Over 100 years after its first discovery, several new aspects of the biology of the redox co-factor NAD are rapidly emerging. NAD, as well as its precursors, its derivatives, and its metabolic enzymes, have been recently shown to play a determinant role in a variety of biological functions, from the classical role in oxidative phosphorylation and redox reactions to a role in regulation of gene transcription, lifespan and cell death, from a role in neurotransmission to a role in axon degeneration, and from a function in regulation of glucose homeostasis to that of control of circadian rhythm. It is also becoming clear that this variety of specialized functions is regulated by the fine subcellular localization of NAD, its related nucleotides and its metabolic enzymatic machinery. Here we describe the known NAD biosynthetic and catabolic pathways, and review evidence supporting a specialized role for NAD metabolism in a subcellular compartment-dependent manner.
The central role of nicotinamide adenine dinucleotide (NAD) as co-enzyme in redox reactions has been known since the 1930s, when Warburg demonstrated the capability of the originally called ‘cozymase’ to act as a hydride-accepting and hydride-donating molecule . Both the balance of cellular redox potential and the synthesis of ATP rely on these hydrogen transfer properties of NAD. Only several decades after elucidation of its redox properties were NAD non-redox reactions, i.e. protein post-translational modifications such as deacetylation by sirtuins  and mono- and poly(ADP-ribosyl)ation by mono(ADP-ribosyl)transferases (ARTs) and poly(ADP-ribosyl)polymerases (PARPs) [3, 4] discovered. NAD also participates in Ca2+ signalling through its derivative nicotinic acid adenine dinucleotide phosphate (NAADP) and its degradation products ADP-ribose (ADPR) and cyclic ADP-ribose (cADPR) .
NAD non-redox reactions are involved in various biological processes, including cell differentiation and survival, apoptosis and chromatin stability. The variety of biological effects derived from these reactions illustrates the fundamental role of NAD as a signalling molecule and the importance of NAD homeostasis in cellular metabolism.
Signalling mechanisms downstream of NAD synthesis have been described in individual cell compartments , and the biological functions that these signals elicit within a particular compartment, as well as the role that NAD biosynthetic enzymes play at those levels, are starting to be elucidated. This review illustrates the main NAD metabolic reactions and the diversification of NAD biological roles that is achieved by targeting these reactions to specific subcellular locations.
NAD metabolic reactions
From NAD precursors to mononucleotides
When used as a co-enzyme, NAD is repeatedly converted between its oxidized form (NAD+) and its reduced form (NADH), thus its levels remain constant. By contrast, in non-redox reactions, the NAD molecule is ‘split’, and consequently its concentration decreases. Therefore, to prevent depletion of the cellular pool of NAD, continuous synthesis of the dinucleotide is required.
Depending on the organism, four building blocks and four corresponding pathways may be used for biosynthesis of NAD: quinolinic acid (QA) in the de novo pathway and nicotinamide (NAM), nicotinic acid (NA) and nicotinamide riboside (NR) in the salvage pathways [7, 8]. QA, NA and NAM are used by three distinct phosphoribosyltransferases [QA phosphoribosyltransferase (QAPRT), NA phosphoribosyltransferase (NAPRT) and NAM phosphoribosyltransferase (NAMPT)] for production of the respective mononucleotides, i.e. NA mononucleotide (NAMN) from QA and NA, and NAM mononucleotide (NMN) from NAM. NAMN and NMN are subsequently converted to NA adenine dinucleotide (NAAD) and NAD, respectively, by NMN adenylyltransferase (NMNAT). NAAD is converted to NAD by NAD synthetase (NADS) (Fig. 1).
The tissue expression pattern of the NAD biosynthetic enzymes, mainly of the enzymes catalysing the rate-limiting formation of NMN and NAMN [9, 10], may mirror specific NAD requirements of different cell types and tissues. For example, hepatocytes are highly active NAD-synthesizing cells, and QAPRT, NAPRT and NAMPT are all expressed in the liver , suggesting that both the de novo and salvage pathways are used for NAD synthesis in this tissue . NAPRT is expressed in tissues that preferentially use NA for NAD synthesis, such as heart, kidney and red blood cells [12-14]. NAMPT activity has been detected in all analysed tissues , where its expression may be influenced by extracellular stimuli such as nutrient restriction availability [15, 16].
In mammalian cells, the NAM salvage pathway is the most relevant, while the de novo pathway appears to have a marginal role. This observation is supported by several lines of evidence. First, NAM is a by-product of NAD non-redox reactions (Fig. 2). Second, NAM is the most abundant NAD precursor in the bloodstream . Third, NAMPT is expressed in all mammalian tissues , and Nampt gene deletion is embryonically lethal ; instead, QAPRT is found mainly in liver, kidney, brain , thyroid  and blood cells , and Qaprt inactivation does not affect organism development . Fourth, tryptophan alone, which is converted to QA through the kynurenine pathway, is not sufficient to maintain the physiological NAD concentration of the cell ; in extra-hepatic tissues, tryptophan may be diverted from its original role as a precursor for NAD biosynthesis to a novel immuno-related role , arguing against the redundancy of the de novo and salvage pathways. Finally, vertebrates lack the enzyme nicotinamidase, which converts NAM to NA (expression of nicotinamidase is mutually exclusive with expression of NAMPT ), making the NAM salvage pathway the only possible NAM recycling pathway.
The catalytic activity of NAMPT is enhanced by ATP and regulated by a feedback mechanism: it is inhibited when NAD levels have reached 90% of their normal physiological concentration [12, 23]. In contrast, NAPRT activity is not inhibited by NAD [12, 24], and NA is a more effective precursor than NAM for NAD synthesis in some tissues [12, 25, 26]. Therefore, NA is also relevant for NAD biosynthesis, probably playing a distinct and complementary role to NAM.
In addition to QA, NAM and NA, NR has recently been identified as an essential precursor for NAD biosynthesis , suggesting a further diversification of tissue-specific NAD synthesis. NR is phosphorylated to NMN by the enzyme nicotinamide riboside kinase (NRK) (Fig. 1). In humans, two NRK isoforms, NRK1 and NRK2, have been identified and characterized, and their ability to use both NR and NA riboside as substrates led some authors to propose that NR and NA riboside are NAD precursor vitamins as well as NA and NAM . NRK1 is ubiquitously expressed, whereas NRK2 is expressed in the heart, skeletal muscle, brown adipose tissue and liver, in keeping with the higher ability of these tissues to increase NAD levels after NR administration .
The biosynthetic pathways from all four NAD precursors converge at the level of dinucleotide formation, catalysed by the enzyme NMNAT, which is able to use both NMN and NAMN with comparable efficiency .
From mononucleotides to dinucleotides
The enzyme NMNAT catalyses the nucleophilic attack by the 5′-phosphate of NMN (or NAMN) on the α-phosphoryl of ATP, yielding NAD (or NAAD) and pyrophosphate (PPi)  (Fig. 1). Three human NMNAT isoforms, differing in their oligomeric state, subcellular localization and catalytic properties, have been identified and characterized .
Human NMNAT1 is a 279 residue nuclear protein (31.9 kDa) containing two conserved ATP-recognition motifs: GXXXPX(T/H)XXH and SX(T/S)XXR [32, 33]. Consistent with its essential role in cellular metabolism, NMNAT1 is expressed in all tissues . Human NMNAT2 consists of 307 amino acids (34.4 kDa) and shares 34% sequence identity with human NMNAT1 [33, 35]. NMNAT2 is enriched at the surface of the Golgi apparatus and is particularly abundant in the heart, skeletal muscle and brain [31, 35-38]. Two functional cAMP-response elements (CRE) have been identified in the mouse Nmnat2 promoter region, suggesting that NMNAT2 is a direct target of the transcription factor CRE-binding protein (CREB) under physiological conditions . Human NMNAT3 is a 252 residue mitochondrial protein (28.3 kDa) with 50% and 34% sequence identity to human NMNAT1 and NMNAT2, respectively. NMNAT3 is present in red blood cells , lung, spleen, and, to a lesser extent, the placenta and kidneys .
The 3D structures of both human NMNAT1 and NMNAT3 have been solved in their apo form and in complex with substrates and products [33, 41-43], while the structure of NMNAT2 has yet to be determined . These enzymes belong to the ‘nucleotidyltransferase α/β phosphodiesterase’ family, and comparative analysis of monomer structure has revealed a common Rossman-like fold with a six-strand β-sheet. While NMNAT1 forms a globular barrel-like hexamer, NMNAT3 forms a tetramer and NMNAT2 behaves as a monomer . A solvent channel crosses both NMNAT1 and NMNAT3 oligomers, and the active site is located in a wide cleft facing the channel [33, 41]. The molecular determinants that enable mammalian NMNAT isoforms to participate in both the de novo and salvage pathways (i.e. to accept both NAMN and NMN as substrate) have been elucidated , as well as the structural determinants of their subcellular localization. The nuclear localization signal of NMNAT1 [PGRKRKW(123–129)] lies in the outer surface of the channel, and is able to interact with the nuclear transporting proteins . Cys164 and Cys165 of NMNAT2 form a distinctive structural pair within the NMNAT family  that is responsible for anchoring the protein to the Golgi membrane via palmitoylation [38, 46], whereas an N-terminal targeting sequence is responsible for mitochondrial localization of NMNAT3 . The rigorous subcellular localization of mammalian NMNAT isoforms suggests a predominant role for this enzyme in determining the subcellular NAD pool distribution.
NAD is used in non-redox reactions such as protein (ADP-ribosyl)ation and protein deacetylation (Fig. 2), in which the NAD glycosidic bond linking NAM to ribose is cleaved. In these reactions, the resulting ADPR is bound to an acceptor protein (ADP-ribosylation) or to an acetyl group (deacetylation), and the released NAM may be used to re-synthesize NAD.
During protein mono(ADP-ribosyl)ation, the ADPR from NAD is transferred to a specific amino acid of the acceptor protein by ART enzymes (Fig. 2A). This reversible protein modification modulates critical cellular functions such as muscle cell development, actin polymerization and cytotoxic T-lymphocyte proliferation [47, 48]. Poly(ADP-ribosyl)ation involves transfer of several ADPR molecules to an acceptor protein by PARP enzymes, leading to formation of branched polymers (Fig. 2B) . In mammals, the PARP family has 17 members (including 10 putative members), mainly involved in the expression of inflammatory genes, DNA repair and modification, telomere length regulation, mitosis and apoptosis . Poly(ADP-ribose) is a structurally complex macromolecule whose mass may exceed that of the protein acceptor; therefore, in poly(ADP-ribosyl)ation, the non-covalent interactions of anchored poly(ADP-ribose) with other macromolecules may become preponderant .
During protein deacetylation, the ADPR derived from NAD is bound to the acetyl group of a lysine of the target protein, generating O-acetyl-ADP-ribose and NAM and deacetylating the protein (Fig. 2C). This reaction is catalysed by sirtuins, a class of proteins that have either histone deacetylase or mono(ADP-ribosyl)transferase activity . Sirtuins are involved in several cellular processes, including transcriptional silencing, DNA repair, DNA recombination and lifespan regulation [53, 54]. In mammals, seven sirtuins (SIRT1–SIRT7) have been identified [55, 56], whose subcellular localization parallels that of NMNAT isoforms. The sirtuin product O-acetyl-ADP-ribose has also been implicated as a second messenger .
Another important family of ‘NAD-consuming’ enzymes are ADP-ribosyl cyclases, also known as NAD glycohydrolases, which generate Ca2+-mobilizing compounds (cADPR, ADPR and NAADP) from NAD and nicotinamide adenine dinucleotide phosphate (NADP) .
Extracellular NAD and its biological role
The NAD plasma concentration is of the order of 10–50 nm, even up to 100 000 times lower than its intracellular levels [59, 60]. This very low extracellular concentration is maintained in three ways: (a) NAD is exported from the cell in minimal amounts rather than being directly synthesized, (b) NAD is rapidly metabolized and its products are also biologically active, and (c) NAD interacts directly with receptors and then is catabolized rapidly to inactivate its action.
Several biological functions mediated by extracellular NAD or its metabolites have been described [61-63]. For example, NAD is a ligand for various subtypes of purinergic P2 receptors. In human mesenchymal stem cells, NAD, which in whole cells can be released extracellularly via the connexin 43 channels , binds to the purinergic receptor P2Y11 causing opening of L-type Ca2+ channels and activation of a cAMP/cADPR/[Ca2+]i signalling cascade, finally causing activation of cellular responses such as proliferation and migration . In T cells and monocytes, activation of P2X7 receptors generally results in internalization of Ca2+ via opening of a non-selective, large membrane pore, causing apoptotic cell death [66-68]. Extracellular NAD also acts as a neurotransmitter released by terminals of mammalian central nervous system and peripheral nervous system neurons upon stimulation, and subsequently binding to post-synaptic P2Y1 receptors, similar to ATP [69, 70].
NAD catabolic enzymes are present on the surface of the cell [71, 72], suggesting that NAD derivatives may mediate cellular responses in the extracellular environment (Fig. 3). Extracellular NAD is degraded by three main groups of specific ectoenzymes: ADP ribosyl cyclases/NAD glycohydrolases (CD38 and CD157 [73, 74]), ARTs [75, 76] and ectonucleotidases, including ectonucleotide pyrophosphatase/phosphodiesterases and ecto-5′-nucleotidase (ENPP, CD73 [72,77,77a]). ADPR, the product of NAD glycohydrolases and ARTs, is subsequently converted to adenosine and may bind to P2Y1 receptors, eliciting signals through elevation of cAMP levels . The catabolic products of NAD may also regulate insulin receptor signalling, as over-expression of plasma cell membrane NAD catabolic enzyme glycoprotein-1/ectonucleotide pyrophosphatases/phosphodiesterase-1 inhibits insulin signalling in an enzymatic activity-dependent manner [79, 80].
In addition to a direct effect of NAD and its metabolites, the enzymes involved in synthesis and degradation of NAD also have important extracellular functions. For example, ARTs post-transcriptionally modify and activate P2 receptors . Another example is extracellular NAMPT (eNAMPT), a highly active protein characterized as a soluble factor that is up-regulated upon activation in lymphoid cells and during stimulation of immune cells by various stimuli .
eNAMPT has a variety of biological functions: (a) it acts as a cytokine that modulates the immune response, (b) it has anti-apoptotic effects on immune cells, including neutrophils and macrophages, and (c) it plays a critical role in the regulation of glucose-stimulated insulin secretion in pancreatic β cells . Originally identified as a cytokine acting as a pre-B-cell colony-enhancing factor , its identity as an NAM phosphoribosyltransferase was later recognized . The nature of eNAMPT as a cytokine that modulates inflammatory processes is complex: it is synthesized and released in response to inflammatory stimuli, a process that requires efficient translation of the intracellular protein . There is a controversy about whether the immune effects of eNAMPT depend on its enzyme activity, i.e. the synthesis of NMN. eNAMPT promotes macrophage survival after induction of endoplasmic reticulum (ER) stress, exerting pro-inflammatory functions, an effect that is mediated by secretion of interleukin-6 and phosphorylation of signal transducer and activator of transcription 3 (STAT3). This effect cannot be reproduced by exogenous NMN . On the other hand, eNAMPT induces inducible nitric oxide synthase (iNOS) and activates extracellular signal-regulated kinases 1/2 (ERK1/2) and NF-κB pathways in human vascular smooth muscle cells, and this action is mimicked by NMN . eNAMPT enhances the expression of extracellular matrix metalloproteinase inducer, activating matrix metalloproteinase-9 in macrophages by a mechanism involving NF-κB and mitogen-activated protein kinase signalling, all effects reproduced by NMN . eNAMPT is up-regulated in obesity: it is enriched in visceral fat, in which it is secreted by adipocytes and acts as an adipokine, and it has therefore been named visfatin . While it was initially claimed to bind and activate insulin receptors, it was later shown that eNAMPT's insulino-mimetic properties are more likely related to its enzymatic activity and the extracellular formation of NMN. This nucleotide is transported, probably via previous conversion to NR [ and ], inside pancreatic β cells where it is converted to NAD, which is used as substrate for sirtuins that regulate gene expression and stimulate insulin secretion [18, 83, 90]. In line with this observation, systemic administration of NMN to aged mice or mice subjected to a high-fat diet restores normal NAD levels in white adipose tissue and liver, and ameliorates glucose intolerance and type II diabetic syndrome [91, 92]. eNAMPT also exerts angiogenic effects that are mediated by induction of monocyte chemoattractant protein-1 and its receptor .
Given its central homeostatic role, it is not surprising that eNAMPT has attracted interest as a target for therapeutic treatments of metabolic and immunological disorders. Although the future of eNAMPT as a target for drug development is promising, it was recently reported that this enzyme is unable to synthesize NMN in the plasma due to the limited availability of ATP and phosphoribosylpyrophosphate (PRPP), both of which are required for NAMPT activity. This casts some doubt as to its exact modus operandi .
In contrast to eNAMPT, no association of NMNAT isoforms with the extracellular environment has been described to date. This raises some intriguing questions about the physiological significance of putative NMN synthesis outside the cell. For example, if NMN synthesis is not followed by conversion to NAD, what is the ‘economical’ reason for the presence of an energy-consuming NAD-export process rather than local synthesis?
The intracellular environment
Biological functions of NAD in the nucleus
By controlling the enzymatic activity of the two nuclear enzymes SIRT1 and PARP1 (Fig. 3), nuclear NAD regulates cellular longevity as well as gene transcription and circadian rhythm. The presence of a dedicated nuclear enzyme for NAD biosynthesis, NMNAT1, and of protein–protein interactions between NAD metabolic enzymes [95-97], suggests that a localized NAD pool must be maintained in close proximity to its site of action. In agreement with the notion that nuclear NMNAT1 has an essential role in cellular metabolism, its deletion is incompatible with organism survival . NMNAT1 loss-of-function mutations cause retinal neuron degeneration, indicating a direct link between this nuclear enzyme and neuron survival [99-102]. Interestingly, the absence of SIRT1 is also characterized by retinal defects .
NAD and PARP1
Nuclear levels of NAD control the activity of PARPs. Among the 17 PARP members, nuclear PARP1 is the best characterized . PARP1 is activated in response to genotoxic stress, oxidative stress and DNA breakage, leading to DNA repair by poly(ADP-ribosyl)ation of target proteins. Although generally this has a beneficial outcome, DNA repair in transformed cells may favour abnormal proliferation of genetically aberrant cells. Based on this observation, inhibitors of PARP1 activity are being tested for treatment of certain types of cancer . PARP1 causes depletion of its nuclear substrate NAD when hyper-activated, and this loss of NAD is associated with cell death [105, 106]. PARP1 hyper-activation also influences the nuclear translocation of substrates such as p65 NF-κB, as their poly(ADP-ribosyl)ated forms become sequestered in the nucleus, where they cause abnormal gene transcription and ultimately cell death [107, 108]. PARP1 hyper-activation promotes translocation of apoptosis-inducing factor from the mitochondrion to the nucleus, an event that is required for PARP1-induced apoptosis  and that follows NAD depletion . Therefore, to avoid abnormal cell proliferation, but also to inhibit excessive apoptosis, PARP1 activity must be carefully balanced. This balance is influenced by NMNAT1, which regulates PARP1 activity both via controlling NAD nuclear levels and by phosphorylation-dependent direct interaction [96, 97].
NAD and SIRT1
Nuclear NAD is also used as a substrate for the activity of the three nuclear sirtuins (SIRT1, 6 and 7), among which SIRT1 is the best characterized and studied . Nuclear sirtuins control the expression of a number of genes by deacetylating histones. In addition, they physically bind and deacetylate various other substrates, mainly transcription factors, directly controlling their activity. The yeast homologue of human sirtuins, Sir2 (silencing information regulator protein 2), the prototypical enzyme of this group, regulates the replicative lifespan in yeast mother cells [53, 54, 112]. Importantly, this property is conserved in higher organisms, although recent reports using Caenorhabditis elegans and Drosophila melanogaster have highlighted the influence of the genetic background . In mammals, SIRT1 does not directly increase longevity, but its over-expression reduces the adverse effects of a high-fat diet or ageing on glucose metabolism, and correlates with a reduced incidence of metabolic diseases, therefore contributing to healthy ageing [54, 114]. SIRT1 plays a crucial role in cellular homeostasis, and Sirt1 deletion is not compatible with organism survival . However, in an outbred genetic background, Sirt1 deletion has less severe consequences but may affect the immune system and abolish the beneficial effects of caloric restriction [115, 116]. SIRT1 has beneficial effects on several types of cancer, although it may act as both an oncogenic factor and a tumour suppressor . The protective effect of SIRT1 extends to cell and mouse models of neurodegeneration: SIRT1 over-expression reduces cell death in models of Alzheimer's disease and amyotrophic lateral sclerosis, an effect that is mimicked by the Sirtuin activator resveratrol  and is due to a reduction in tau acetylation . Increasing NAD synthesis by over-expressing NAMPT has similar effects, supporting the concept that SIRT1 activity is regulated by the levels of its substrate NAD [10, 83]. Indeed, NAD, its biosynthetic machinery and SIRT1 are tightly correlated. Nuclear NAD levels control SIRT1 activity in a variety of cells, for example vascular smooth cells , skeletal myoblasts  and neurons . The gene expression profiles of NAMPT- and SIRT1-over-expressing cells are correlated , and knockdown of NAMPT or NMNAT1 has similar effects to SIRT1 knockdown on the expression profile of co-regulated genes . Interestingly, SIRT1 activity also requires that the nuclear levels of the NAD precursor NAM, which acts as a potent SIRT1 inhibitor, remain low; therefore, it has been proposed that some positive effects of NAMPT on SIRT1 activity are mediated through NAM removal, rather than NAD synthesis . For example, glucose restriction elevates NAMPT activity by transcriptional regulation, and the subsequent increase in NAD levels and decrease in NAM levels promote SIRT1 activity, mediating the cellular response to nutrient availability .
NAD and circadian rhythm
Recent studies have demonstrated a role for NAMPT in the regulation of circadian rhythm. The transcription factors circadian locomotor output cycles kaput (CLOCK) and brain and muscle aryl hydrocarbon receptor nuclear translocator-like (BMAL1) form heterodimers that bind to the promoters of the circadian genes encoding Period 2 and cryptochrome, and activate their expression. When their levels increase, Period 2 and cryptochrome bind to the CLOCK/BMAL1 complex and inactivate it, therefore completing a feedback loop to regulate their own expression . In turn, NAD-dependent SIRT1 binds to CLOCK and terminates the activity of the CLOCK/BMAL1 heterodimer via deacetylation , decreasing the expression of circadian genes. NAD, NAMPT and SIRT1 levels follow a circadian cycle, and are regulated by the core clock machinery. CLOCK binds to and up-regulates Nampt, thereby providing NAD for SIRT1, which in turn represses CLOCK/BMAL1, causing a decrease in Nampt transcription, a decrease in NAD availability and consequently a reduction of CLOCK/BMAL1 inhibition [125, 126]. Increasing systemic and cellular NAD levels by down-regulating the NAD catabolic enzymes PARP1 or CD38 disrupts NAD oscillations and desynchronizes the circadian clock [127, 128]. In agreement with the regulation of NAD levels in response to dietary changes, circadian rhythm genes and pattern are modified by the availability of nutrients . As this function is highly dependent on local NAD levels, it appears likely that the localization and expression of NMNAT1 are also tightly controlled, contributing to the regulation of the circadian rhythm loop.
NAD metabolism in mitochondria
In contrast to yeasts and plants [130, 131], cytoplasmic NAD cannot be transported across the mitochondrial membrane in mammalian cells, which therefore use cytoplasmic NMN to generate the mitochondrial NAD pool . Once NMN has entered the mitochondria, synthesis of NAD may only be accomplished by the mitochondrial NMNAT isoform, i.e. NMNAT3 (Fig. 3). An alternative, mitochondrial-specific NAD salvage pathway involving a mitochondrial NAMPT has also been suggested, but the presence of NAMPT within the mitochondrion is controversial [15, 132, 133].
Some reports suggest that mitochondrial NAD is the largest subcellular pool of NAD , particularly in tissues that require a high energy level. This is in agreement with the pivotal role of NAD as an electron donor during oxidative phosphorylation and ATP biosynthesis, as well as its redox functions in metabolic processes including the Krebs cycle and β-oxidation. Interestingly, the NAD concentration in mammalian mitochondria is up-regulated by nutrient restriction and is independent of the total cell NAD levels .
In addition to its role in mitochondrial redox mechanisms, numerous studies have demonstrated the importance of NAD in non-redox reactions that are also fundamental in regulating mitochondrial function. Mitochondrial NAD regulates the activity of SIRT3, SIRT4 and SIRT5, which act as metabolic sensors of the energy status of the cell [134-138] by modulating the activity of several mitochondrial enzymes, including carbamoyl phosphate synthetase , acetyl CoA synthetase  and glutamate dehydrogenase . SIRT3 is the major mitochondrial deacetylating protein, and its loss, with the concomitant dysregulation of mitochondrial protein acetylation, contributes to the metabolic syndrome . Protein mono-(ADP-ribosyl)ation in mitochondria has been demonstrated. For example, glutamate dehydrogenase is subjected to mono(ADP-ribosyl)ation . Poly-(ADP-ribosyl)ation of mitochondrial DNA and proteins has also been observed: it is related to mitochondrial DNA repair and is involved in both oxidative stress and brain injury [143-146].
The effect of NAD in mitochondria is reproduced by its precursors. For example, NR improves mitochondrial function via SIRT3 activation  and protects against metabolic damage and neurodegenerative diseases .
Cytoplasmic NAD and signalling
Although it is still debated whether NMNAT1-synthesized NAD may be exchanged between the nucleus and cytoplasm via nuclear pores, it is known that NAD synthesis is independently regulated in these compartments and the cytoplasmic NAD pool is maintained primarily by NMNAT2 (Fig. 3). As NMNAT3 is strictly confined to the mitochondria, it is unlikely that it contributes to cytoplasmic NAD production.
In addition to its role as a redox co-factor, NAD has an important function in the cytoplasm as a signalling molecule. To perform this function, NAD is converted to cADPR and NAADP by ADP-ribosyl cyclases [148-150]. cADPR and NAADP bind to specific receptors on the surface of the endoplasmic reticulum and stimulate the release of Ca2+, causing an overall increase in intracellular Ca2+ concentration. As NMNAT2 is enriched at the surface of the Golgi apparatus [31, 46], the close communication between the Golgi and the endoplasmic reticulum confines NAD synthesis and its use as a precursor of Ca2+-mobilizing second messengers both spatially and temporally. The exact identity of the cytoplasmic ADP-ribosyl cyclase responsible for converting NAD into cADPR and NAAD remains elusive; however, one candidate for this role is CD38. Originally described as an extracellular NADase, CD38 is also present in specific intracellular compartments, most notably associated with the endoplasmic reticulum, the nuclear membrane and the surface of the mitochondria . This suggests that, as for its synthesis, degradation of NAD by CD38 occurs in specific locations [31, 46].
As in the other cell compartments, NAD acts as a substrate for protein deacetylation and (ADP-ribosyl)ation in the cytoplasm. For example, NAD is used by SIRT2, which deacetylates and stabilizes microtubules , and by two PARP family members, PARP4 and tankyrase . Initially described as a regulator of telomere function , tankyrase is enriched at the Golgi apparatus, where it co-localizes with glucose transporter type 4 vesicles and influences the dynamics of the organelle .
NAD synthesis and axon degeneration
The NAD-synthesizing enzyme NMNAT2 has acquired a specialized function in the axoplasm (Fig. 4), where it maintains axon integrity under normal conditions. NMNAT2 is constantly delivered to the axon by fast axonal anterograde transport [155, 156], but has a short half-life; therefore, when axons are injured and new protein can no longer be delivered locally, NMNAT2 levels fall below a critical threshold, triggering axon degeneration . This process, known as ‘Wallerian degeneration’ after Augustus Waller who was the first to describe it , is characterized by mitochondrial swelling and axonal cytoskeleton degradation. The role of the NAD biosynthetic pathway in Wallerian degeneration was first revealed by studies of the spontaneous mutant mouse Wallerian degeneration slow (WldS), which is characterized by a tenfold delay in Wallerian degeneration . The WldS spontaneous mutation causes generation of a fusion protein in which full-length NMNAT1 is joined to an N-terminal sequence that relocates the nuclear protein to the axon [159-161]. As NMNAT1 is a stable isoform, it may compensate for loss of NMNAT2 after an injury when it is relocated to the axon, maintaining axonal integrity [161-163]. Down-regulation of NMNAT2 causes degeneration of uninjured axons, but has a much weaker effect at the cell-body level, consistent with its predominant axonal location [155, 156]. Axon integrity after an injury is also maintained when NMNAT2 degradation is reduced [155, 156, 164, 165]. The essential role of NMNAT2 in axon maintenance has been recently demonstrated in vivo: Nmnat2 gene deletion causes embryonic or perinatal death due to lack of peripheral innervation . Nuclear NMNAT1 does not affect axon degeneration even if over-expressed, in keeping with its tight subcellular localization [98, 167].
As NMNAT enzyme activity is required for axon protection , synthesis and maintenance of high NAD concentrations has been considered crucial to axon integrity, probably by keeping levels of ATP high [168, 169]. Indeed, NAD, its precursors and over-expression of its biosynthetic enzymes have all been shown to confer neuroprotection [168-170]. However, more recent work, despite confirming the requirement for NMNAT enzyme activity to induce axon protection, failed to demonstrate its NAD dependence [161, 167, 171], indicating an alternative role for NMNAT enzymatic activity in the axon.
In addition to its role in the axon, NMNAT2 is also involved in protection in neurodegeneration models. For example, Nmnat2 gene transcription is decreased in mouse models of taoupathies, while over-expression of NMNAT2 ameliorates the pathology .
Interplay between NAD synthesis and degradation at specific compartments
Mammalian cells have evolved various systems to regulate NAD supply for NAD-dependent reactions. First, they express enzymes that catalyse the formation of NMN and NAMN in a tissue-specific pattern; second, they regulate NAMPT activity through a feedback mechanism that depends on NAD levels (and also on ATP levels); third, they express various NMNAT isoforms that are specifically localized to particular cellular compartments, thus enabling the cell to maintain highly distinct functional pools of NAD in different organelles.
The intracellular localization of NAD within the cell strongly reflects the spatial distribution of NAD-dependent processes . Indeed, in specific organelles, local NAD production may be strictly modulated by recruitment of NAD-biosynthetic enzymes to sites of NAD-consuming reactions [6, 31, 172]. For example, as mentioned above, a complex interplay of protein–protein interactions between SIRT1, PARP1 and the NAD synthetic enzymes NAMPT and NMNAT1 at particular locations within the nucleus dictates NAD-mediated control of gene transcription, circadian rhythm and cell death [95, 96, 125, 126, 173]. The intracellular NAD/NADH level, which reflects the metabolic status of the cell , increases in low energetic states via up-regulation of NAMPT, causing an increase in nuclear SIRT1 and mitochondrial SIRT3 activities, regulating energy metabolism and decreasing oxidative damage [15, 174]. The activity of NRK also positively influences sirtuins, as NR treatment increases SIRT1 and SIRT3 activities by enhancing NAD bioavailability in the nuclear and mitochondrial compartments, respectively, without affecting cytoplasmic NAD levels .
As NAD-consuming reactions catalysed by PARP and sirtuins compete for the same substrate, cross-talk of these reactions through modulation of NAD levels, which may have pathophysiological implications, has been proposed [55, 175, 176]. This is in agreement with the reduction of sirtuin activity in myocytes caused by NAD depletion, which results from PARP1 activation . Conversely, reducing the activity of NAD-consuming enzymes such as PARP1 and CD38 leads to an increase in the availability of NAD and thus an enhancement of SIRT1 activity and mitochondrial metabolism [178, 179].
The studies mentioned here and many others have contributed to our current understanding of NAD functions in the cell. The last 20 years of research have revealed that NAD, its biosynthetic enzymes and its related nucleotides achieve a high degree of specialization and diversification of their biological role through compartmentalization. Despite the advances in our understanding of compartmentalized NAD function, some crucial questions remain unanswered. For example, what are the minimum quantities of NAD needed for cell survival? Do NAD precursors have specialized biological roles? Are NAD compartmentalization and pathological conditions linked? Understanding the specific roles of NAD within its varied subcellular locations and subsequently manipulating NAD levels at specific compartments may offer a novel approach for treatment of various diseases.
We apologise to all the authors whose work could not be cited due to space restrictions. We are grateful to Nadia Raffaelli, Universita' Politecnica delle Marche, Ancona and Sally Wheatley for critically reading the manuscript. This work was supported by a Non-Clinical Senior Fellowship from the Faculty of Health and Life Science, University of Nottingham, and by a Marie Curie Intra-European Fellowship within the 7th European Community Framework Program.