Feeding RNAi, RNAi dilution testing, and lifespan analyses were performed and analyzed exactly as previously described (Rea et al., 2007).

Exometabolome collection

We have described elsewhere detailed methods for exometabolome collection from C. elegans cultured under aerobic or anaerobic conditions (Butler et al., 2010, 2012). In the present studies we utilized 120 000 one-day-old gravid adult worms per replicate. Data for all strains were collected from multiple independent experimental replicates. Briefly, 120 000 arrested L1 larvae were cultured on 12 × 10-cm BNGM agar plates spread with E. coli (OP50). BNGM agar plates consisted of 1% w/v peptone, 2% w/v agar, 50 mm NaCl, 1 mm CaCl2, 1 mm MgSO4, 25 mm phosphate, and 5 μg/mL cholesterol. When animals became gravid adults they were collected in S-Basal (100 mm NaCl, 50 mm KH2PO4, pH 6.8), washed extensively in the same buffer (6 × 50 mL), stripped of residual bacteria by sucrose flotation (Foll et al., 1999), then again washed extensively in S-Basal (3 × 15 mL). Worm pellets were resuspended in S-Basal to a final concentration of 1 mg/mL. 1.2 mL of worm slurry was then transferred into a 3 cm glass dish and rotated (100 rpm) for the relevant length of time after which the supernatant containing the worm exometabolome was filtered [0.2 μm, Life Science Products (Denver, CO, USA), Cat &hash;6502-413X] and retained at −80°C until further use.

Multiple strains containing the same mutation were tested when possible to discount background effects. Strains used to generate GC-MS datasets were as follows: Fig. 1A: N2 Bristol, MQ887, SLR0032; Fig. 1B: N2 Bristol, TJ564Bx1, MQ1333, TM2285, TK22, CB1370, DA465; Fig. 1C & Supplementary Figure 6: N2 Bristol, NL1832, SP506, TJ5032, TJ564Bx1, TK22; Fig. 2D: N2 Bristol; Fig. 3A: N2 Bristol, MQ770, MQ887, MQ1333, SLR0032; Supplementary Figure 8: N2 Bristol, TJ564Bx1, MQ130, MQ1333, SLR0032, NL1832.

GC-MS acquisition

Aliquots of relevant worm supernatants (100 μL) were evaporated overnight using vacuum centrifugation. Excreted worm metabolites were then derivitized using methoxylamine HCl (Sigma Aldrich, Cat. No. 226904) and N-methyl-N-(tert-butyldimethylsilyl)trifluoroacetamide containing 1% tert-butyldimethylchlorosilane [Thermo Scientific (Waltham, MA, USA), Cat. No. TS-48927].

GC-MS analyses were undertaken using either an ion trap (University of Colorado, Boulder, CO, USA) or a quadrupole (University of Texas Health Science Center, San Antonio, TX, USA) mass spectrometer, with an electron impact ionization source. GC separation was performed using a ZB-5MS column (5%-phenyl-arylene–95%-dimethylpolysiloxane), 30 × 0.25 mm, 0.25 μm film thickness (Phenomenex; Torrance, CA, USA). MS analyses performed at the UT Health Science Center were conducted on a TRACE DSQ single quadrupole mass spectrometer (Thermo Fisher, San Jose, CA, USA). GC conditions for standard exometabolome samples (UT) were as follows: carrier gas, helium; linear velocity, 1 mL/min (constant flow); injection, split, 10 mL/min split flow; injector temperature, 220°C; column temperature program, initial temperature of 70°C held for 1 min followed by an increase to 310°C at 5°C/min. MS conditions were as follows: ionization, electron impact (70 eV); detection, positive ion; full scan analyses, m/z 50 – m/z 700 at two scans/s. Volatile metabolites eluted with the solvent front using this method, so GC separation of these analytes started with an initial temperature of 50°C held for 1 min, followed by an increase to 80°C at 10°C/min. The temperature was maintained for 3 min at 80°C after which it was increased to 275°C at a rate of 30°C/min.

Samples acquired at University of Colorado were analyzed on a Finnigan Polaris Q ion trap (Thermo Fisher). Injections onto the GC were performed manually using a hot needle injection technique with 1 μL of sample sandwiched between two cushions of air in the syringe body. The inlet temperature was 230°C with split ratio of 1:10. Chromatographic conditions were as follows: 1 mL/min constant helium flow with an initial oven temperature of 70°C held for 5 min followed by a ramp to 310°C at 5°C min−1 with a final hold of 1 min. The transfer line was kept at 250°C and the source was operated at 200°C and 70 eV. Masses were scanned from 50 to 650 m/z at ~2 scans/s. Data acquisition was performed using Xcalibur (Thermo Fisher).

MS data analysis

Peak integration

Following data acquisition, gas chromatograms were deconvoluted using AMDIS (Stein, 1999), then extracted ion chromatograms were integrated using MET-IDEA (Broeckling et al., 2006). Peak data were sequentially normalized to total worm protein content, and then to the peak area of an exogenously added internal standard (3,4-dimethoxybenzoate). Protein analyses were performed using the bicinchoninic acid-based protein assay (Pierce, Rockford, IL, USA).

Cluster analysis

Clustering was performed using the Hierarchical Clustering module of the GenePattern software suite (http://www.broad.mit.edu/cancer/software/genepattern/) (Reich et al., 2006). Metabolites were clustered using the pairwise complete-linkage method. Distance measures were calculated using Pearson's correlation coefficient. Low abundance and noisy peaks were removed from the analysis prior to clustering. For Fig. 1C & Supplementary Figure 6, clustering was performed using the rank orders of metabolite intensities.

Self-organizing map (SOM)

Time-course data were analyzed using the SOM algorithm of Mayday (http://microarray-analysis.org/) (Dietzsch et al., 2006). Prior to analysis, data were Z-score transformed (mean centered and scaled by standard deviation). Mayday settings were as follows: cycles, 250; Kernel function, Gaussian; Initial kernel radius, 2.0; Final kernel radius, 0.1; initializer of the SOM-units, random data point; distance measure, Euclidean. The grid topology and number of clusters were altered heuristically until a satisfactory result was obtained. Cluster quality was assessed by silhouette plots.

Significance testing

To identify metabolites that differed significantly between strains (or groups of strains), normalized GC-MS peak areas were generally rank ordered across all relevant cases and then analyzed using a General Linear Model (GLM). No effort was made to control for variation in experimenter or sample collection date, but data collected on different MS instruments were segregated. Specific details of all comparisons are provided in the legends of relevant Supplementary Figure files. Contrast coding was performed using SPSS 17.0 (IBM). In all instances significance thresholds were Bonferroni corrected to adjust for the number of contrasts interrogated.

Quantification of metabolites

Absolute levels of pyruvate, 2-ketobutyrate, 2-ketoisocaproate, lactate, 2-hydroxybutyrate, 3-hydroxypropionoate, 3-hydroxybutyrate, and 2-hydroxycaproate, present in the exometabolome of N2, MQ887 [isp-1(qm150)], and MQ1333 [nuo-6(qm200)] animals, were determined by comparison with standard curves. Solutions of metabolites of interest were prepared at a range of concentrations (1, 10, 50, 100, and 250 μm) and analyzed by GC-MS. Data presented in Fig. 2G represent averages from multiple independent test samples – N2 (n = 4), MQ1333 (n = 3), and MQ887 (n = 7).

Western blotting

Protein lysates and western analyses were collected and performed, respectively, exactly as described previously (Bhaskaran et al., 2011). DLD was analyzed using a rabbit polyclonal antibody raised against native pig heart DLD [1:5000, Abnova (Taipei City, Taiwan), Cat &hash; PAB10259]. β-actin was used as a loading control and was analyzed using a mouse monoclonal antibody [1:2000, Sigma (St. Louis, MO, USA), Cat &hash; A5441].

DLD activity assay

DLD activity in whole-worm extracts was determined spectrophotometrically, exactly as we have previously described (Bhaskaran et al., 2011).

Acknowledgments

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

We thank Drs. Shawn Ahmed (UNC, Chapel Hill, NC, USA), Benjamin Eaton, Kathleen Fischer, Milena Girotti, and Gregory Macleod (UTHSCSA, San Antonio, TX, USA) for critical reading of the manuscript. Mass spectrometry analyses were conducted in the Institutional Mass Spectrometry Laboratory at UTHSCSA. Technical support was provided by Yvonne Penrod, Adwitiya Kar, Melissa Little, and Meghan Cain. Financial support was provided by the National Institute on Aging (RJM & SLR) and the Ellison Medical Foundation (SLR).

Author contributions

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

JAB, RJM, and SLR designed the experiments, collected the data, analyzed the data, and cowrote the manuscript. SB was also involved in data collection.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Results
  5. Discussion
  6. Experimental procedures
  7. Acknowledgments
  8. Author contributions
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
acel12029-sup-0001-FigS1-S18.pdfapplication/PDF30703K

Fig. S1. Analysis of metabolite differences between wild-type (N2), isp-1(qm150), and mev-1(kn1) worms. The exometabolome data used to generate Fig. 1A were analyzed using a General Linear Model (GLM) to capture the major changes in concentration over time between each worm strain. Briefly, normalized MS peak area data for each metabolite were rank ordered across the 18 collection times and then grouped by strain. Significance testing, using strain as the predictor variable and rank as the dependent variable, was then performed (P < 0.05). This test was only sensitive to metabolite changes that were monotonic across time (up or down, see Fig. 1A). Nonetheless, even at this level of resolution, both isp-1 and mev-1 mutants are seen to produce several compounds in significantly altered amounts relative to wild-type worms (left two columns). Also, long-lived isp-1 Mit mutants markedly, and significantly, overproduce a variety of α-ketoacids and α-hydroxyacids relative to both N2 and mev-1 mutants (right column).

Figs. S2–S4. Cluster analysis of exometabolite data shown in Fig. 1A. A self-organizing map (SOM) algorithm (described under Material and Methods) was used to cluster metabolites that correlated temporally within the exometabolome data of N2, isp-1(qm150), and mev-1(kn1) animals; data are presented in Supplementary Figures 2, 3, and 4, respectively. In each figure, individual metabolite clusters are shown in (A). Metabolites that contributed to each cluster are listed in (B). In (A), individual metabolites are shown in gray. Black lines represent cluster averages. Metabolite levels were Z-score transformed before clustering.

Fig. S5. Exometabolome analysis of C. elegans Age mutants. A GLM analysis of the 47 exometabolome datasets collected from the strains listed in the lower Table (number of independent replicates is also shown) was undertaken to assess the differences and similarities between a variety of Age mutants and Mit mutants. Metabolites were collected from populations of 120 000 worms (per replicate) after an 18-h incubation period either in the presence or absence of oxygen. For each metabolite, normalized peak areas were rank-ordered across the 47 datasets, then binned into percentiles of 5%. A GLM was then employed for significance testing using strain as the predictor variable and rank as the dependent variable. No effort was made to test additional terms controlling for changes in collection date or experimenter. Contrasts were performed using SPSS (17.0). Significance was set at P < 0.0042 for each metabolite to account for the Type I error following the 12 tested comparisons. Notable is the almost indistinguishable nature of isp-1 and nuo-6 mutants, and of mev-1 and ucr-2.3 mutants. Only volatile metabolites (2-methylbutyric acid and all metabolites listed below it) were collected for clk-1(qm30). Nonvolatiles were collected using a different MS instrument and are analyzed separately in Supplementary Figures 6 and 7 below.

Figs. S6 and S7. Exometabolome analysis of additional C. elegans Age mutants. The +18-h exometabolomes of the following strains were collected and analyzed by GC-MS: wild-type worms (N2), wild-type worms exposed to anoxia during the collection period, isp-1(qm150) and clk-1(qm30) Mit mutants, short-lived mev-1(kn1) and ucr-2.3(pk732) ETC mutants, long-lived clk-2(qm37), and finally short-lived rad-5(mn159) mutants. The latter two mutants are allelic variants of the same genetic locus. We have segregated this data from that shown in Supplementary Figure 5 because it was collected using a different GC-MS instrument (refer to Materials and Methods for details). Columns represent independent experimental replicates (120 000 worms/replicate). Metabolite data were analyzed as described in Supplementary Figure 5, except binning was not used. In Supplementary Figure 6, data has been presented using hierarchical clustering – left panel, metabolite variation across row; right panel, metabolite variation relative to the entire array. Statistical analyses are provided in Supplementary Figures 7. Significance was set at P < 0.0062 for each metabolite to account for the expected increase in Type I error following the 10 tested comparisons.

Figs. S8 and S9. Mit mutants do not subsist on a metabolism normally reserved for growth under anaerobic conditions. Exometabolites were collected from the following strains after an 18-h incubation period either in the presence or absence of oxygen: N2, isp-1(qm150), clk-1(qm30), nuo-6(qm200), mev-1(kn1), and ucr-2.3(pk729). Metabolite levels were quantified using GC-MS and the effect of oxygen interrogated using a GLM. To monitor highly volatile compounds, a modified collection procedure was employed (refer to Materials and Methods). For the GLM, normalized peak areas were rank ordered across the 44 datasets then binned into percentiles of 5%. Oxygen served as the predictor variable and rank as the dependent variable. Contrasts were undertaken using SPSS (17.0). In Supplementary Figure 8, data has been segregated into (A) highly volatile (evaporative) compounds, and (B) standard assay compounds. Data are presented using hierarchical clustering – blue-yellow, metabolite variation across row; blue-red, metabolite variation relative to the entire array. Scale bars apply to all panels. Wild-type worms produce a distinct set of volatile compounds when exposed to anoxia – including butyric acid, isobutyric acid, 2-methylbutyric acid, isovaleric acid and tiglic acid (most pronounced in the lower display panel of (A)). There is an overt absence of these compounds from all normoxically cultured ETC mutants; suggesting that Mit mutants do not ectopically activate a metabolism normally reserved for survival under anoxia. Statistical analyses are presented in Supplementary Figure 9. The number of independent strain replicates is listed on the right. Significance testing was set at P < 0.017 and P < 0.0125 for the standard and volatile compounds, respectively. Although the data for isp-1, mev-1, and tpk-1 are clearly underpowered, even with this number of samples, the same compounds altered in the N2, nuo-6, ucr-2.3, and clk-1 following anaerobia are also trending toward significance in these animals. (−) indicates metabolite was either not detected or, for clk-1, samples were not collected.

Figs. S10–S15. Cluster analysis of exometabolite data shown in Fig. 2D. SOM clustering was used to identify groups of metabolites that correlated temporally across the exometabolome samples of each feeding RNAi condition in Fig. 2D. All metabolite levels were Z-score transformed before clustering. Treatment conditions included empty vector (pL4440), and DLD RNAi to empty vector ratios of 1:200, 1:20, 1:5, and 1:0. These five conditions are presented sequentially in Supplementary Figures 10 to 14. For each figure, the individual panels of (A) show unique metabolite clusters. Individual metabolites are shown in gray, and black line represents cluster averages. Only metabolites that contributed to each cluster are listed in (B). In Supplementary Figure 15, metabolite peak area data were analyzed similar to how the data for Supplementary Figure 1 were analyzed. That is, a GLM was used to capture the major changes in concentration over time between vector-treated worms and each of the DLD RNAi treatments. Briefly, normalized MS peak area data for each metabolite were rank ordered across the 30 collection times and then grouped by RNAi dilution. Significance testing, using RNAi concentration as the predictor variable and rank as the dependent variable, was then performed (P < 0.05). Contrast coding was performed with SPSS (17.0). As in Supplementary Figure 1, this test was only sensitive to metabolite changes that were monotonic across time (up or down, see Fig. 2D). Nonetheless, there is a significant and progressive change in the exometabolome profile of N2 worms as they suffer increasingly more severe loss of DLD. As predicted, the α-ketoacid levels are among the first metabolites affected.

Fig. S16. Correlation matrix showing metabolic similarity between long-lived Mit mutants, long-lived dld-1 disrupted animals, and long-lived tpk-1(qm162) mutants. The exometabolome profile of each of the strain replicates listed on the ordinate axis was used to calculate a distance measure between it and all other profiles. Distances were calculated using the CORRELATION algorithm of SPSS (17.0). In this procedure, degree of similarity was calculated as CORRELATION (x, y) = Σi(ZxiZyi)/N, where Zxi is the Z-score value of x for the ith metabolite, and N is the number of metabolites. Metabolites with missing peak area data were excluded from the calculation. No effort was made to adjust raw metabolite data for differences in collection date or experimenter. Within-strain correlations should approach 1.0; our data averaged 0.904 ± 0.089 (± 1 SD) (khaki). We set 0.726 as the lower (95%) boundary for the mean of same-strain comparisons (light blue). Red box highlights the similarity of tpk-1 mutants with nuo-6 and isp-1 Mit mutants, and with neat DLD RNAi-treated animals. Box in pure blue (bottom row) highlights similarity of neat DLD RNAi-treated animals also with nuo-6 and isp-1 Mit mutants.

Fig. S17. Summary statistics for survival curves shown in Fig. 2F. Increasing doses of RNAi targeting DLD (RNAi to empty vector ratio – 0:1, 1:1000, 1:500, 1:200, 1:100, 1:50, 1:20, 1:10, 1:5, and 1:0) were fed to wild-type C. elegans from the time of hatching and the effects on lifespan measured. Lifespan data for four independent replicates are shown. red indicates significantly different from vector, p < 0.005 (Bonferroni corrected).

Fig. S18. Individual disruption of the E1 subunits of α-ketoacid dehydrogenases does not increase wild-type lifespan. (A) RNAi targeting the E1α subunits of BCKADH and PDH or the E1 subunit of αKGDH was fed to worms from the time of hatching. All three RNAi treatments reduced adult size. (EV – empty vector (pL4440) control; all worms are of the same chronological age; scale bar: 200 μm). (B, C) Increasing doses of RNAi targeting individual E1 subunits (target gene to empty vector ratio – 0:1, 1:100, 1:50, 1:20, 1:10, 1:5, 1:2, and 1:0) were fed to wild-type C. elegans from the time of hatching and their effect on lifespan measured. Lifespan data represent a single dataset, except for vector control which was collected in triplicate (n = 60 worms/condition/replicate). In (B), asterisks indicates significantly different from vector average (pink line). Significance was set at P < 0.0024 (Bonferroni corrected).

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.