Mitochondrial metabolism is intimately connected to the universal coenzyme NAD. In addition to its role in redox reactions of energy transduction, NAD serves as substrate in regulatory reactions that lead to its degradation. Importantly, all types of the known NAD-consuming signalling reactions have been reported to take place in mitochondria. These reactions include the generation of second messengers, as well as post-translational protein modifications such as ADP-ribosylation and protein deacetylation. Therefore, the availability and redox state of NAD emerged as important factors in the regulation of mitochondrial metabolism. Molecular mechanisms and targets of mitochondrial NAD-dependent protein deacetylation and mono-ADP-ribosylation have been established, whereas poly-ADP-ribosylation and NAD-derived messenger generation in the organelles await in-depth characterization. In this review, we highlight the major NAD-dependent reactions occurring within mitochondria and describe their metabolic and regulatory functions. We also discuss the metabolic fates of the NAD-degradation products, nicotinamide and ADP-ribose, and how the mitochondrial NAD pool is restored.
General control of amino acid synthesis protein 5-like 1
nucleoside diphosphate-linked moiety X
reactive oxygen species
transient receptor potential cation channel 2
Mitochondria are vital organelles that carry out important functions in almost all eukaryotic cells. They contain their own genetic material and protein translation machinery and possess two membranes which enclose specialized subcompartments. The mitochondrial matrix, the lumen enclosed by the inner mitochondrial membrane, accommodates enzymes of key metabolic pathways including Krebs cycle, β-oxidation of fatty acids, amino acid catabolism and the urea cycle. Mitochondria are the major site of energy transduction and ATP production as the oxidative phosphorylation machinery is embedded in the inner mitochondrial membrane. Given the central role of mitochondria in energy metabolism, their dysfunction has been associated with various diseases including neurodegenerative disorders, diabetes and cancer [1-3].
Many mitochondrial processes depend on the universal coenzyme nicotinamide adenine dinucleotide (NAD) or its phosphorylated counterpart NADP (Fig. 1). The redox reactions involve the reversible hydride transfer at the nicotinamide moiety of NAD(P), resulting in a switch between oxidized (NAD+, NADP+) and reduced (NADH, NADPH) forms of the nucleotides (Fig. 1). Mitochondria usually contain a major portion of the cellular NAD content, with up to 70% of the cellular pool, depending on tissue and cell type [4, 5]. Mitochondrial oxidative metabolism results in a rather low NAD+/NADH ratio of ~ 7–8 compared with the cytosolic redox ratio of up to 700 . Therefore, both the availability and redox state of NAD have a pronounced influence on all major mitochondrial metabolic processes (Fig. 2).
In addition to participation in redox reactions, NAD is a versatile cellular signalling molecule . All NAD-dependent signalling reactions involve cleavage of the N-glycosidic bond between the nicotinamide ring and the ribose. The remaining ADP-ribose (ADPR) is transferred to an acceptor molecule, which can range from a polypeptide chain to an acyl group or a water molecule, or even the adenine of the ADPR moiety molecule itself in a cyclization reaction (Fig. 1). These reactions comprise the generation of second messenger molecules by the activity of NAD glycohydrolases (NADases) as well as post-translational protein modifications including NAD+-dependent protein deacetylation and mono- and poly-ADP-ribosylation of acceptor proteins.
Although the different types of NAD-dependent signalling reactions do not seem to be present throughout the cell, all of them have been discovered in mitochondria. As outlined below, some of the protein modifications in particular have been intensely studied. For example, protein deacetylation appears to be involved in the regulation of all key metabolic pathways within mitochondria. Therefore, it is not surprising that NAD-dependent signalling pathways are important to maintain mitochondrial function and may become promising targets for the treatment of metabolic diseases.
Because NAD-dependent signalling conversions include cleavage of the nicotinamide moiety, only NAD+, but not NADH, is used as a substrate. In light of the rather low mitochondrial redox ratio, several of the enzymes that catalyse NAD+-dependent signalling reactions have therefore been speculated to be redox sensors detecting the metabolic state of the cell [7, 8].
In this review, we provide an overview over the current knowledge about NAD-dependent signalling in mitochondria and highlight its involvement in processes that are essential for proper mitochondrial function.
NAD+-dependent protein deacetylation – global regulation of mitochondrial metabolism
All major mitochondrial processes, such as Krebs cycle, fatty acid metabolism, antioxidant response, oxidative phosphorylation and amino acid catabolism, are regulated by N-ε-lysine acetylation [9-16]. Acetylated lysine residues have been detected in many mitochondrial enzymes and their NAD-dependent deacetylation has been recognized as a key mechanism to control their activities. However, despite its obvious importance, establishment of acetylation marks within mitochondria has not been thoroughly studied. General control of amino acid synthesis protein 5-like 1 (GCN5L1, also known as BLOC1S1), a homologue of the nuclear acetyltransferase GCN5, was suggested to be an indispensable component of the mitochondrial protein acetylation machinery . Knockdown of GCN5L1 was shown to reduce lysine acetylation of mitochondrial fractions, yet this enzyme can account for this modification only in part. Therefore, other acetylation mechanisms need to be considered. For example, earlier reports suggested that nonenzymatic modification might be possible in the presence of high acetyl-CoA concentrations, which may be found in mitochondria [18, 19].
In general, acetylation of mitochondrial enzymes is associated with inhibition of their function, hence, deacetylation serves to activate mitochondrial processes [7, 12, 20]. Deacetylation within mitochondria is primarily carried out by NAD+-dependent deacetylases of the sirtuin family, three of which are located in the mitochondrial matrix (SIRT3–5) . The deacetylation mechanism involves cleavage and release of nicotinamide from NAD+, followed by the transfer of the acetyl moiety from acetylated lysine onto the 2″-position of ADPR, yielding 2″-O-acetyl-ADP-ribose (2″-OAADPR) [22, 23]. The latter can undergo a nonenzymatic transesterification to 3″-OAADPR and to a lesser extent 1″-OAADPR . To date, no function of the acetylated ADPR product in mitochondria has been identified.
It is noteworthy that the three mitochondrial sirtuins belong to different phylogenetic classes within the sirtuin family  and in the light of the extended understanding of their enzymatic properties, class-specific functions have been suggested [26, 27]. Whereas SIRT4 and SIRT5 exhibit mono-ADP-ribosyl transferase and deacylase activity, respectively, SIRT3 has emerged as the master regulator of the mitochondrial acetylome (Table 1) [28-31].
Table 1. Functions and targets of mitochondrial sirtuins
Mice knockout studies suggested that SIRT3 is the major mitochondrial protein deacetylase in vivo . Lombard et al.  showed that the absence of SIRT3 leads to protein hyperacetylation in mitochondrial fractions of liver and other tissues. This notion was confirmed by mitochondrial acetylome studies that identified more than 1000 SIRT3 targets [12, 16]. Functional association analyses of the identified proteins revealed that SIRT3 is involved in the regulation of all major mitochondrial processes, including maintenance of mtDNA integrity, protection from reactive oxygen species (ROS) and metabolic pathways . Based on this broad spectrum of targets, sirtuins were suggested as sensors for the redox state, to trigger adjustment of metabolic activities [7, 8, 32]. SIRT3 expression levels are elevated under conditions of calorie restriction or prolonged fasting and reduced in response to nutrient excess. The state of mitochondrial protein acetylation changes accordingly [33-36]. The importance of these changes in the acetylome has become more and more apparent over the last years, as critical functional consequences of protein acetylation/deacetylation in mitochondria have been documented.
For example, mitochondrial acetyl-CoA production is stimulated by SIRT3-mediated deacetylation, and hence catalytic activation of very-long-chain- and long-chain-specific acyl-CoA dehydrogenase as well as acetyl-CoA synthase 2 [33, 37-39]. Acetyl-CoA can then enter the Krebs cycle, which is further regulated by SIRT3-dependent deacetylation of succinate dehydrogenase [40, 41]. Alternatively, acetyl-CoA may be utilized for the formation of ketone bodies via SIRT3-activated hydroxymethylglutaryl-CoA synthase . Studies on mouse embryonic fibroblasts from SIRT3−/− mice revealed that electron transport chain complex I is also subject to acetylation-dependent regulation [43, 44]. Furthermore, the amino acid catabolism is stimulated by deacetylation of glutamate dehydrogenase (GDH) . Accumulation of ammonia is prevented by activation of the urea cycle via SIRT3-dependent ornithine carbamoyltransferase deacetylation [14, 46].
In addition to its role in metabolic regulation, SIRT3 can also counteract the accumulation of ROS. This is achieved by deacetylation of isocitrate dehydrogenase 2 and superoxide dismutase leading to the production of NADPH and direct superoxide degradation, respectively [47, 48]. Skeletal-muscle- and liver-specific SIRT3 knockout in mice showed that mild stress conditions neither altered ROS levels nor influenced metabolic homeostasis in comparison with control animals . These observations are in apparent contrast to the results obtained with germline knockout mice [33, 34, 43, 47, 50] and therefore pose questions regarding the mechanism of SIRT3 action, e.g. its tissue specificity, the interplay of multiple tissues or the presence of compensatory mechanisms.
Even though knowledge about SIRT3-dependent deacetylation has expanded rapidly, it has remained elusive why this enzyme has such a broad spectrum of targets. Future studies will have to establish whether SIRT3 activation always leads to a global rearrangement of the mitochondrial acetylome or if there are mechanisms that mediate a more discriminative action, e.g. by directing SIRT3 to specific targets.
SIRT5 – the mitochondrial de-acyl-ase
In contrast to SIRT3, the number of known targets for SIRT5 has remained rather limited. However, as described below, SIRT5 appears to preferentially remove acylation marks in proteins other than acetyl groups. Originally, SIRT5 was also suggested to be an NAD+-dependent deacetylase and the first function associated with its activity was the regulation of the urea cycle by deacetylation of carbamoylphosphate synthase 1 (CPS1) . More recently, SIRT5 was found to stimulate purine catabolism in mice via deacetylation of urate oxidase (UOX) . However, UOX is not expressed in hominoid primates including humans .
Closer examination of the catalytic properties of SIRT5 revealed a surprisingly weak deacetylation activity [29, 54]. However, this finding is in line with the lack of hyperacetylation in mitochondrial fractions from SIRT5 knockout mice . It turned out that SIRT5 exhibits rather de(acyl)ase activity with higher affinity for malonylated and succinylated targets [29, 31]. In fact, it was demonstrated that CPS1, the identified target of SIRT5, is succinylated in vivo and SIRT5 catalyses the removal of this modification . Analogous to the deacetylation reaction, the deacylation reaction releases O-malonyl-ADPR or O-succinyl-ADPR as products.
For the third mitochondrial member of the sirtuin family, SIRT4, no detectable deacetylation activity could be found [21, 55]. Although this may be due to a very specific substrate spectrum for the deacetylation reaction, another NAD+-dependent enzymatic activity, namely mono-ADP-ribosylation, was identified for SIRT4 [28, 30].
Mitochondrial protein ADP-ribosylation
Mono-ADP-ribosylation of proteins designates the transfer of a single ADPR moiety from NAD+ onto an acceptor amino acid residue. In poly-ADP-ribosylation, the protein-bound ADPR moiety serves as target for the attachment of further ADPR units resulting in long, branched polymers containing up to 200 ADPR units.
Mono-ADP-ribosylation in mitochondria has been reported both as enzymatic and nonenzymatic reaction. Enzymatic ADP-ribosylation was shown to preferentially modify cysteine residues in target proteins . GDH, a central enzyme of amino acid metabolism, was identified as the first, and so far only, acceptor protein , with cysteine 119 as the major acceptor site . Although GDH is a homohexamer, stochiometric analyses suggested that the modification of a single subunit is sufficient to potently inhibit its enzymatic activity . The ADP-ribosyl transferase mediating the modification was identified as SIRT4 [28, 30]. Thus, in addition to complex allosteric regulation, GDH activity is controlled by two members of the sirtuin family (SIRT4 and SIRT3). Physiologically, the SIRT4-dependent decrease in GDH activity was found to repress amino acid-stimulated insulin secretion  both in cell culture systems and mouse SIRT4 knockout models. The recent observation that GDH and SIRT4 antagonistically influence the growth of glial cells corroborates the importance of this regulatory cycle . ADP-ribosylation of GDH is reversible. A Mg2+-dependent ADP-ribosyl cysteine hydrolase activity was detected in mitochondria which removes the modification and restores GDH activity . However, the molecular identity of the enzyme has remained unknown.
Nonenzymatic ADP-ribosylation is chemically distinct from the enzymatic modification and may contribute to the accumulation of advanced glycation end products during aging . Moreover, nonenzymatic mono-ADP-ribosylation of mitochondrial proteins was found to be involved in calcium release from organelles. Treatment with pro-oxidants led to a shift in the mitochondrial redox ratio, the degradation of pyridine nucleotides, generation of ADPR, the modification of a ~ 30 kDa protein by ADP-ribosylation and release of Ca2+ . In a similar way, hormonal control of mitochondria was described. Triiodothyronine is a thyroid hormone that binds to nuclear receptors and influences gene transcription (reviewed in [62, 63]). However, it has long been appreciated that it also acts directly on mitochondria [64-66]. Evidence has been presented suggesting that the effect of triiodothyronine on mitochondrial metabolism may be regulated via mono-ADP-ribosylation of a ~ 30 kDa acceptor protein . The possibility to generate intramitochondrial ADPR, potentially enabling nonenzymatic ADP-ribosylation, was supported by the demonstration of a mitochondrial NAD+ glycohydrolase activity .
Poly-ADP-ribosylation has a prominent role in various nuclear processes. Nevertheless, several studies reported mitochondrial poly-ADPR formation or detection of poly-ADP-ribosylated proteins within the organelles [69-71]. However, this subject remains controversial (reviewed in [62, 63]). For example, poly-ADP-ribose polymerase 1 (PARP1), a protein with an undisputed nuclear localization, was partially localized to mitochondria [69, 74]. However, this observation was not confirmed by others . Interestingly, a small isoform of the poly-ADP-ribose (PAR)-degrading enzyme poly-ADP-ribose glycohydrolase (PARG) was proposed to be present in mitochondria [76, 77]. Detailed analyses revealed, however, that this isoform is catalytically inactive . By contrast, ADP-ribosylhydrolase 3 (ARH3) partially localizes to mitochondria  and exhibits weak PAR-cleaving activity [77,78]. However, its major activity appears to be hydrolysis of OAADPR to ADPR and acetic acid  (see below).
Mitochondrial NAD-derived calcium messengers – synthesis of cADPR and NAADP
Cyclic ADPR (cADPR) and nicotinic acid adenine dinucleotide phosphate (NAADP) are two potent intracellular calcium-mobilizing second messengers. Both molecules are generated by the same class of enzymes, NADases. The best characterized mammalian NADase is the plasma membrane protein CD38 . NAADP is generated by a base-exchange reaction that replaces the nicotinamide ring of NADP with nicotinic acid, whereas cADPR is generated by an intramolecular cyclization of ADPR between N1 of the adenine ring and the anomeric carbon of the terminal ribose following cleavage of the nicotinamide moiety (Fig. 1) [81, 82]. The established enzyme mechanism also accounts for the direct conversion of NAD+ to ADPR, as well as the cleavage of cADPR to ADPR. Mitochondria were repeatedly shown to possess NADase activity, implying that they are able to generate these calcium-mobilizing agents [83-88]. Accordingly, mitochondrial fractions were found capable of synthesizing both cADPR and NAADP [89, 90].
The identity of the mitochondrial NADase has not been established. Liang and colleagues  reported that the mitochondrial enzyme activity was distinct from CD38 both by western blot analyses and specific inhibition of the mitochondrial ADP-ribosyl cyclase activity, but not CD38, with Zn2+ ions. However, only two NADases have been found in mammals, CD38 and CD157. Both are plasma membrane proteins and therefore not readily suspected to play a role in intracellular processes. Nevertheless, CD38 has repeatedly been suggested to have intracellular functions . Identification of a bovine mitochondrial NADase, based on partial peptide sequences of a ~ 31 kDa protein, resulted in a perfect match with bovine CD38 protein, whose sequence was not available at the time of the study . However, plasma membrane CD38 has a molecular mass of ~ 45 kDa due to glycosylation of the protein [93, 94]. In fact, the amino acid sequence alone predicts a theoretical mass of ~ 31.5 kDa. The presence of CD38 in mitochondria has been further supported by immunoreactivity towards CD38 associated with mitochondria in rat brain samples  and in studies with CD38 knockout mice . Moreover, a truncated CD38 protein lacking the N-terminal tail and transmembrane domain showed predominant localization in mitochondria . The intriguing question remains: is the mitochondrial enzyme a nonglycosylated form of CD38?
Mitochondrial ADPR metabolism
Although the NAD-dependent signalling processes differ in their biological roles and molecular mechanisms, they share at least two general features. First, a common step in the modifying reactions is the release of nicotinamide from NAD. That is, all these reactions produce nicotinamide, which in turn may act as inhibitor, at least of sirtuins, NADases and PARPs [97, 98]. However, because nicotinamide is membrane permeable and readily recycled to nicotinamide mononucleotide (NMN) by nicotinamide phosphoribosyltransferase, it is not known whether the inhibition is relevant for endogenously produced nicotinamide. The second property shared is the generation of free ADPR as a result of ‘signal deactivation’ (Fig. 2). Protein mono-ADPR bonds, poly-ADPR, OAADPR and cADPR are cleaved by specific hydrolases to liberate ADPR [79, 99]. ADPR can directly influence mitochondrial metabolism, because it inhibits complex I of the electron transport chain . Consequently, proper disposal of this molecule is important to maintain mitochondrial function.
Evidence has been presented that ADPR can be released from mitochondria and potentially function as a second messenger [101, 102]. Indeed, cytosolic ADPR stimulates Ca2+ signalling by activating the transient receptor potential cation channel 2 (TRPM2) via its C-terminal NUDT9 homology domain (NUDT9-H) [103, 104]. The mechanism of ADPR release from mitochondria is not yet understood. It might involve opening of the mitochondrial permeability transition pore or a hitherto unidentified ADPR transporter .
Interestingly, OAADPR itself is also an activating ligand of the TRPM2 channel . Hydrolysis of OAADPR to ADPR within mitochondria is possible due to the enzymatic activity of ADP-ribosyl hydrolase 3 (ARH3) [77, 79, 106]. With the discovery of O-succinyl-ADPR and O-malonyl-ADPR as products of SIRT5 deacylation reactions  it is tempting to speculate that ARH3 might also hydrolyse these intermediates. Polymers of ADPR, when artificially generated in mitochondria, are also degraded by ARH3 . By contrast, ARH3 failed to hydrolyse any of the tested protein–ADPR conjugates at cysteine, arginine and aspartate residues . Given the uncertainty regarding endogenous PAR in the organelles and the rather low activity towards the polymers, the major physiological role of ARH3 in mitochondria appears to be the cleavage of OAADPR (or other acylated ADPR derivatives) generated by mitochondrial sirtuins.
In addition, the recently identified mitochondrial macro-domain-containing protein, MacroD1 (also known as LRP16)  exhibits OAADPR hydrolase activity . Macro domains are protein modules able to bind NAD metabolites such as OAADPR and ADPR and are found both in the nucleus and mitochondria .
ADPR is also subject to further conversion both within mitochondria and the cytosol (Fig. 2). It is cleaved by nucleoside diphosphate-linked moiety X (NUDIX) hydrolases. Members of the NUDIX family often exhibit substrate selectivity for specific nucleotide derivatives [110, 111]. NUDT9, a mitochondrial enzyme with preference for ADPR as substrate, produces ribose-5-phosphate and AMP [112-114]. In the cytosol, ADPR can also be degraded into ribose-5-phosphate and AMP by another NUDIX hydrolase, NUDT5 (Fig. 2) .
Establishment and maintenance of the mitochondrial NAD pool
The functionality of mitochondrial metabolism is highly dependent on the maintenance of the organellar NAD pool [116, 117]. Given the multitude of NAD-degrading signalling reactions, it can be expected that the turnover of this molecule is significant. Nevertheless, the pathway of mitochondrial NAD generation in mammalian cells is still not fully understood.
Although in Saccharomyces cerevisiae  and Arabidopsis thaliana  NAD is imported from the cytosol by specific mitochondrial NAD transporters, such a mechanism has not been identified in mammals. In fact, the closest human homologue of these transporters was unable to carry NAD , but rather transports folate  and FAD . Moreover, whereas the outer mitochondrial membrane is permeable for NAD, in vitro studies on isolated mammalian mitochondria suggested that the inner mitochondrial membrane was indeed impermeable for NAD .
A potential alternative to mitochondrial NAD uptake is intraorganellar biosynthesis of the molecule. Indeed, a mammalian isoform of NMN adenylyltransferase (NMNAT), NMNAT3, has been identified  and localized to the mitochondrial matrix . NMNATs catalyse the final step of NAD synthesis by forming the dinucleotide from NMN and ATP [124, 125]. Interestingly, neither S. cerevisiae nor A. thaliana have a mitochondrial NMNAT isoform. Consequently, in mammals, generation of mitochondrial NAD might rely on the uptake of NMN from the cytosol. However, a corresponding mitochondrial transporter has not been identified so far. Nevertheless, the ability of isolated rat liver mitochondria to generate NAD from added NMN supports such a possibility . Moreover, it has been recently reported that mitochondria from U937 and HeLa cells contain considerable amounts (1–2 nmol·mg−1 protein) of NMN . Taking further into account that none of the other known NAD-biosynthetic enzymes in mammals localizes to mitochondria [116, 123], it appears plausible that mitochondrial NAD is generated from imported NMN, even though direct NAD uptake from the cytosol cannot be fully ruled out.
Conclusion and outlook
The understanding of NAD-dependent processes within mammalian mitochondria has been greatly expanded over recent years revealing an unexpected complexity of regulatory networks that include and combine NAD-dependent post-translational modifications, second messenger generation and redox state. However, several important questions remain elusive. For example, how the mitochondrial NAD pool is established requires further investigation, especially with regard to the mechanisms of NMN or NAD transport into the organelles. Moreover, the existence of mitochondrial poly-ADPR metabolism is still debated and requires clarification. The potential signalling functions, metabolism and export of NAD-derived intermediates also need further investigation. Finally, exploring the possible roles of global acetylation/deacetylation in mitochondria as compared to target-specific regulation of mitochondrial activities by acylation will be a demanding task for the years to come.
We gratefully acknowledge support from the Norwegian Cancer society (Kreftforeningen) and the Norwegian Research Council.