Non-enzymatic protein acylation as a carbon stress regulated by sirtuin deacylases

Introduction

Acetylation is an evolutionarily conserved post-translational modification occurring on lysine residues. Using acetyl-CoA as a donor, acetyltransferase enzymes catalyze site-specific acetylation of several proteins; a large family of deacetylase enzymes opposes this reaction by removing acetylation. This balanced regulation was first discovered on histones (Allfrey et al., 1964; Isenberg, 1979; Strahl and Allis, 2000), but has since been shown to regulate several biological processes, including gene and protein expression, cell survival, and cell death (Choudhary et al., 2009; Verma et al., 2012; Yi et al., 2011).

Sirtuins are one family of deacetylases that plays a role in a variety of important homeostatic and stress-related cellular processes. They have a unique ability to remove acetyl and other acyl groups from protein lysine side chains in a nicotinamide adenine dinucleotide (NAD⁺)-dependent manner (Imai et al., 2000). The founding member of the Sirtuin family, yeast Sir2p, was originally identified in a genetic screen for gene silencing factors (Rine et al., 1979). Interest in the sirtuin family was stimulated by the findings that yeast lifespan is extended by increasing Sir2 gene dosage and that Sir2 utilizes the metabolic coenzyme NAD⁺ to deacetylate its targets, thereby providing a link between metabolism, gene expression, and lifespan (Imai et al., 2000; Kaeberlein et al., 1999).

Mammalian sirtuins are grouped into four phylogenetic classes: Class I contains SIRT1-3, class II contains SIRT4, class III contains SIRT5, and class IV contains SIRT6 and SIRT7 (Frye, 2000). All mammalian sirtuins were originally thought to have deacetylase activity owing to their homology to yeast Sir2. Indeed, class I sirtuins (SIRT1-3) have robust deacetylase activity on a wide variety of substrates (Michishita et al., 2005). Because the remaining classes of sirtuins (SIRT4-7) have low deacetylase activity on tested substrates, they were thought to be either highly specific deacetylases or ADP-ribosyltransferases. Recently, SIRT5 was shown to have robust lysine desuccinylase and demalonylase activities, expanding the known enzymatic activities of the sirtuin family (Du et al., 2011; Peng et al., 2011). These findings suggested that the different sirtuins in different phylogenetic classes could harbor not-yet-discovered enzymatic activities (Hirschey, 2011). This idea is supported by recent studies demonstrating that SIRT6 has efficient lysine depalmitoylase and demyristoylase activity (Jiang et al., 2013). Another recent study describes deglutarylase activity for SIRT5 (Tan, 2014). Remarkably, another study showed in vitro that many of the sirtuins are capable of hydrolyzing long-chain acyl, including octanoyl-, decanoyl-, and dodecanoyl-lysines (Feldman et al., 2013). However, the extent to which sirtuins are removing these longer-chain acyl modifications in vivo has not yet been explored. Together, these studies show that mammalian sirtuins are not merely deacetylases, but are NAD⁺-dependent deacylases capable of acting on a several acyl substrates.

Sirtuins have important roles in a wide range of mammalian biological processes, including metabolic homeostasis (e.g. fasting, CR, high-fat diet feeding), and resistance to stresses (e.g. DNA damage, ROS, protein misfolding) (Cheng et al., 2003; Finley et al., 2011; Hirschey et al., 2011; Toiber et al., 2013; Westerheide et al., 2009). Oftentimes sirtuin-mediated deacylation is associated favorable cellular processes (Hall et al., 2013), and models of sirtuin ablation or knock-down are associated with increased incidence or acceleration of disease (Finkel et al., 2009). However, sirtuin ablation is not always pathogenic. For example, mice lacking SIRT2 are markedly protected from ischemic injury (Narayan et al., 2012). Additionally, SIRT1 has been described to both promote and suppress tumorigenesis, depending on the molecular and cellular context (Yuan et al., 2013). However, the mammalian sirtuins are collectively viewed to increase stress resistance by remodeling metabolism, altering inflammatory processes, and enhancing oxidative stress responses (Hall et al., 2013). Yet, how and why removing an acyl-group from a protein is generally associated these processes is unclear.

Despite the number of studies on the sirtuins and their function, less is known about the family of acetyltransferase or acyltransferase enzymes opposing sirtuin enzymatic activity. Lysine acetyltransferases such as those in the GNAT, MYST, and P300/CBP superfamilies have been well-studied in the context of chromatin regulation, but also regulate the stability, localization, and interactions of several extra-nuclear protein targets (Jiang et al., 2011; Lin et al., 2013; Zhao et al., 2013); for a recent review, see (Friedmann and Marmorstein, 2013). Remarkably, none of these acetyltransferases have known roles in the mitochondria where protein acylation is highly abundant. In this compartment, little is known about the mechanism of acylation. One report described a component of an acetyltransferase, or the acetyltransferase machinery, in the mitochondria called GCN5L1 (Scott et al., 2012). Another recent report described an acetyltransferase activity for the mitochondrial enzyme ACAT1 (Fan et al., 2014). However, more studies on the enzymology of these reactions are required to determine how these proteins catalyze acetyl transfer to proteins.

In contrast, recent and historical evidence supports the idea that non-enzymatic acetylation and acylation can lead to protein modification (Wagner and Payne, 2013; Weinert et al., 2013b). Indeed, the idea of non-enzymatic mechanisms is emerging as a possible contributor to the protein acetylation and acylation landscape (Guan and Xiong, 2011; Newman et al., 2012). In this review, we analyze the evidence for non-enzymatic protein acetylation and acylation, and discuss the ability of the sirtuins to regulate these nonenzymatic processes. Finally, we propose a model where the sirtuins are regulators of nonenzymatic acylation, and are therefore important mediators of overall protein quality control (Figure 1).

Non-enzymatic Protein Acetylation

The use and interconversion of carbon substrates is a near universal property of what defines life. Within this simplistic concept of cellular metabolism, chemical reactivity is a fundamental and often overlooked requirement for metabolism to operate. Whether metabolism is pro-growth (anabolic), pro-breakdown (catabolic), or both (amphibolic), metabolites and metabolic intermediates must be chemically reactive for physiological use.

The reactivity of a cellular metabolite is often dependent on the presence or absence of nucleophilic or electrophilic moieties and/or leaving groups. For example, the carbonyl group is electrophilic and reactive, and certain metabolic enzymes can further “activate” carbonyl-containing compounds by covalently adding additional electronegative functional groups, such as thiols or phosphates. These additional groups are also known as leaving groups for their tendency to “leave” the compound after a nucleophilic attack, and the inherent reactivity of a given chemical or metabolite can often be determined by whether or not it contains a “good” leaving group (Figure 2A).

Compounds containing a reactive thioester group in the form of coenzyme A (CoA) are particularly prevalent throughout metabolism in diverse organisms. The activation of metabolites with a labile thioester bond facilitates many biochemical processes including de novo lipogenesis, the TCA cycle, ketogenesis, amino acid metabolism, as well as protein lysine acetylation. The chemical versatility of the thioester bond inspired the “Thioester World” hypothesis, which suggests a central role of this chemical group in the process of abiogenesis and early chemical evolution (De Duve, 1995).

While organisms have evolved to rely on intrinsically reactive metabolites to carry out their vital cellular functions, the utility of reactive metabolites in metabolism also poses a challenge: reactive metabolic intermediates could spontaneously react with nearby proteins. For example, acetylation in bacteria correlates strongly with fluctuating levels of the highly reactive intermediate acetyl-phosphate (Figure 2A) (Weinert et al., 2013b). In this study, the majority of protein lysine acetylation in E. coli occurs independently of its only known acetyltransferase, YfiQ. Furthermore, incubating bacterial proteins with acetyl-phosphate in vitro causes acetylation at nearly all the sites identified in vivo, and approximately 10% of the putatively non-enzymatic acetylation events are targeted by the sirtuin deacetylase CobB (Weinert et al., 2013b). This study provides strong evidence that acetyl-phosphate mediates global shifts in protein acetylation via non-enzymatic mechanisms, and that the sirtuin CobB regulates only a subset of these chemical acetylation events.

Non-enzymatic protein lysine acetylation mediated by reactive acetyl-thiol compounds (i.e. acetyl-CoA) has been reported on human serum albumin, purified histones, lysine-containing polypeptides, and recombinant bacterial proteins (Baddiley et al., 1952; d’Alayer et al., 2007; Paik et al., 1970). In fact, acetyl-CoA has historically been described as a suitable “acetylating agent” (Delpech et al., 1983; Paik et al., 1970). More recently, nonenzymatic reactions between acetyl-CoA and protein lysines have been shown and discussed in context of their ability to contribute to the acetylation landscape in biological systems.

For example, one study showed that while the acetyltransferase GCN5 can catalyze acetylation of proteins, neither the enzyme nor its activity is required for acetylation in some conditions (Tanner et al., 2000; Tanner et al., 1999). Specifically, the catalytic mechanism of the GCN5 acetyltransferase involves deprotonating the epsilon amino group of a lysine residue, which enables it to act as a nucleophile towards the electrophilic carbonyl carbon in acetyl-CoA, resulting in an acetylated lysine. Because deprotonation of lysine residues is pH-dependent, acetyl-CoA incubation with catalytically-dead GCN5 or no-enzyme control reactions performed at alkaline pH still leads to lysine acetylation, primarily by increasing the number of amino groups acting as nucleophiles towards the inherently reactive acetyl-CoA. These data suggest that some cytoplasmic and/or nuclear protein acetylation could occur without its acetyltransferase.

As another example, the environment of the mitochondrial matrix, which has an alkaline pH and abundant acetyl-CoA, could make non-enzymatic acetylation favorable (Casey et al., 2010; Garland et al., 1965). One recent study demonstrated that reproducing these conditions in vitro is sufficient to cause non-enzymatic acetylation of mitochondrial and non-mitochondrial proteins using acetyl-CoA as a substrate (Wagner and Payne, 2013). An additional study showed that acetylation accumulates in growth-arrested Saccharomyces cerevisiae that depended on acetyl-CoA generation in distinct subcellular compartments (Weinert et al., 2013a), suggesting that non-enzymatic protein acetylation contributes to the overall acetylation landscape.

Further supporting this idea, recent findings from eukaryotic systems indicate that the stoichiometry of acetyl-lysine modifications is low. For example, 86% of acetylation sites identified in yeast are estimated to have less than 1% stoichiometry (Weinert et al., 2013a). These findings are in contrast to the best measurements of stoichiometry of phosphorylation in yeast, where only 11% of phosphorylation sites show less than 1% stoichiometry (Wu et al., 2011). Interestingly, the most abundant (high stoichiometric) acetylation sites in yeast occurred on histones, proteins present in histone deacetylase (HDAC) or histone acetyltransferase (HAT) complexes, and on transcription factors, supporting the idea that acetylation can be regulated in these complexes (Weinert et al., 2013a). In contrast, baseline mitochondrial protein acetylation levels are low in yeast, and increase markedly upon increasing acetyl-CoA concentrations in vivo by manipulating acetyl-CoA consuming enzymes (citrate synthase), or were reduced when an acetyl-CoA generating enzyme was ablated (pyruvate dehydrogenase). These studies indicate that a large portion of protein lysine acylation lacks the high levels of stoichiometry typically associated with regulatory signaling modifications such as phosphorylation, supporting the idea that some protein acetylation could occur in a non-targeted, non-signaling manner (i.e. non-enzymatically).

An important caveat is that these studies were performed in yeast, and the stoichiometry of acetylation was only estimated indirectly. Therefore, absolute levels of mammalian protein acetylation are not known. The major reason for this lack of understanding is because measuring the stoichiometry of global acetylation sites accurately would require thousands of isotopically labeled, custom-synthesized, internal standards and is therefore impractical with current technology. Thus, studies will generally determine “maximal median stoichiometry” by normalizing acetyl-lysine ratios to protein ratios and reporting the findings as relative values (Hebert et al., 2013; Rardin et al., 2013b). Using these techniques, low overall stoichiometry of protein acetylation is still observed.

Together, these studies support the notion that non-enzymatic protein acetylation by acetyl-CoA could occur, and would be especially favorable in the mitochondria. Importantly, low-level acetylation mediated by a promiscuous acetyltransferase that is sensitive to changes in acetyl-CoA levels cannot be formally excluded. However, the historical ability of acetyl-CoA to acetylate a wide variety of substrates, coupled with recent evidence of the dependence on the generation of distinct acetyl-CoA pools for acetylation, are highly consistent with a non-enzymatic mechanism contributing to the global protein acetylation landscape.

Non-enzymatic Protein Acylation

This model further suggests that non-enzymatic acetylation might not be limited to acetyl-CoA alone. Like acetyl-CoA, succinyl-CoA and malonyl-CoA contain a thioester bond which is intrinsically reactive toward nucleophiles (Figure 2A). Indeed, protein lysine succinylation and malonylation was found on a wide-range of mitochondrial, cytosolic, and nuclear proteins (Du et al., 2011; Peng et al., 2011; Zhang et al., 2010).

The reactive nature of succinyl-CoA suggests that manipulating concentrations of this intermediate in vivo might alter protein succinylation. Indeed, historical studies demonstrated non-enzymatic protein succinylation could be regulated by succinyl-CoA concentrations and was a potential control mechanism for ketogenesis (Lowe and Tubbs, 1985; Quant et al., 1990). More recently, succinyl-CoA has been shown to non-enzymatically succinylate protein lysine residues in a pH-dependent manner at micromolar concentrations (Wagner and Payne, 2013; Weinert et al., 2013c). Furthermore, deleting succinyl-CoA generating and consuming enzymes in the TCA cycle, in yeast, caused significant global decreases and increases in succinylation, respectively (Weinert et al., 2013c).

Similarly, a new protein modification called lysine glutarylation is enriched on mitochondrial proteins (Tan, 2014). Like acetyl-CoA and succinyl-CoA, glutaryl-CoA is primarily generated in mitochondria and protein glutarylation is found on a wide-range of metabolic proteins. In vitro, incubation of heat-inactivated mitochondrial proteins with micromolar concentrations of glutaryl-CoA, but not glutarate, increases protein glutarylation. In vivo, glutarylation increases when flies are fed glutaryl-CoA precursors or when protein catabolism is increased.

The prevalence of both lysine acetylation and malonylation/succinylation/glutarylation on proteins involved in mitochondrial metabolism suggests something unique about this cellular compartment. Furthermore, similar to lysine acetylation, a significant proportion of succinylated proteins are localized to mitochondria, with less succinylation among cytosolic and nuclear proteins (Park et al., 2013). Indeed, the mitochondrial matrix is the primary site of acyl-CoA metabolism, where acetyl-CoA and succinyl-CoA intermediates reach estimated steady-state concentrations in the millimolar range (Garland et al., 1965; Hansford and Johnson, 1975).

Interestingly, early estimates on the levels and stoichiometry of succinylation have shown 68% of succinylation sites quantified from a subset of proteins in mouse embryonic fibroblasts display less than 10% stoichiometry (Park et al., 2013). These results are in agreement with the low stoichiometry observations of acetylation, described above. Additionally, despite the different chemical properties of acetyl-lysine and succinyl-lysine modifications, significant overlap is found between the protein lysine residues that are acetylated and succinylated (Rardin et al., 2013a; Weinert et al., 2013c), supporting the idea that some lysines are susceptible to non-enzymatic acylation by reactive acyl-CoAs.

Lysine malonyltransferase, succinyltransferase, or glutaryltransferase enzymes have not been identified in any subcellular compartment (Park et al., 2013; Weinert et al., 2013c). In the mitochondria, the biochemical properties of the protein microenvironment, such as metabolite concentration and pH, may be facilitating non-enzymatic acylation reactions. While enzymatic acyltransfer cannot be ruled out, acyl-CoA concentrations in the mitochondria are several hundred-fold higher than the kinetic binding constants (K_m) of characterized lysine acetyltransferases (Albaugh et al., 2011). Thus, if an acyltransferase resides in the mitochondria that is kinetically comparable to known acetyltransferases, its catalytic activity would be fully saturated at all times. This would either lead to high stoichiometric levels of acylation (which has not yet been observed), or would leave a substantial excess of reactive acetyl-CoA in a steady state to potentially react with protein amine groups.

Together, these studies suggest that non-enzymatic protein lysine acylation may occur in diverse organisms and is closely correlated with perturbations of acyl-CoA metabolism. Furthermore, protein acylation would be particularly favorable in the mitochondrial compartment. Studies in this area so far suggest that non-enzymatic reactions between metabolites and proteins might be a common mechanism of protein acylation.

Protein Acylation as Carbon Stress

One of the major challenges of studying protein acetylation and acylation is understanding the overall function of these post-translational modifications. Indeed, the seemingly infinite number of possible combinations of modifications, on a given protein, amplified across the proteome is daunting. So far, investigations on acylation have either quantified the acyl proteome using mass spectrometry-based proteomics, taken a one-target-at-a-time approach, or both. Large-scale proteomic surveys have identified metabolic proteins are over-represented in acetylation surveys, suggesting that acetylation could regulate cellular metabolism. Acetylation of metabolic proteins outside the mitochondria can be activating, inhibitory, or have no measureable consequence. However, one theme that is emerging across mitochondrial metabolic studies is that the effect of protein acetylation is consistently inhibitory, especially to mitochondrial energy-producing processes (Ghanta et al., 2013). Interestingly, protein acetylation and acylation have significant overlap in the mitochondria (Rardin et al., 2013a). Together, these observations support the idea that protein acylation in the mitochondria is largely inhibitory.

Based on a non-enzymatic mechanism of action, we propose that protein acylation is one form of “carbon stress,” in which the accumulation of intrinsically reactive carbon metabolites can directly and negatively alter protein function and disrupt cellular homeostasis. Using this definition, other endogenous examples of carbon stress include protein glycation, carbamylation (Figure 2B), and the recently identified 3-phosphoglycerylation (Moellering and Cravatt, 2013). All of these modifications can be driven by non-enzymatic reactions of metabolites with proteins. Under basal conditions, these modifications might occur at a low level. During physiological or pathological states that increase the amounts of reactive metabolite, non-enzymatic reactions could also increase and have a detrimental effect on protein function; in this setting, these reactions would be considered a form of “carbon stress.”

Acylation could alter protein function by a variety of mechanisms. For example, lysine acetylation neutralizes a positive charge and alters hydrogen bonding potential, which could disrupt protein-protein or protein-subunit interactions, enzyme-substrate and cofactor binding, and allosteric regulation. Some of these mechanisms have been demonstrated experimentally (Bharathi et al., 2013), while others are predicted through protein structural analyses (Hebert et al., 2013; Shimazu et al., 2010; Tao et al., 2010). Together, these studies support the idea that unregulated protein acylation could broadly interfere with protein function and compromise cellular health, potentially leading to disease.

Carbon Stress and Sirtuin-Mediated Protein Quality Control

Sirtuins are a conserved family of NAD⁺-dependent protein deacylases that are important in response to cellular stress and homeostasis. Mounting evidence shows that the sirtuins act on a wide variety of substrates. While the role of sirtuins is complex, and can be unique in certain tissues or organ systems, sirtuin-mediated deacylation is often associated with increased protein function and metabolic health; for a review, see (Finkel et al., 2009).

Indeed, increases in mitochondrial protein acetylation (by loss or ablation of SIRT3) leads to acceleration and/or development of several disease states, including cardiac hypertrophy, hearing loss, the metabolic syndrome, and reduced regenerative capacity of hematopoietic stem cells (Brown et al., 2013; Hafner et al., 2010; Hirschey et al., 2011; Someya et al., 2010). Furthermore, hypersuccinylation (by ablation of SIRT5) of numerous mitochondrial and non-mitochondrial proteins (Park et al., 2013; Yu et al., 2013) leads to hyperammonemia (Nakagawa et al., 2009) and reduced fatty acid oxidation and less ketogenesis (Rardin et al., 2013a). Finally, protein hyperglutarylation (by ablation of SIRT5) is also associated with reduced carbamoyl phosphate synthase 1 activity (Tan, 2014). Thus, protein acylation may be a form of non-enzymatic chemical damage which alters protein function and compromises cellular health.

Interestingly, studies investigating the effects of acetylation on specific SIRT3 target enzymes indicate that many known acetylation events do not affect enzymatic activity (Bharathi et al., 2013; Hallows et al., 2006; Hirschey et al., 2010; Shimazu et al., 2010; Tao et al., 2010; Yu et al., 2012). Large-scale proteomic surveys support this idea, and have shown that the overall amount of acetylation predicted to be regulated by SIRT3 in the mitochondria is low (~20%) compared to the majority of mitochondrial protein acetylation, which remains relatively static (Hebert et al., 2013). These findings support a model in which a subset of protein acetylation has consequences for protein function, and are regulated by SIRT3, whereas others sites of protein acetylation have no relevance on protein function. Interestingly, this model is analogous to how the bacterial sirtuin CobB regulates protein acetylation during stationary phase growth: when concentrations of reactive acetyl-phosphate and global protein acetylation concurrently increase, CobB removes a subset of protein acetylation (Figures 3A and 3B) (Weinert et al., 2013b).

Importantly, this does not imply that only acylation events targeted by sirtuins have functional consequences. Indeed, some acylation events could occur on catalytic lysines, which are not accessible to or targeted by sirtuin deacylation, but would have an irreversible negative effect on enzymatic activity. However, examples of these types of acylation (i.e. non-sirtuin targeted but functionally relevant) have not yet been described in the literature.

Together, these studies suggest that if non-enzymatic acylation is a contributor to mitochondrial protein acylation and a subset of these modifications are detrimental to protein function, then this form of stress would require a protein quality control response to remove spurious chemical modifications on proteins; indeed, the sirtuins could serve this exact role.

Supporting this idea, some sirtuins might respond to or be upregulated during carbon stress when protein acylation increases. During calorie restriction in mice, global hepatic mitochondrial protein acetylation increases while whole tissue acetyl-CoA levels fall, suggesting increased utilization of acetyl-CoA in mitochondria for energy requiring processes (Hebert et al., 2013). Interestingly, SIRT3 expression increases during calorie restriction, suggesting that SIRT3 may be responding to a condition when increased non-enzymatic protein acetylation would be expected due to an increased mitochondrial acetyl-CoA burden. Mice fed a high-fat diet for one week also show increased SIRT3 expression, further suggesting that SIRT3 is responding to conditions predicted to generate more mitochondrial acetyl-CoA (Hirschey et al., 2011).

While mitochondrial SIRT3 and SIRT5 are well-positioned to play a role in managing non-enzymatic lysine acylation, one of the major questions remaining is whether these proposed roles are broadly applicable to the rest of the sirtuin family. Acetyl-CoA and succinyl-CoA are particularly abundant metabolic intermediates; however there are many other acyl-CoA intermediates utilized throughout cellular metabolism, all of which contain the inherently reactive thioester bond and many of which function outside the mitochondrial compartment (Figure 2). For example, many other endogenous acyl-lysine modifications have been validated, including propionylation, butyrylation, crotonylation, and glutarylation. Importantly, these modifications can be removed by HDACs and sirtuins in vitro (Feldman et al., 2013; Madsen and Olsen, 2012). Additionally, longer chain acyl-lysine modifications occur in cells and are targeted for removal by SIRT6 in vivo (Jiang et al., 2013).

While no dedicated lysine propionyltransferase or butyryltransferase has been identified, some acetyltransferases can catalyze propionylation and butyrylation of histones at substantially lower efficiency than acetylation (Albaugh et al., 2011). Importantly, the influence of potential unidentified lysine acyltransferase enzymes cannot be formally excluded. However, the common reactive nature of acyl-CoA intermediates, coupled with recent findings indicating that sirtuins are capable of removing a broad range of acyl-lysine modifications, provides the rationale for the sirtuins to serve a role to remove protein acylation, reduce carbon stress, and maintain overall protein quality control.

Analogous Stress and Protein Quality Control Systems

Several analogous stressors and protein quality control systems are found throughout biology. Organisms experience several types of physiological stress and have evolved systems to maintain protein quality control and overall protein function. Oxidative stress is a quintessential example, which is generated from reactive oxygen species (ROS). For aerobic organisms, ROS are an unavoidable consequence of using molecular oxygen as the terminal election acceptor during cellular respiration (Davies, 2000). Cellular ROS, such as superoxide and hydroxyradicals, can spontaneously react with and damage cellular macromolecules including DNA, lipids, and proteins (Figure 3C). Biological systems have evolved a suite of quality control strategies to mitigate the harmful effects of ROS. For example, catalase, superoxide dismutases, peroxiredoxins, thioredoxins, glutaredoxins, glutathione, glutathione peroxidases, and methionine sulfoxide reductases all represent families of enzymes which have evolved to directly or indirectly cope with harmful ROS and thereby mitigate potentially damaging non-enzymatic modification to macromolecules (Meyer et al., 2009; Muller et al., 2007).

The importance of these endogenous antioxidant systems in mammalian organisms is demonstrated by the pathological phenotypes elicited by genetically ablating their individual components. For instance, mice lacking the ability to detoxify mitochondrial superoxide due to homozygous deletion of manganese superoxide dismutase (MnSOD or SOD2) develop cardiomyopathy, acidosis, and are neonatal lethal (Li et al., 1995). An inability to detoxify lipid peroxides via germline ablation of glutathione peroxidase 4 (GPX4) results in embryonic lethality, while inducible deletion of GPX4 in adult mice causes rapid degeneration and death (Seiler et al., 2008). Furthermore, genetic ablation of mitochondrial methionine sulfoxide reductase (MsrA) in mice causes an increased sensitivity to oxidative stress, increased protein carbonyl adducts, and a 40% decrease in lifespan (Moskovitz et al., 2001). Moreover, impairments or alterations in many of these antioxidant systems are associated with tumorigenesis in mice and humans (De Luca et al., 2010; Muller et al., 2007). Indeed, oxidative stress and reduced protein quality control have been implicated in many diseases.

Remarkably, its counterpart called “reductive stress” also contributes to reduced protein quality control. Reductive stress is an abnormal increase of reducing equivalents (e.g. reduced glutathione [GSH] or NADPH) that contributes to the molecular pathogenesis of disease (Rajasekaran et al., 2007). In one example, a transgenic mouse model that mimics a human cardiomyopathy caused by a mutation in the gene encoding αB-crystallin (R120GCryAB) had enhanced generation of the reducing equivalents NADPH and GSH, increased protein aggregation, and developed cardiomyopathy.

Additionally, nitrosative stress is a unique type of redox stress induced by nitric oxide or NO-related species (e.g. including S-nitrosothiols) which lowers intracellular thiol levels (Figure 3D) (Hausladen et al., 1996). One cellular and genetic response to this stress is the direct activation of transcription, which constitutes an adaptation to nitrosative stress.

Thus, organisms that experience oxidative, reductive, and nitrosative stresses, and perhaps carbon stress, all require mechanisms to counteract these ongoing stressors. Several systems have evolved to do so, and genetic ablation of these quality control pathways reduces the ability to respond to stress and overall organismal fitness; remarkably, ablation of the sirtuin genes is associated with reduced ability to respond to stress and reduced overall organismal fitness.

Role of Carbon Stress in Disease

Increased carbon stress and/or dysregulation of a carbon stress response system could be a general feature of disease, with reduced protein quality control and therefore reduced protein function as a common underlying mechanism. Indeed, several examples of endogenous carbon metabolites undergoing non-enzymatic reactions with protein amine groups contribute to disease. In the setting of diabetes mellitus, prolonged elevation of aldehyde-containing sugars in the blood and intracellular dicarbonyl compounds cause excessive non-enzymatic glycation of protein lysine and arginine residues which can further react to form advanced glycation end products (AGEs) (Figure 2B) (Thornalley, 2005). Non-enzymatic protein glycation and AGE formation contribute to the pathophysiology of diabetes and are associated with cardiovascular and retinal complications by interfering with protein function and cellular signaling (Brownlee, 2001; Fukami et al., 2013).

Remarkably, Louis-Camille Maillard, who first reported the protein glycation process over 100 years ago using an in vitro system, predicted that this reaction may be relevant in diabetes (Maillard, 1912). However, it was not until the 1980s that the role of endogenous non-enzymatic protein glycation and AGEs in mediating diabetic complications was recognized (Garlick et al., 1984; Ulrich and Cerami, 2001). Interestingly, metabolism of glycation-inducing sugars is predicted to be a key evolutionary driver for the Glyoxylase I and II enzyme system, which serves to detoxify intracellular glycation-inducing dicarbonyl compounds such as methylglyoxal (Rabbani and Thornalley, 2012). Thus, aldehyde-containing sugars are carbon sources essential for viability of many cell-types, but they and their byproducts can react with and modify cellular proteins and therefore require coordinate detoxification mechanisms.

In addition to sugar metabolism, lipid metabolism can lead to protein modification. For example, oxidation of polyunsaturated fatty acids generates lipid aldehydes such as malondialdehyde and 4-hydroxy-2-nonenal which are capable of forming harmful adducts with protein lysine, cysteine, and histidine residues (Dalle-Donne et al., 2003). Furthermore, the metabolism of ethanol produces the toxic reactive intermediate acetaldehyde which can similarly form harmful adducts with protein or DNA (Langevin et al., 2011). The ongoing formation of toxic aldehyde species during metabolism could have provided the evolutionary impetus for the aldehyde dehydrogenase superfamily of enzymes, which together can detoxify both lipid and ethanol-derived aldehyde-containing metabolites (Koppaka et al., 2012). The importance of aldehyde detoxification in mammals is evinced by studies of the aldehyde dehydrogenase 2 (ALDH2) knockout mouse or humans carrying the near inactive variant of ALDH2 (ALDH2*2), which demonstrate severe ethanol sensitivity and an increased risk of several cancers (Yokoyama et al., 1998). Furthermore, excessive protein aldehyde adducts are associated with many human diseases including Alzheimer’s disease, Parkinson’s disease, and diabetes (Dalle-Donne et al., 2003).

Other non-enzymatic reactions involving metabolites are implicated in disease but lack known quality control or detoxifying mechanisms. Similar to the mechanism of protein glycation in diabetes, prolonged elevation of cyanate in the blood of smokers or during renal failure can cause harmful non-enzymatic protein lysine carbamylation (Figure 2B) (Fluckiger et al., 1981; Stark, 1965). Excessive protein carbamylation may be a common mechanism underlying the pathophysiology of inflammation and coronary artery disease (Wang et al., 2007).

Separately, recent work has demonstrated that the glycolytic metabolite and acyl-phosphate 1,3-bisphosphoglycerate (Figure 2) can non-enzymatically modify protein lysine residues by virtue of its highly reactive electrophilic carbonyl carbon (Moellering and Cravatt, 2013). This novel modification, 3-phosphoglyceryl-lysine, was further shown to impair the activity of select glycolytic enzymes, but is not yet associated with any disease states.

More commonly, drug toxicity is often caused by electrophilic drug metabolites nonenzymatically reacting with cellular proteins, and minimizing this problem is one of the foremost challenges during the development and optimization of small-molecule pharmaceuticals (Evans et al., 2004; Horng and Benet, 2013). Interestingly, the reactivity and toxicity of many drug metabolites is dependent on their bioactivation to acyl-CoA thioesters (Sallustio et al., 2000). Collectively, these studies demonstrate that non-enzymatic reactions of endogenously-generated metabolites with cellular proteins contributes to the pathophysiology of diabetes, neurodegeneration, kidney failure, inflammation, coronary artery disease, and drug toxicity.

Carbon Stress, Protein Quality Control, and Aging

Many of the diseases associated with carbon stress and reduced protein quality control (diabetes, cardiovascular disease, neurodegeneration, cancer; described above) are also associated with aging. Sirtuins are also implicated in regulating diseases of aging, as well as in mediating the salutary effects of caloric restriction (Finkel et al., 2009). When sirtuins are examined from the perspective of protein quality control, perhaps it is not surprising that sirtuin-mediated deacylation of cellular proteins has generally protective benefits for cellular and organismal health. Just as genetic loss of conserved protein quality control mechanisms accelerates the development of age-related diseases and/or shortens lifespan in models systems, so does the loss of many sirtuin family members. Conversely, pharmacologic or genetic stimulation of protein quality control machinery can extend lifespan in diverse organisms (Powers et al., 2009).

Recent evidence indicates that a stoichiometric imbalance between mitochondrial genome-encoded and nuclear genome-encoded subunits of the respiratory chain induces mitochondrial protein quality control machinery, which extends lifespan in C. elegans through a hormetic effect and is correlated with lifespan in mice (Houtkooper et al., 2013). Interestingly, further evidence indicates that the longevity-promoting effects of mitochondrial protein quality control are stimulated by SIRT1 and by raising cellular NAD⁺ levels, suggesting that in addition to the proposed direct effects of sirtuin-mediated deacylation on protein quality control, sirtuins also indirectly regulate the protein quality control response (Mouchiroud et al., 2013). Separately, C. elegans lifespan can be extended by respiratory chain impairment which also induces mitochondrial protein quality control systems (Durieux et al., 2011). These studies offer strong evidence that the aging process in lower organisms and perhaps the development of age-related disease in mammals is tightly linked to protein quality control, one component of which may be sirtuin-mediated deacylation.

Sirtuins are often implicated in the physiological adaptations to calorie restriction (CR), which extends lifespan and delays the onset of age-related disease in mammals (Guarente, 2013). Interestingly, gene expression profiling of CR mice suggests that cellular mechanisms of proteostasis and macromolecular quality control are key mediators of the CR-induced delay in age-associated pathological changes (Lee et al., 1999). Thus, the involvement of sirtuins in mediating the beneficial effects of CR is consistent with the model that sirtuins support health by removing harmful non-enzymatic acyl-modifications and enhancing protein quality control.

Future Studies

While this model might be useful for understanding the role of acetylation as a post-translational modification compared to other signaling modifications (i.e. phosphorylation), several questions and challenges remain. For example, this model relies on the concept of non-enzymatic acylation, which is challenging to test experimentally. Even in the most rigorous in vivo studies, the possibility of enzymatic transfer by a not-yet-discovered transferase remains. However, despite similar experimental challenges for protein modifications like glycation and carbamylation, these reactions have become accepted as non-enzymatic due to the large body of evidence supporting their reactive nature and their correlation with disease states in which the reactive metabolite is elevated.

In the setting of the cytoplasm or nucleus, where acyltransferases play well-established roles in acetylating several cellular proteins, disentangling the balance between enzymatic and non-enzymatic acyl transfer will prove to be difficult. Based on in vitro studies, the mitochondrial compartment has a pH that is permissive for non-enzymatic acetylation, whereas the cytoplasmic and nuclear compartments are less permissive. However, again based on in vitro evidence, all three compartments could be permissive for non-enzymatic acylation, such as succinylation (Wagner and Payne, 2013). Therefore, the difference between enzymatic and non-enzymatic acylation with respect to sub-cellular compartments as well as the difference in these events on regulation and signaling, will be an important area of future study.

Conclusions

If life originated as a primordial soup of stochastic organic chemical reactions, it is not surprising that some unintended reactions occur between metabolites and proteins. In this setting, the endogenous formation of proteotoxic carbon conjugates is an ongoing and unavoidable consequence of carbon metabolism, which can contribute to protein dysfunction and disease. To prevent and/or reverse this, a variety of cellular quality control and stress resistance programs have evolved. We propose that the sirtuin enzymes are part of the global protein quality control family, and they serve to remove suppressive and potentially damaging non-enzymatic acyl-modifications. This conceptual framework could explain the broad physiological roles of the sirtuins in stress resistance, healthspan, and in some cases organismal lifespan. Continued work in this exciting area will surely uncover new areas of biology regulated by the sirtuin family, and how a myriad of cellular processes are regulated by acylation and deacylation.