Chronic depletion of subcellular NAD pools reveals their interconnectivity and a buffering function of mitochondria


 The coenzyme NAD is consumed by signaling enzymes including poly-ADP-ribose-polymerases (PARPs) and sirtuins. Understanding the mechanisms of aging-associated NAD decline and how cells cope with decreased NAD concentrations requires model systems reflecting chronic NAD deficiency. To evoke compartment-specific over-consumption of NAD, we have engineered cell lines expressing PARP activity in mitochondria, the cytosol, endoplasmic reticulum, or peroxisomes. Irrespective of the compartment targeted, total cellular NAD concentrations declined by ~40%. Isotope-tracer flux measurements and mathematical modeling showed that the lowered NAD concentration limits total NAD consumption kinetically. Moreover, NAD biosynthesis rate and capacity remained unchanged, thereby also precluding an increase of total NAD turnover. The chronic NAD deficiency was surprisingly well tolerated unless the mitochondria were targeted. Oxidative phosphorylation and glycolysis were little affected by NAD over-consumption in the other compartments. Likewise, peroxisomal NAD over-consumption was balanced by mitochondrial NAD decrease to maintain beta-oxidation of very long chain fatty acids in peroxisomes. We propose that subcellular NAD pools are interconnected, with mitochondria acting as a rheostat to facilitate NAD-dependent processes in organelles with excessive consumption.


Introduction
Among the most versatile and widely represented biomolecules in all living cells is NAD. As a redox factor, it participates in most, if not all branches of metabolism 1 . In metabolic reactions, NAD serves as electron carrier being interconverted between its oxidized (NAD + ) and reduced (NADH) form [2][3][4][5] . In addition to its redox functions, a variety of signaling processes require the cleavage of NAD + . Most prominently, NAD + is the substrate of ADP-ribosyltransferases (ARTs), sirtuins and ADP-ribosylcyclases such as CD38 and SARM1 [6][7][8][9][10] . The most thoroughly studied ART, poly-ADP-ribose polymerase 1 (PARP1), modifies itself and other proteins with polymers of ADP-ribose derived from cleavage of NAD + accompanied by the release of nicotinamide (Nam). PARP1 resides in the nucleus and its function is primarily associated with DNA damage detection and facilitation of the DNA repair process [11][12][13][14][15] . Sirtuins, which reside in the nucleus, cytosol and mitochondria, remove acyl groups from proteins by cleaving NAD + and transferring the acyl group onto the ADP-ribose moiety, thus generating O-acyl-ADP-ribose and Nam 10,16,17 . The functions of sirtuins range from epigenetic and transcriptional regulation to direct modulation of the activity of metabolic enzymes [18][19][20][21][22][23][24] . ADP-ribosyl cyclases generate the calcium messengers cyclic ADP-ribose and NAADP from NAD + and NADP + , respectively, reactions which are also accompanied by the release of Nam [25][26][27] .
This remarkable array of NAD + -consuming processes needs to be counterbalanced by NAD + biosynthesis. The major route of NAD + synthesis in mammals recycles Nam produced in the signaling reactions 28,29 . A limited amount of NAD is also generated from quinolinic acid, a product of tryptophan degradation, and from nicotinic acid [30][31][32] . Nam is converted to nicotinamide mononucleotide (NMN) by nicotinamide phosphoribosyl transferase, NAMPT 28,33,34 , an enzyme that has evolved an unusually high affinity to Nam to prevent excessive loss of the precursor 35 . NMN is also produced by nicotinamide riboside kinases (NRKs) using the nucleoside form of nicotinamide (NR) as substrate 36 . To form NAD + , NMN combines with the adenylyl moiety of ATP in a reaction catalyzed by NMN adenylyltransferases (NMNATs) 37 .
There are three NMNAT isoforms in mammals that provide the only obvious indication of compartment-specific NAD + biosynthesis, as they have been localized to the nucleus (NMNAT1), the Golgi complex, facing the cytosol (NMNAT2) and the mitochondrial matrix (NMNAT3) 38 .
The mitochondrial NAD pool has been shown to be of particular importance to protect cells in stress situations 45,46 . However, the way in which mitochondrial NAD might contribute to counteract stress originating from other cellular components is not known. It appears that the mitochondrial NAD pool has a certain degree of autonomy, as indicated by these observations, but also by the presence of NMNAT3 within this organelle 38 . Nevertheless, how this pool is generated and maintained is still a matter of debate and investigation. Previously, evidence has been presented suggesting the uptake of NMN or NAD + into the organelles 41,47,48 . Import of NMN appears plausible, given the presence of NMNAT3 in the matrix, capable of using it to form NAD + . However, the recent discovery of SLC25A51, or MCART1, as a mammalian mitochondrial NAD + transporter has established the molecular basis for the major route of generation of the mitochondrial NAD + pool 49,50 . Additionally, NMNAT3 has been demonstrated to be dispensable in mice, further suggesting that this enzyme is not essential for mitochondrial NAD + synthesis 51 .
Besides the nucleus, cytosol and mitochondria, NAD + has also been detected in other subcellular compartments including the peroxisomes and the endoplasmic reticulum (ER) 52,53 .
For peroxisomes, a carrier has been described as a possible candidate 54 , whereas the route of entry of NAD into the ER remains unknown. A fundamental question therefore relates to the connectivity between the individual NAD pools. To what extent are they autonomous or independent? That is, does local consumption of NAD + remain a local affair or does excessive consumption in one organelle affect other NAD pools? This question becomes particularly relevant in view of the observed decline of tissue NAD + content in some diseases such as mitochondrial myopathies 55,56 and, especially, in aging [57][58][59] . Decreased NAD + content is most often interpreted as a result of excessive NAD + consumption or decreased NAD + biosynthesis.
As a cause for the initial change, this is probably true. However, at equilibrium, synthesis and consumption must have equal rates. Therefore, cause and consequence cannot be readily identified when the change has become stably manifested. In line with this notion, Liu et al. have recently established NAD + turnover rates in cells and tissues by means of stable isotope labeling and LC-MS based quantification 60 . They demonstrated that NAD + levels per se are unsuitable to evaluate the activities of cellular NAD + synthesis and consumption 60 .
So far, the consequences of NAD depletion on physiological functions have been largely studied through pharmacological blockage of NAMPT, for example, using the inhibitor FK866 45,[61][62][63][64] . This approach has provided important insights into the acute consequences of the essentially irreversible inhibition of NAD + biosynthesis. Increased levels of protein acetylation were observed owing to the ensuing inactivation of sirtuins 65,66 . However, the relevance of these observations to chronically decreased NAD + levels, as observed in aging, is unclear.
Here, we present a tunable model system for chronic, compartment-specific NAD + depletion in human cells. Expression of the catalytic domain of PARP1 resulted invariably in a ~40% lowered total cellular NAD content, irrespective of the subcellular compartment targeted. NAD + depletion was surprisingly well compensated as long as the mitochondrial pool was not directly targeted, consistent with a role of this pool as a cellular NAD buffer. The excessive NAD + -degrading activity was kinetically counterbalanced by the observed NAD decrease to maintain total NAD turnover constant, a mechanism that could plausibly explain aging-related NAD decline. Our study thus revealed distinct cellular responses to chronically lowered NAD levels that are rather different from acute NAD depletion.

PARP1cd expression decreases cellular NAD + content by ~40%, irrespective of the subcellular compartment targeted
We have previously demonstrated that HEK293 (293) cells stably expressing the catalytic domain of PARP1 (PARP1cd) in the mitochondria (293mitoPARP1cd or mP cells) or the cytosol (293cytoPARP1cd or cP cells) represent powerful systems to study these subcellular NAD + pools 52,67,68 . To comprehensively analyze the consequences of compartment-specific chronic NAD + depletion, we generated two additional cell lines, stably overexpressing PARP1cd in peroxisomes (pexPARP1cd or pP) and the ER (ER-PARP1cd or erP), respectively. PARP1cd, endowed with an N-terminal EGFP-tag, was targeted to these organelles by adding a Cterminal SKL peroxisomal targeting signal or the ER-targeting sequence of BiP and the KDEL ER retention signal for the ER (Fig. 1A). The resulting proteins consume NAD + as a substrate for PAR formation in the targeted compartment of interest 52,53 (Fig. 1B). Immunodetection of PAR in the pP cells (Figs. 1C and 1D), and peroxisomal localization were confirmed by colocalization with the peroxisomal marker PMP70 (Fig. 1D). Surprisingly, unlike in cells transiently expressing ER-PARP1cd (Dölle et al., 2010;VanLinden et al., 2017), PAR formation was undetectable in the stably expressing erP cells (Fig. 1E). Addition of nicotinic acid (NA) to the cell culture did not lead to detectable polymers. However, concomitant overexpression of NA phosphoriboslyltransferase (NAPRT) led to readily detectable PAR accumulation (Fig. 1E).
Presumably, PAR degrading activity in the ER is high and the activation of the Preiss-Handler pathway through NAPRT expression led to sufficient NAD + supply to maintain a detectable PAR level in this compartment. The correct targeting of the erP construct was confirmed by colocalization with the ER marker calnexin (Fig. 1F).
To exploit PARP1cd as a compartment-specific NAD + consumer requires a PAR-degrading activity, that is, a dynamic equilibrium of polymer synthesis and degradation. Otherwise, PARP1cd would be mostly in its maximally PARylated form and inactive. The presence of PARdegrading activities has been previously verified for mP and cP cells 48,67 and here for erP cells (see above, Fig. 1E). To determine whether PAR degrading activity is present within peroxisomes, pP cells were incubated in presence of the PARP inhibitor 3-aminobenzamide (3AB) and the effect on PAR levels was monitored over time. Since mitochondrial PAR turnover has been previously described for mP cells 67 , these cells were used as a positive control. In presence of 3AB, the signal for PAR in pP cells weakened in a time-dependent manner, albeit at a somewhat slower rate than that observed in the mP cells (Fig. 1G). Pre-incubation of the cells in presence of 3AB for 48 hours followed by release of the inhibition showed that polymer levels were readily restored in the pP cells at 24 hours (Fig. 1H). Thus, in pP cells, polymers were restored faster than in mP cells (Fig. 1H). Collectively, these results indicate the presence of PAR-degrading activities and dynamic turnover in all selected organelles. Notably, this is the first demonstration of PAR metabolism in the ER and peroxisomes.
Homogenous expression of the PARP constructs for each cell line used in this study was confirmed by flow cytometry based on the EGFP signal harbored by the recombinant proteins ( Fig. S1). However, the level of PARP1cd expression varied somewhat depending on the targeted compartment (Fig. 1I).
Strikingly, the total cellular NAD + levels were lowered by approximately 40 % regardless of the compartment targeted (Fig. 1J). Importantly, incubation of the PARP1cd cell lines with 3AB restored cellular NAD + concentrations, indicating that the observed NAD + depletion was exclusively due to the catalytic activity of the overexpressed constructs (Fig. 1J). Therefore, the established cell lines represent a suitable model system to study chronic, compartmentspecific NAD + depletion.

Chronic NAD + depletion is tolerated surprisingly well by human cells
To assess functional alterations in the PARP1cd cell lines, we first conducted a quantitative proteomics analysis using TMT labeling to establish individual and common changes compared to the parental 293 cell line. Three independent experiments were conducted as replicates for each cell line. Principal component analysis of the protein abundances showed good separation between the different cell lines and clustering of the replicates (Fig. S2A). As shown in Fig. 2A, cP and mP cells exhibited by far the largest amount of proteins whose abundance changed. This might be expected, given that the majority of known NAD + -dependent processes take place in the cytosol (and nucleus) and the mitochondria. Interestingly, 112 affected proteins were common for cP and mP, but the direction of change (increase or decrease of abundance) did not always coincide (https://github.com/MolecularBioinformatics/NADpools). Network enrichment analyses ( Fig.   S2 B-F) showed that among the observed changes across all cell lines, alterations related to protein folding and metabolic processes, especially central carbon metabolism, were most abundant. Some of the effects align with organelle-specific functions for the targeted compartment. For example, carbon metabolism was most affected in mP and cP cells, while protein processing in the ER and peroxisomal processes were among the most abundantly changed in erP and pP cells, respectively. Nevertheless, the changes we observed overall were rather small (https://github.com/MolecularBioinformatics/NADpools) and therefore indicate that chronic NAD + depletion evokes relatively mild changes in protein profiles.
We next sought to investigate functional consequences of the subcellular PARP1cd expression.
First, we analyzed how NAD + -dependent processes were affected by NAD + depletion in the PARP1cd cell lines. We measured cell proliferation rates and observed no noticeable differences in the PARP1cd cell lines compared to the parental 293 cells (Fig. 2B, solid lines).
Blocking NAD + biosynthesis by FK866, a NamPT inhibitor, leads to rapid cellular NAD + depletion and eventually cell death 48,61,69 . However, in the time frame of the experiment (48h), some loss of cells was observed for cP cells only, while the cell number of the other cell lines remained fairly unchanged indicating growth arrest (Fig. 2B, dashed lines). 3AB caused a partial rescue (Fig. 2B, dotted lines), as it inhibits endogenous and overexpressed PARP activities and thereby reduces NAD + consumption and, consequently, the need for its biosynthesis.
Owing to the NAD + dependency of sirtuins, changes in cellular NAD + levels may severely affect the acetylation state of a variety of proteins [70][71][72] . Therefore, we compared the protein acetylation state following acute NAD + depletion evoked by FK866 treatment in parental cells to that during chronic NAD + deficiency in the PARP1cd cell lines (Fig. 2C-E). Acute FK866mediated NAD + depletion in the 293 cells promoted prominent changes. In particular, a strong increase in the acetylation of α-tubulin (Fig. 2C), which is deacetylated by SIRT2 65,73 , as well a decrease in the acetylation of H3K9, likely caused by cell cycle arrest in response to energy depletion 74,75 . In contrast, no major changes were detectable when comparing protein acetylation in whole cell lysates of parental and PARP1cd cell lines (Fig. 2D). The same held true for the acetylation state of proteins in isolated mitochondria (Fig. 2E). These data suggest that the chronically NAD + -depleted cells have adapted to maintain NAD + -dependent deacetylation activities even when the concentration of the dinucleotide is substantially decreased.

Direct mitochondrial NAD + depletion affects bioenergetics
To assess the potential impact of compartment-specific NAD + depletion on energy metabolism, we monitored mitochondrial respiration and glycolysis by measuring the oxygen consumption rate (OCR) and the extracellular acidification rate (ECAR), respectively (Figs. 3A and S3). mP cells displayed considerably lower basal and maximum (uncoupled) respiration rates, as noted previously 68 , whereas cP and erP cells exhibited only a slightly reduced maximum respiration (Fig. 3A). Conversely, glycolysis was strongly elevated in mP cells, whereas cP and erP cells were far less affected (Fig. 3A). None of the modified cell lines presented increased leak respiration, supporting that the mitochondrial inner membrane integrity and respiratory control remained intact ( Fig S3). PARP1cd expression in peroxisomes (pP cells) affected neither respiration nor glycolysis. Consequently, the major bioenergetic functions appear to be well maintained during constitutive NAD + deficiency, unless the depletion originates from the mitochondria, as is the case in mP cells.
These observations prompted us to employ the resazurin assay, which is an indicator of the activity of mitochondrial NAD + -dependent dehydrogenases 76 and often used to assess cell viability. Mitochondrial dehydrogenase activities appeared to be maintained in all cell lines regardless of the origin of NAD + depletion. Similar to the proliferation measurement in Fig. 2B, the PARP1cd cells were then challenged with FK866. Both mP and pP cells were highly sensitive to this condition displaying a marked decrease already after 24 hours with a further decline after 48 hours whereas cP and erP cells tolerated the treatment for the first 24 hours (Fig. 3B). Regarding pP cells, this finding was surprising, as these cells exhibited normal oxygen consumption rates. Therefore, we assessed the functionality of -oxidation in the PARP1cd cell lines, an NAD + -dependent process taking place both in peroxisomes (preferential catabolism of long and very long-chain fatty acids) and mitochondria (preferential oxidation of short/medium chain fatty acids) 77 . As shown in Fig. 3C, under conditions of mitochondrial NAD + depletion in mP cells, a dramatic accumulation of medium-chain fatty acids was observed, whereas peroxisomal NAD + depletion (pP cells) resulted only in a slight increase of very long-chain fatty acids. Since the NAD + dependency of peroxisomal and mitochondrial fatty acid oxidation is similar, it appears that peroxisomal NAD + depletion is well compensated, whereas mitochondrial NAD + depletion results in a comparatively severe deficiency of oxidation.
Taken together, PARP1cd cells cope surprisingly well with the constitutive, targeted depletion of NAD + . These observations indicate that, unlike during acute NAD + depletion, cells can adapt to chronic NAD + deficiency and efficiently maintain physiological functions under these conditions. However, the pronounced functional deficiencies in mP cells demonstrate that targeted depletion of the mitochondrial NAD + pool may have severe consequences on cellular homeostasis.

Total cellular NAD + turnover is not increased by PARP1cd overexpression
Next, we wished to understand the mechanism how the PARP1cd cells maintained their phenotypic homeostasis despite the lowered NAD + concentrations. These are stable cell lines that have attained a steady state. Therefore, to achieve this equilibrium, an augmented NAD + consumption could be counterbalanced by an equal increase in biosynthesis. To test this possibility, we determined the NAD + turnover using stable isotope labeling-based flux measurements. We substituted the respective unlabeled cell culture medium components by 15 N Nam (32M) and 13  indicates that NMN is readily formed from Nam and PRPP (resulting in NAD + M+6), and that label incorporation from PRPP into ATP proceeds more slowly, presumably because the cellular ATP concentration is high and therefore takes more time to become isotope-labeled.
We then applied this approach to the PARP1cd cell lines. As can be inferred from Figures 4C and S4B, the NAD + half-lives were considerably shorter than that measured for the 293 control cells. Strikingly, mP and pP cells showed the strongest and nearly identical decrease in their half-lives (2h13  3 min and 2h17  3 min). NAD + half-lives were also shorter for cP and erP, although to a smaller extent, indicating that, depending on subcellular localization, NAD + depletion may affect cellular NAD + turnover differently. However, to appropriately compare the NAD + turnover (the number of NAD + molecules synthesized and consumed per unit of time per cell), we had to take into account the actual NAD + contents in the different cell lines (Fig.   1J). As shown in Figure 4D, the observed decrease in NAD + half-life in the PARP1cd cell lines can be primarily ascribed to their lower NAD + content. Remarkably, NAD + turnover did not increase in any of the engineered PARP1cd cell lines, despite the constitutive overexpression of an additional NAD + consumer. The mP and pP cells exhibited NAD + turnover similar to that of the control cells. That is, their NAD + biosynthesis and consumption are essentially the same as those measured for the parental cell lines. On the other hand, cP and erP cells displayed an even slower NAD + turnover (Fig. 4D). Intuitively, this may seem paradoxical, since PARP1cd overexpression should increase cellular NAD + consumption and, in turn, biosynthetic rates to maintain the steady state. To explain this phenomenon, we used a kinetic model of NAD + metabolism (https://github.com/MolecularBioinformatics/NADpools) to simulate PARP1cd overexpression. As shown in Figure 4E, the model recapitulates the observations of the experiments (Fig. 4D) when arbitrarily assuming different maximal NAD + consuming activities (10 and 20% increase). However, in the steady state, these maximal activities are not attained, because of a kinetic counterbalancing due to the lowered NAD + concentration. That is, the Km values of the NAD + consumers (both endogenous and PARP1cd) are in the range of the (free) NAD + concentrations in the cell 41,78 . Therefore, a decrease of the NAD + concentration will cause a decrease of the overall NAD + consumption rate until the biosynthetic rate is matched (Fig. 4F). These observations indicate that cells might maintain a rather stable NAD + biosynthesis rate. In turn, when the total capacity (Vmax) of NAD + consumers rises (for example, increased expression of NAD + -dependent signaling enzymes), unchanged NAD + synthesis would limit total cellular NAD + consumption kinetically.

Stable NAD + biosynthesis kinetically limits NAD + consumption
To consolidate this surprising finding, we analyzed the expression of enzymes involved in NAD + biosynthesis. Under the conditions of our experiments, the only precursor added was nicotinamide, NAD + biosynthesis proceeded via NamPT and NMNAT, the only intermediate being NMN (Fig. 5A, right). The contribution of de novo NAD + synthesis from tryptophan in our cell lines was likely negligible because inhibition of NamPT by FK866 led to rapid growth arrest and eventually cell death (Figs. 2B and 3B). As shown in Figure 5B, the expression of NamPT was similar in all PARP1cd overexpressing cell lines and seemed unchanged compared to the parental 293 cell line. The proteomics analyses confirmed this evaluation and revealed a decrease only in cP cells (~25%, Fig. S2G). Mathematical modelling of NAD + metabolism indicates that even a small change in NamPT activity could have noticeable consequences on NAD + turnover (Fig. 5C). Next, we analyzed the presence of the individual NMNAT isoforms in the different cell lines. We did not consistently detect peptides from these enzymes in the proteomics analyses, presumably because tryptic digestion of these proteins generates many small and few rather large peptides. Therefore, we used Western blotting to reveal potential changes in NMNAT abundances. The most noticeable and only significant adjustment in the expression of NMNATs was observed in mP cells where NMNAT2 was upregulated by approximately 3-fold, suggesting an increased NAD + forming capacity in the cytosol (Figs. 5D, E, F).
We wondered whether there may have been constraints on the activity/expression of NamPT that precluded an upregulation of NAD + biosynthesis. To test this, we included an alternative precursor, nicotinic acid (NA). This precursor enters NAD + biosynthesis through the Preiss-Handler pathway 31,79 , which is independent of NamPT ( Fig. 5A). It was previously established that NA can sustain NAD + synthesis and cell growth in 293 and mP, when NamPT is inhibited by FK866 48 . Under this condition, the NAD pool is replenished using NA ( Fig. S5) indicating that this pathway is functional and active in the cell lines used in the present study. To measure the contribution of NA consumption to overall NAD + biosynthesis in our cell lines, we made use of 13 C-labeled NA 80 . We did not observe any increase in the use of NA as NAD + precursor in the cells expressing PARP1cd compared to the parental cells ( Fig. 5G) confirming that upregulation of NAD + biosynthesis was generally not used as a compensatory mechanism to counteract increased NAD + consumption. Surprisingly, utilization of NA was even significantly diminished in mP cells (Fig. 5G). Indeed, the proteomics analyses revealed a decrease in abundance of NAPRT in these cell lines by about 50% (Fig. S2G). On the other hand, overexpression of NAPRT can increase NAD + synthesis from NA, because it enabled the accumulation of PAR in the erP cells (Fig. 1E).
Collectively, these results indicated that, independent of the compartment harboring the overexpressed NAD + consumer, NAD + biosynthesis was not increased to counterbalance the increased NAD + consumption. Rather, by maintaining the expression level of NamPT, the ratelimiting enzyme 81 , unchanged NAD + consumption was kinetically limited, which is consistent with the observations from the turnover experiments (Fig. 4).

Direct, but not indirect depletion of mitochondrial NAD + is detrimental to cellular homeostasis
As opposed to the cells expressing PARP1cd in the cytosol, peroxisomes or ER, mP cells exhibited noticeable phenotypic changes that indicated adverse effects including deficiencies in energy metabolism, fatty acid oxidation and increased NMNAT2 expression. Consequently, the mitochondrial NAD + pool appears to play a particular role for cellular NAD + homeostasis.
To assess the mitochondrial contribution to NAD + metabolism of the entire cell, we compared the time courses of labeled Nam incorporation into mitochondrial and whole-cell NAD + in 293 cells. Cells were grown on 10 cm dishes, harvested at the times indicated and mitochondria isolated, while a sample was taken to determine label incorporation into total cellular NAD + .
As shown in Figure 6A, left panel, label incorporation into the mitochondrial NAD + pool closely followed that of total cellular NAD + , albeit with a small but reproducible delay. When the same experiment was carried out with mP cells, the relative delay of mitochondrial NAD + labeling became appreciably larger (Fig. 6A, middle panel). For comparison, we also analyzed the pP cells ( Fig. 6A, right panel), because they exhibited a similar total cellular NAD + turnover (Fig.   4D). The relative delay between total and mitochondrial NAD + synthesis was similar to that in the control 293 cells (Fig. 6A). To enable calculation of mitochondrial NAD + turnover, we estimated the mitochondrial NAD + concentration in 293, mP and pP cells. The NAD + content in mitochondria isolated from mP cells was dramatically lowered by approximately 80% (Fig.   6B). Strikingly, a strong NAD + decrease was also found for mitochondria isolated from pP cells.
To validate these measurements, we made use of HeLa cells constitutively expressing a genetically encoded NAD + biosensor in mitochondria 41 . Transient expression of the mitochondrial or peroxisomal PARP1cd constructs in these cells demonstrated a similar decline of the free mitochondrial NAD + concentration as observed in the corresponding 293derived PARP1cd cell lines (Fig. 6C). Inhibition of PARP activity by 3AB restored original NAD + levels in mitochondria. Treatment of the sensor-cells alone with FK866 strongly diminished free mitochondrial NAD + concentration, as expected. Based on these measurements, we calculated the mitochondrial and extramitochondrial (that is, total minus mitochondrial) NAD + turnover in these cell lines (Fig. 6D). While the total turnover in 293, mP and pP cells was similar (Fig. 4D), the subcellular distribution of NAD + turnover turned out to be very different.
In the parental cells, mitochondrial NAD + turnover is almost as fast as the extramitochondrial one, while in mP cells, mitochondrial NAD + turnover is ten times slower compared to the extramitochondrial one. That is, even though mP cells express an additional NAD + consumer (PARP1cd) in the mitochondria, NAD + turnover is drastically slowed. Even more striking, mitochondria from pP cells showed a similar behavior (Fig. 6D). These observations suggest that mP cells have adapted to the presence of the PARP1cd construct in mitochondria by dramatically decreasing the mitochondrial NAD + content, thereby kinetically minimizing the activity of the construct. On the other hand, in pP cells we also observed a mitochondrial NAD + depletion. This would indicate that mitochondria might compensate for the elevated NAD + consumption in peroxisomes by providing NAD + . Indeed, despite a similar total decrease of NAD + in the pP and mP cells, targeting the PARP1cd construct to peroxisomes resulted in only subtle functional deficiencies.

NMNAT3 is not required for mitochondrial NAD + generation
A putative buffering function of the mitochondrial NAD + pool would be intrinsically linked to the mechanisms underlying the establishment and maintenance of that pool. Recent studies have shown that the main route for the maintenance of mitochondrial NAD + levels is associated with SLC25A51/MCART1, an NAD + carrier in the mitochondrial inner membrane 49,50 . Therefore, the role of the previously identified mitochondrial NMNAT3 38,82 in mitochondrial NAD + homeostasis is unclear. Moreover, it was recently shown that NMNAT3 is not vital in mice 51 . To verify whether NMNAT3-deficient mitochondria from human cells contain sufficient NAD + , we used CRISPR/Cas9-edited HAP1-Nmnat3 -/cells. As shown in Figure  the total cellular NAD + in WT and KO cells is also similar (Fig. 7D). These results are consistent with the notion that, even in a cell normally expressing NMNAT3, this enzyme is not required to maintain mitochondrial NAD + levels. Consequently, the enzyme may not be solely involved in mitochondrial NAD + biosynthesis. In fact, the NMNAT-catalyzed reaction is fully reversible 38,83 , the equilibrium favoring the production of NMN and ATP from NAD + and pyrophosphate (PPi) (Fig. 7E). Therefore, in cooperation with SLC25A51/MCART1, NMNAT3 could act as a rheostat buffering cellular NAD + levels according to demand (Fig. 7E).

Discussion
The present study has provided fundamentally new insights into the physiology of intracellular NAD + pools, their interconnectivity, dynamics and plasticity. To explore NAD + biology on a subcellular level, we established and analyzed human cell lines stably expressing an NAD + consumer targeted to various subcellular compartments. The use of the catalytic domain of PARP1 as targeted NAD + consumer has several advantages. The protein can be easily targeted (as it lacks the nuclear localization signal of full-length PARP1). Moreover, it retains its catalytic activity when fused to GFP, a feature that makes it easy to monitor proper subcellular localization and homogenous expression within a cell population. The possibility to immunodetect PAR as an indicator of the catalytic activity is a further advantage, even though PAR is not detectable in cP and erP cell lines, despite the activity of the construct 52 . Perhaps, the most important advantage is the availability of specific PARP inhibitors permitting to control the catalytic activity, that is, making it a tunable system that does not rely on druginducible expression.

Chronic cellular NAD + deficiency reveals unexpected adaptive mechanisms
In addition to the organelle-specific NAD + depletion, our model system enabled the experimental study of chronic rather than acute NAD + deficiency in human cells. Many investigations have made use of potent NamPT inhibitors such as FK866 to identify the consequences of acute NAD + depletion 45,[61][62][63][64] . Treatment of cells with FK866 causes rapid NAD + loss in 293 cells, commensurate with the NAD + turnover rate determined in the present study. Under these conditions, the equilibrium between synthesis and degradation is disrupted, as NAD + synthesis is essentially switched off, while degrading enzymes continue cleaving the remaining NAD + , albeit with decreasing activity owing to the declining NAD + concentration. The diminished activity can be noted, for example, by increased levels of protein acetylation owing to decreasing sirtuin-dependent deacylation. In the end, the cells die because of insufficient NAD + supply 84,85 . Given the inevitable outcome, it is not obvious whether the observed effects of acute NAD + depletion reflect adaptive mechanisms or unregulated consequences of cell death.
In contrast, chronic NAD + shortage, as evoked in our cell lines, requires the establishment of a new equilibrium between NAD + synthesis and consumption. In principle, when the total number of NAD + consumers (Vmax) increases, this could be counteracted by increasing the activity of NAD + biosynthesis to match the higher demand. Such a response would result in an increased cellular NAD + turnover ( Fig. 4E and F). Surprisingly, our measurements revealed that the common response to PARP1cd overexpression was no change or even a slight decrease in NAD + biosynthesis (erP and cP cells) and, hence, NAD + turnover. Mathematical modelling showed that the robust maintenance of the NAD + biosynthesis flux (as primarily reflected by unchanged expression of NamPT, the rate-limiting enzyme) is an appropriate adaptive mechanism, because, in the steady state, consumption can only be as fast as biosynthesis (Fig.   4). However, as a consequence, the total cellular NAD + level needs to decrease. Thereby, NAD + -consuming enzymes are kinetically downregulated, due to diminished substrate saturation, until the consumption rate again matches that of NAD + synthesis. Thus, a new steady state is attained with the same total turnover as originally present in the parental cells.
Strikingly, the fact that the changes observed in the PARP1cd cell lines can be well described by a purely kinetic model of NAD + metabolism (Fig. 4F) indicates that there are no discernable signaling mechanisms governing NAD + homeostasis. We conclude, therefore, that the general cellular maintenance of NAD + homeostasis is regulated primarily based on enzyme kinetics.
Consistent with this notion, according to the KEGG database 86 , as many as ~25% of all biochemical reactions use NAD(P) as cofactor, a number that would make it inconceivable to invoke regulation of specific enzymes to regulate global NAD + homeostasis. The only noticeable adaptation of NAD + synthesis was the increased expression of NMNAT2 in mP cells.
Interestingly, NMNAT2 has been shown to be upregulated during adipocyte differentiation to support cytosolic NAD + synthesis 42 . Therefore, this enzyme seems to emerge as a key player in controlling the distribution of NAD + into different subcellular pools.

Subcellular NAD + pools interact to maintain global NAD + homeostasis
A surprising finding was that, regardless of its subcellular location, the constitutive activity of PARP1cd led to a decrease of the total cellular NAD + content to about 60% of the parental cell line. A plausible interpretation of this observation is that the remaining amount of NAD + represents a minimum to maintain viability. Cells that might acutely tolerate even lower NAD + contents cannot survive the long-term depletion during the generation and maintenance of stably transfected PARP1cd cell lines. On the other hand, the convergence to the same remaining NAD + content in all the PARP1cd cell lines points towards a hitherto undetected dynamic interplay between the different subcellular NAD + pools and suggests the presence of mechanisms that sense and redistribute the total cellular NAD + content upon local depletion.
Indeed, the concentration of NAD + in mitochondria is known to be substantially higher than in other subcellular compartments 41,78 , and the NAD + metabolism of these organelles has been proposed to be autonomous 45,46 . Nevertheless, even in mitoPARP1cd cells, the same total cellular NAD + concentration was attained. Thus, our data provide evidence that a redistribution of cellular NAD + pools takes place to match the essential demands in all compartments. Of note, however, despite being the pool with the highest NAD + concentration under normal conditions 41,78 , direct constitutive depletion of mitochondrial NAD + causes severe functional deficiencies. Consequently, mitochondria can provide NAD + to maintain other pools, but are insufficiently supported when in critical demand themselves.

Mitochondrial NAD + homeostasis is critical to maintain cellular functions
We found that the incorporation of precursors into mitochondrial NAD + is slightly delayed relative to the other pools (Fig. 5). This observation indicates that the uptake of NAD + into mitochondria is rate-limiting for the maintenance of mitochondrial NAD + . When mitochondrial NAD + is subjected to degradation by the expression of mitoPARP1cd, the mitochondrial NAD + turnover is strongly reduced, and the delay becomes more pronounced (Fig. 5). This may seem paradoxical, since overexpression of an NAD + consumer should increase the turnover.
However, it turned out that the mitochondrial NAD + content in the mP cells is dramatically diminished. Presumably, to minimize NAD + cleavage by the overexpressed construct, the concentration of mitochondrial NAD + was strongly decreased. Thereby, even the activities of other NAD + -dependent processes were affected resulting in the overall decrease of mitochondrial NAD + turnover. Importantly, when assessing mitochondrial NAD + content and turnover in pP cells, these were also strongly decreased. Notably, mP cells exhibited a strong accumulation of short-and medium chain fatty acid esters (Fig. 3C) and a substantially lower rate of oxidative phosphorylation (Fig. 3A), both NAD + -dependent mitochondrial processes. In contrast, despite the considerable mitochondrial NAD + depletion, pP cells did not exhibit any measurable OXPHOS deficiency, while the resazurin assay revealed only a slight increase of sensitivity of mitochondrial NAD + -dependent dehydrogenases when challenged with FK866 ( Fig. 3B). Moreover, the abundance of long-chain fatty acids appeared to be only slightly elevated indicating that a major NAD + -dependent peroxisomal process was only marginally affected despite cellular NAD + depletion originating from these organelles. Likewise, we did not detect any noticeable defects in bioenergetic or other NAD + -dependent functions in cP or erP cells. Since the overall NAD + decrease was similar in all engineered cell lines, mitoPARP1cd cells clearly stand out with their vulnerability when these organelles represent the origin of enhanced NAD + degrading activity. As exemplified by the pP cells, mitochondria appear to have the capacity to support NAD + -dependent functions in other subcellular compartments.

Mitochondrial NAD + metabolism is compatible with a function as cellular NAD + rheostat
Mitochondria are known to have the highest subcellular NAD + concentration in a variety of cell lines, including in 293 cells 41,78 . So far, the reason for this NAD + accumulation in mitochondria has remained unclear. Considering the results from the present study, we propose that mitochondria may serve as an NAD + reservoir to buffer fluctuations in cellular NAD + concentrations. This suggestion is consistent with previous studies demonstrating the important role of the mitochondrial NAD + pool for cell vitality and the detrimental effects when this pool is impaired 45,46,68 .
It has recently been reported that SLC25A51 is a mitochondrial NAD + carrier essential to maintain the mitochondrial NAD + pool in mammals 49,50 . The majority of the SLC25 family of mitochondrial carriers are known to be reversible exchange carriers 87,88 . Therefore, in principle, SLC25A51 might also mediate the export of NAD + from mitochondria, on demand ( Fig. 7E). On the other hand, NMNAT3 has been proposed to catalyze the generation of NAD + from NMN and ATP in mitochondria, which would require NMN uptake into the organelles 48 .
However, it has been shown that Nmnat3 -/mice are viable 51 . Indeed, as we demonstrate here, knocking out NMNAT3 in HAP1 cells (which normally express NMNAT3 and SLC25A51) did not affect mitochondrial or whole cell NAD + content ( Fig. 7B-D). Given the reversibility of both the mitochondrial NAD + carrier and the reaction catalyzed by NMNATs, it appears plausible that mitochondria may fulfill their function as cellular NAD + buffer by maintaining an appropriate balance based on the cooperation between SLC25A51 and NMNAT3. In such a scenario, depending on the extramitochondrial NAD + demand, mitochondrial NMNAT3 would generate NAD + for export when needed or, when demand outside the organelles is low, imported NAD + is cleaved to NMN and ATP, thereby maintaining a readily available reserve for rapid NAD + production. sensitivity, 50,000 cells were seeded in triplicate in the cavities of a 12-well plate prior to treatment with FK866 (2 μM) with or without 3-aminobenzamide (3AB, 1 mM)). After incubation for the indicated time points, samples were prepared as described above. Cell counting was performed on a BD Accuri C6 flow cytometer according to the manufacturer's guidelines. Data analysis was conducted using FlowJo X. Cells were gated to exclude debris, followed by standard doublet exclusion.

Resazurin-based in vitro toxicology assay
To assess mitochondrial NAD-dependent dehydrogenase activity, a resazurin-based in vitro Images were taken using a Leica TCS SP8 STED 3x confocal laser scanning microscope equipped with a ×100 oil immersion objective (numerical aperture 1.4).

Measurement of cellular glycolytic rate and oxygen consumption rate
Oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) in 293 cells were µM rotenone (complex I-independent respiration) and 1 µM antimycin A (non-mitochondrial respiration) were conducted to investigate mitochondrial respiratory function. For measuring glycolysis function, the basal ECAR was measured in unbuffered assay medium (pH 7.4, containing 2 mM L-glutamine). This was followed by an injection of 10 mM glucose to measure glycolysis-dependent ECAR. Next, 3 µM oligomycin (glycolytic capacity) and finally 100 mM 2deoxyglucose (non-glycolytic ECAR) were added. All data were normalized to protein content, measured using BCA reagent as previously described. Prep. Elysian, MN, USA) that was run as every 6th sample in the analytical sequence.

NAD + biosensor calibration
NAD + biosensor calibration was conducted essentially as described earlier 41 . 500,000 HeLa cells stably expressing the NAD biosensor in the cytosol or the cpVenus control 41  On the day before the experiment started, medium was exchanged for medium containing dialyzed serum. On the next day, at T0, medium was exchanged for medium containing the labelled compounds and samples were collected at the indicated time points.
For sample collection, cells were placed on ice, the medium was removed, and the cells were For normalization to protein content, the pellet was dried overnight and resuspended in 100 μl lysis buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 2% (w/v) SDS, 1% (w/v) EDTA). Protein concentration was subsequently determined using the BCA assay kit (Thermo Scientific) according to the manufacturer's recommendations.

Isotopolog Correction
To correct for natural isotope abundance the python package PICor (https://github.com/MolecularBioinformatics/PICor) was used. It applies a theoretical isotopolog correction approach based on statistical distributions of each isotopolog. The detailed approach has been described previously 93 .

Determination of cellular half-life and turnover rates
The half-life of NAD was estimated based on fitting an exponentially decay functions (see equation 1) to the measured ratios of unlabeled to total NAD over time ( ). The SciPy 97 optimization function curve fitting was used for determining the prefactor and the exponential factor along with their respective standard deviations, see equation 2 and 3.
The ratio of unlabeled to total NAD amount ranges between 0 and 1 with no label being present at time point zero. The exponential curve of the unlabeled compound should therefore start at 1 and approaches 0 over time. The prefactor should therefore be close to 1 and its deviation is an indication for the quality of the fit.

Determination of mitochondrial half-life and turnover rates
The cellular NAD pools are connected but have distinct concentrations and turnover rates. As we can only determine whole cell and mitochondrial NAD concentrations and turnover, the parameters for the nucleo-cytoplasm (extramitochondrial) turnover were determined in a multistep process. First the mitochondrial labeling dynamics were fitted based on the ratio of unlabeled to total NAD in mitochondria, as above: To estimate the extramitochondrial turnover, we calculated the distribution of NAD between the nucleo-cytoplasm and mitochondria as follows: The volume of mitochondria and nucleocytoplasm were determined in 293 cells using confocal microscopy. Free NAD concentrations in both compartments were measured using the NAD biosensor expressing HeLa cells as described above. These measurements were then used to calculate the mitochondrial and nucleo-cytoplasmic amount of NAD per cell, assuming a ratio of free to bound NAD of 1:10.
The amount of mitochondrial NAD in 293 cells stably transfected with PARP1cd was calculated accordingly using the relative differences in mitochondrial NAD measured by LC-MS and the NAD concentrations measured in whole cells.
In a second step, whole cell measurements were used to estimate the turnover rate of the extramitochondrial fraction of the NAD pool, which is here referred to as "nucleocytoplasmic", to simplify descriptions. A sum of two exponential function was used to describe the combined cellular pool using subcellular NAD amounts as scaling factor: The turnover for the nucleo-cytoplasmic and the mitochondria pool were calculated accordingly:

Quantitative proteomics
Cells were cultured as described above. Following trypsinization, 10 6 cells were counted and pelleted by centrifugation for 10 min at 400 x g.
Quantification of proteins and peptides was achieved by an MS/MS-based analysis strategy using isotopomer labels as described previously 98 . Analysis was conducted by the Proteomics Unit at the University of Bergen (PROBE) 99 . To identify significant differences in the proteome of different cell lines, only proteins identified with high confidence (FDR < 1%) were included in the further analysis.
We performed independent parametric t-tests for each protein comparing the abundance ratios of the PARP1cd cell lines versus the 293 cells. Proteins with p < 0.001 were interpreted to be significantly differentially expressed. For these significantly changed proteins we calculated mean log2 fold changes between each PARP1cd cell line and 293 cells.
We then performed Principal Component Analysis (PCA) on the proteins changed with high confidence (FDR > 1%) using scikit-learn 0.23.1 100 . To ensure equal feature variance, the protein abundances were normalized to their Z-Scores. Components were added until 95% of total variance was accounted for, which resulted in a total of 21 components. The results were plotted using the Python packages seaborn 0. 10

Model simulations
To simulate the effect of changes in NAD consumption and biosynthesis we constructed two mathematical models. To simulate the label integration in cell lines in a steady state situation, we created a mathematical model including a reaction for NAD consumption connected to NamPT and NMNAT to simulate the biosynthesis and consumption cycle. Only the major isotopologs were included, assuming instant labeling of glucose, Nam and PRPP. Competition between different isotopologs were simulated using competitive binding in random irreversible bimolecular reactions as described earlier 103 . In addition, ATP synthesis and degradation were simulated using mass action kinetics to be able to resemble the dynamic behavior of all major NAD isotopologs.

Figure 3: Direct mitochondrial NAD + depletion affects bioenergetics
A) Activity and integrity of mitochondrial respiration in terms of oxygen consumption rate (OCR), and glycolysis in terms of extracellular acidification rate (ECAR) in stably transfected PARP1cd cell lines compared to parental 293 cells as determined by extracellular flux analysis. Data are presented relative to parental 293 cells as mean ± SEM where n = 3. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 (Student's t-test). B) Activity of mitochondrial NAD-dependent dehydrogenases as measured by resazurin assay in parental 293 cells and stably transfected PARP1cd cell lines in the absence or presence of the NamPT inhibitor FK866 (2 µM) or its solvent DMF. Additionally, cells were incubated in the presence of both FK866 (2 µM) and the PARP inhibitor 3aminobenzamide (3AB, 1 mM). Data are presented as mean ± SD where n = 3. C) Relative cellular abundance of fully saturated fatty acids in stably transfected PARP1cd cell lines compared to parental 293 cells as determined by gas chromatography. The abundance of each fatty acid is normalized to total fatty acid content. D) Data are presented as mean ± SD where n ≥ 3. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 (Student's t-test).

but not indirect depletion of mitochondrial NAD + is detrimental to cellular homeostasis
A) Time-dependent appearance of the sum of labelled NAD + species (dashed line) and disappearance of unlabeled NAD + (solid line) in whole cell lysate (WCL, black) and mitochondria (red) upon incubation in the presence of 15N nicotinamide and 13C glucose. Delays are presented as the difference between mitochondrial and extramitochondrial (total minus mitochondrial) half-lives relative to the half-life of extramitochondrial NAD + . Experimental data where n ≥ 3. mP -293mitoPARP1cd, pP -293pexPARP1cd. B) Mitochondrial NAD + content in parental 293 cells and stably transfected 293mitoPARP1cd (mP) and 293 pexPARP1cd (pP) cell lines as determined by mitochondrial isolation followed by methanol extraction and LC-MS analysis. Results were normalized to protein content and data are represented relative to parental 293 cells as mean ± SD where n = 3. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001 (Student's ttest). C) Free mitochondrial NAD + concentration in untreated HeLa cells, as well as upon transient transfection with red mitoPARP1cd (mP) and pexPARP1cd (pP) constructs. NAD + concentrations were calculated from sensor (488/405 nm)/control (488/405 nm) fluorescence ratios determined by flow cytometry (see Figure S6). Measurements were performed in presence or absence of 3-aminobenzamide (3AB, 1 mM, 48h), and incubation of HeLa cells with FK866 (2 µM) served as a positive control. Results are represented relative to untransfected control cells as mean ± SD where n ≥3. D) The mitochondrial and cytosolic NAD + turnover per cell as calculated for parental 293 cells, as well as stably transfected transfected 293mitoPARP1cd (mP) and 293 pexPARP1cd (pP) cell lines taking into account the cellular NAD + distribution.   A) Immunoblot analysis of NMNAT3 expression in HAP1 wild type (WT) and NMNAT3 -/cells (NMNAT3 KO) cells. β-tubulin served as a loading control. B) PAR immunoblot analysis of lysates from HAP1 wild type (WT) and NMNAT3 -/cells (NMNAT3 KO) cells transiently transfected with mitoPARP1cd. Expression of the mitoPARP1cd fusion protein was confirmed by its intrinsic green fluorescence, while β-tubulin served as a loading control. C) Confocal fluorescence micrographs of HAP1 wild type (WT) and NMNAT3 -/cells (NMNAT3 KO) cells transiently transfected with mitoPARP1cd. Nuclei were stained with DAPI, the fusion protein was detected by its intrinsic green fluorescence, and visualization of PAR was achieved by indirect immunocytochemistry using the PAR (10H) antibody. Scale bar 20 µm. D) Determination of the total cellular NAD + content in HAP1 wild type (WT) and NMNAT3 -/cells (NMNAT3 KO) cells by methanol extraction and LC-MS analysis.
Results were normalized to protein content and data are represented relative to HAP1 WT cells as mean ± SD where n = 3. E) Schematic representation of a proposed model for the collaboration of SLC25A51 and NMNAT3 in situations of normal/high and low extramitochondrial NAD.