The oxidized form of nicotinamide adenine dinucleotide (NAD⁺) found in all living cells plays critical roles in a wide range of physiological processes. NAD⁺ acts as a coenzyme for enzymes which are involved in energy metabolism and homeostasis pathways such as glycolysis, TCA cycle, and oxidative phosphorylation (Campbell, 1995). NAD⁺ also acts as a co-substrate for enzymes such as sirtuins (SIRTs) and poly(ADP-ribose) polymerases (PARPs) to regulate a wide array of cellular processes such as survival/cell death, circadian clock, and aging (Imai and Guarente, 2014; Verdin, 2015). In mammals, cellular NAD⁺ is primarily biosynthesized by the NAD⁺ salvage pathway, in which nicotinamide phosphoribosyltransferase (NAMPT) is the rate-limiting enzyme and converts nicotinamide into nicotinamide mononucleotide (NMN), the NAD⁺ precursor (Revollo et al., 2007; Imai and Guarente, 2014; Nakahata and Bessho, 2016).

NAD⁺ amount declines during the aging process, causing defects in nuclear and mitochondrial functions and resulting in many age-associated diseases (Gomes et al., 2013; Mouchiroud et al., 2013). Intriguingly, administration of NAD⁺ precursors such as NMN and nicotinamide riboside (NR) can ameliorate many age-associated pathologies, thereby leading to healthy longevity (Yoshino et al., 2018; Custodero et al., 2020; Hong et al., 2020). Similar to organismal aging, NAD⁺ amount is also reported to decline with cellular aging, i.e., cellular senescence (Khaidizar et al., 2017). Overexpression of NAMPT in primary fibroblast cells delays the onset of replicative senescence (van der Veer et al., 2007; Khaidizar et al., 2017) and confers a protective effect against stress-induced premature senescence (Nuriliani et al., 2020). These reports demonstrate that a decline in NAD⁺ is one of the causes to induce cellular senescence as well as organismal aging.

NAD⁺ in cells and organs demonstrates circadian oscillation due to the circadian clock regulation of Nampt gene expression (Nakahata et al., 2009; Ramsey et al., 2009). Moreover, NAD⁺ is associated with the transcriptional regulation of circadian clock genes through the regulation of SIRT1, SIRT6, and PARP1 activities (Asher et al., 2008, 2010; Nakahata et al., 2008; Masri et al., 2014). These findings demonstrate that circadian clock and NAD⁺ metabolism are mutually regulated (Imai, 2010; Nakahata and Bessho, 2016). Since NAD⁺ amount declines with senescence/aging, circadian clock properties are expected to be altered by senescence/aging. Indeed, circadian clocks in primary cells, tissues, and animals show altered circadian properties by senescence/aging. The circadian amplitude at the transcriptional (Kunieda et al., 2006; Nakamura et al., 2015; Ahmed et al., 2019), neural activity (Nakamura et al., 2011), and locomotor activity levels (Pittendrigh and Daan, 1974; Witting et al., 1994; Valentinuzzi et al., 1997; Sellix et al., 2012) declines with aging. The period of circadian genes alters with senescence/aging in primary human fibroblasts and ex vivo mouse SCN slices (Nakamura et al., 2015; Ahmed et al., 2019), although it is still controversial at the organismal level (Valentinuzzi et al., 1997; Aujard et al., 2001; Kolker et al., 2004). However, only a few researches have looked into the molecular mechanisms of how the circadian clock is affected by senescence/aging (Chang and Guarente, 2014; Zwighaft et al., 2015; Levine et al., 2020).

We have reported that a decline in NAD⁺ keeps activating the transcription of E-box-regulated circadian clock genes via hyperacetylation of lysine 9/14 residues on histone H3, due to the inactivation of NAD⁺-dependent deacetylase, SIRT1 (Nakahata et al., 2008). In addition to that report, some investigators have also demonstrated that NAD⁺ regulates other circadian clock components: SIRT1 deacetylates PER2, resulting in its stabilization (Asher et al., 2008); SIRT1 deacetylates PGC1α to activate Bmal1 transcription (Chang and Guarente, 2014); PARP-1, an NAD⁺-dependent ADP-ribosyltransferase, poly(ADP-ribosyl)ates CLOCK (Asher et al., 2010); and the inhibitor of SIRT1 changes the ratio of PER2 subcellular localization (Miki et al., 2012).

The aforementioned findings prompted us to reveal the molecular mechanisms of how NAD⁺ amount, especially low NAD⁺, affects the circadian clock system, which might give some hints to unravel molecular links between aging and circadian clock. In this study, we reduced intracellular NAD⁺ amount by inhibiting NAMPT enzymatic activity and found that it attenuates the amplitude of Bmal1 promoter-driven luciferase oscillation and promotes cytoplasmic localization of PER2, leading to attenuated CRY/PER-dependent repression, thereby enhancing E-box-regulated circadian gene expressions. At the molecular level, we revealed that the translocation of PER2 with low NAD⁺ is dependent upon SIRT1 activity, but not PARP1 activity.

To investigate how the circadian clock system is affected by low NAD⁺ in cells, we first evaluated how much NAD⁺ was decreased by the inhibition of the rate-limiting enzyme, NAMPT, in the NAD⁺ salvage pathway (Figure 1A). We used FK866 (Hasmann and Schemainda, 2003) to inhibit the NAMPT activity and found that 2.0, 5.0, and 7.5 nM FK866 treatment decreased NAD⁺ to approximately 50, 33, and 25%, respectively (Figure 1B). In addition, the acetylated form of BMAL1, which has been reported to be increased by FK866 treatment (Nakahata et al., 2009), was increased by FK866 treatment in a dose-dependent manner (Supplementary Figure 2A). We further confirmed that the effect of FK866 treatment at 7.5 nM on NAD⁺ amount was abrogated by the co-treatment with the NAD⁺ precursor, NMN (Figure 1C). A similar tendency was observed by another NAMPT inhibitor, GMX1778 (Watson et al., 2009) (Supplementary Figures 3A,B). We then performed a real-time luciferase assay to address whether circadian clock properties are affected by low NAD⁺. To address that, we used NIH3T3 cells stably expressing Bmal1 promoter-driven luciferase (Yoshitane et al., 2012). The amplitude, which was defined in this study as the height from the first peak to the first trough, of Bmal1 promoter-driven luciferase oscillation was attenuated in FK866-treated cells, compared with that in control cells (Figures 1D,E). A delayed phase was also observed in FK866-treated Bmal1-luc oscillation. Intriguingly, the attenuated amplitude and delayed phase were abrogated in FK866 and NMN co-treated cells, indicating that the amplitude and phase are regulated by NAD⁺ amount. The reason why these alterations were abrogated after 60 h Dex treatment is presumably due to the recovery of intracellular NAD⁺ amount (Supplementary Figure 4). We further checked endogenous circadian gene expressions in unsynchronized cells with low NAD⁺. Among core circadian genes, E-box-regulated genes, such as Rev-erbα, Rev-erbβ, Per1, and Per2, were upregulated; however, the RORE-regulated gene, Bmal1, was downregulated (Figure 1F). Moreover, these alterations were abrogated in FK866 and NMN co-treated cells, indicating that the alterations of these genes are regulated by NAD⁺ amount. Furthermore, we demonstrated that circadian profiles of endogenous Bmal1 and Rev-erbα transcripts were also affected, lower Bmal1 and higher Rev-erbα, in synchronized cells with low NAD⁺ (Figure 1G). These results demonstrate that low NAD⁺ alters transcript levels of not only E-box-regulated genes as we have reported previously (Nakahata et al., 2008), but also RORE-regulated genes, which presumably regulate the amplitude of circadian clocks.

As our previous studies have demonstrated that the circadian gene expressions are epigenetically regulated by NAD⁺-dependent deacetylases, SIRT1 and SIRT6 (Nakahata et al., 2008; Masri et al., 2014), in this study, we sought to find out whether the properties of circadian clock proteins, such as protein stability and subcellular localization, are regulated by NAD⁺ amount. First, whether low NAD⁺ affects protein stability of clock proteins was investigated using cycloheximide, which is an inhibitor of protein biosynthesis due to its prevention in translational elongation. However, the stabilities of CLOCK, BMAL1, PER2, and CRY1 were not significantly affected by low NAD⁺ (Supplementary Figure 5). Next, whether low NAD⁺ affects subcellular localization of clock proteins was investigated (Figures 2A,B and Supplementary Figure 6). Among clock proteins tested in this study, subcellular localization of PER2 was clearly affected by FK866 treatment (Figure 2A). Cytoplasmic and nuclear PER2 under control condition were 18.8 ± 3.4 and 49.3 ± 4.2%, respectively, whereas cytoplasmic and nuclear PER2 under FK866-treated condition were 32.0 ± 7.9% (p = 5.1 × 10^–3, compared with cytoplasmic PER2 under control condition) and 34.5 ± 9.2% (p = 3.8 × 10^–3, compared with nuclear PER2 under control condition), respectively. Furthermore, GMX1778 treatment also changed the ratio of PER2 subcellular localization from 22.2 ± 0.5 to 26.3 ± 0.1% in the cytoplasm (p = 1.0 × 10^–4) and from 62.0 ± 1.9 to 50.0 ± 0.6% in the nucleus (p = 2.0 × 10^–4) (Supplementary Figure 7). Importantly, the effect of FK866 or GMX1778 on PER2 subcellular localization was abrogated when cells were treated with both FK866 and NMN or GMX1778 and NMN (Figure 2A and Supplementary Figure 7). These results indicate that PER2 subcellular localizations are controlled by intracellular NAD⁺ amount; lower NAD⁺ promotes the translocation of PER2 from the nucleus to the cytoplasm. Similarly, cytoplasmic and nuclear CRY1 under control condition were 6.3 ± 0.9 and 75.6 ± 3.9%, respectively, while cytoplasmic and nuclear CRY1 under FK866-treated condition were 10.6 ± 2.8% (p = 0.026, compared with cytoplasmic CRY1 under control condition) and 67.1 ± 8.5% (p = 0.12, compared with nuclear CRY1 under control condition), respectively (Figure 2B), indicating the similarity to the change in PER2 by FK866 treatment. These results prompted us to investigate whether the increase in PER2 cytoplasmic localization triggered by FK866 induces the increase in CRY1 cytoplasmic localization, because it is well known that PER2 determines CRY1 localization (Albrecht et al., 2007). Cytoplasmic localizations of CRY1 co-expressing PER2 with or without FK866 were 15.4 ± 0.7 and 10.8 ± 0.8% (p = 4.8 × 10^–3) and nuclear localizations were 71.8 ± 2.2 and 81.7 ± 1.3% (p = 8.6 × 10^–3), respectively (Figure 2C). The alteration of CRY1 subcellular localization by FK866 treatment was more when PER2 was co-expressed, suggesting that the alteration of PER2 subcellular localization leads to that of CRY1.

To confirm whether the changes in PER2 subcellular localization alter CRY1 subcellular localizations, PER2 wild type (WT), C-terminal-tagged nuclear localization signal (-NLS), or C-terminal-tagged nuclear export signal (-NES) (Akashi et al., 2014) was co-expressed with CRY1. Similar to a previous report (Akashi et al., 2014), nuclear localizations of PER2-WT, PER2-NLS, and PER2-NES without CRY1 were 59.1 ± 2.1, 73.6 ± 1.3, and 37.0 ± 4.7%, respectively (Supplementary Figure 8A). As expected, subcellular localizations of CRY1 were followed by those of PER2. Compared with the ratio of nuclear CRY1 co-expressing with PER2-WT (73.1 ± 1.6%), the ratio of nuclear CRY1 was increased (81.1 ± 0.7%, p = 0.011) when PER2-NLS was co-expressed, whereas the ratio of nuclear CRY1 was decreased (59.8 ± 2.2%, p = 0.008) when PER2-NES was co-expressed (Figure 2D). These results are consistent with previous reports that PER2 determines CRY1 localization (Albrecht et al., 2007). Even when CRY1 was co-expressed with these PER2s, PER2 subcellular localizations were not affected (Supplementary Figure 8B). The results of Figures 2A–D suggest that localization of PER2 is directly affected by intracellular NAD⁺ amount and the intracellular localization of CRY1 follows that of PER2.

Since PER2 protein amount and subcellular localization show time-of-day variations, whether circadian subcellular localization of PER2 is affected by NAD⁺ amount was next investigated (Figure 2E). Although nuclear localization of PER2 clearly showed the time-of-day variation under the control condition as reported previously (Lee et al., 2001; Yagita et al., 2001), the nuclear accumulation of PER2 was reduced, but still showed circadian oscillation, under the FK866-treated condition (Figure 2E, left panel). Intriguingly, PER2 in total cell lysate under the FK866-treated condition was comparable with that under the control condition (Figure 2E, right panels), suggesting that the reduction of PER2 nuclear localization under the FK866-treated condition is not due to the reduction of PER2, but the decrease in PER2 nuclear entry and/or the increase in PER2 nuclear export.

CRY acts as a transcriptional repressor with PER against CLOCK/BMAL1-dependent, namely E-box-regulated, transcriptions. As the decrease in nuclear localization of PER2 by FK866 suggested the attenuation of PER/CRY-dependent repression against E-box-regulated genes, we next addressed that possibility. To address that, luciferase reporter assay using Per1 promoter that possesses several E-boxes was assessed using PER2-NES, which mimics PER2 subcellular localizations under the FK866-treated condition. As reported previously, Per1 promoter activity was activated by CLOCK/BMAL1 and this activation was repressed by CRY1 (Figure 3). When PER2-NES was co-expressed, the repression by CRY1 against CLOCK/BMAL1-dependent Per1 expression was attenuated; namely, the luciferase intensities were increased in a PER2-NES dose-dependent manner (Figure 3). On the other hand, when PER2-WT or PER2-NLS was co-expressed, the repression by CRY1 was not affected. The results from Figures 2E, 3 indicate that PER2 subcellular localization determines PER/CRY-dependent repression potential, and support that the mechanism of upregulation of E-box-regulated genes under the FK866-treated condition (Figure 1C) is due to the preferential cytoplasmic localization of PER2 with CRY1.

Finally, to investigate the molecular mechanisms of how PER2 subcellular localization is regulated, we pharmacologically inhibited NAD⁺-dependent enzymes, SIRT1 and PARP1. The SIRT1 inhibitors, 10 mM nicotinamide (NAM) or 5 μM EX-527 (Bitterman et al., 2002; Napper et al., 2005), significantly increased cytoplasmic PER2 from 24.1 ± 1.3 to 39.5 ± 0.5% (p = 4.0 × 10^–4) and 33.5 ± 2.1% (p = 1.9 × 10^–4), respectively, and decreased nuclear PER2 from 42.7 ± 5.5 to 28.7 ± 3.8% (p = 3.0 × 10^–4) and 30.5 ± 4.2% (p = 1.8 × 10^–4), respectively (Figure 4A), showing the same tendency when treated by FK866 (Figure 2A). Both 10 mM NAM and 5 μM EX-527 were confirmed to be effective to inhibit SIRT1 deacetylate activity against BMAL1 (Nakahata et al., 2008; Supplementary Figures 2A,B). However, the PARP1 inhibitor, PJ34 (Abdelkarim et al., 2001), did not change both cytoplasmic and nuclear PER2 populations [17.3 ± 3.8% (p = 0.56) and 54.6 ± 3.3% (p = 0.10), respectively] (Figure 4), and even 10 μM PJ34 was confirmed to be effective to reduce poly-ADP-ribosylation (Supplementary Figure 2C). As FK866 treatment attenuated the amplitude of Bmal1-luc circadian oscillation (Figures 1D,E), we further investigated whether SIRT1 deacetylation activity is involved in the regulation of amplitude of Bmal1-luc circadian oscillation. NAM or EX-527 treatment attenuated the amplitude of Bmal1-luc circadian oscillation (Figures 4B,C,E). Intriguingly, a delayed phase was observed only in cells treated with 10 mM NAM, the concentration at which the enzymatic activity of SIRT1 and also others such as SIRT2 and PARP1 is inhibited (Virág and Szabó, 2002; Peck et al., 2010). However, a delayed phase was not observed in cells treated with 5 μM EX-527, the concentration at which the enzymatic activity of SIRT1 is specifically inhibited (Peck et al., 2010), suggesting that SIRT1 deacetylation activity is involved in the amplitude regulation and other enzymes inhibited by NAM are involved in the phase regulation. Sirt1-knockdown cells also demonstrated the attenuated Bmal1-luc amplitude, but not a delayed phase (Figure 4D). Knockdown efficiencies against Sirt1 and circadian oscillation patterns of other Sirt1 knockdown cell lines are shown in Supplementary Figure 9. These results demonstrate that the NAD⁺/SIRT1 axis regulates PER2 subcellular localization, which further affects the amplitude of the circadian clock.

To further investigate whether the acetylation status of lysine 680 residue of PER2, where PER2 is acetylated and deacetylated by SIRT1 (Levine et al., 2020), affects its subcellular localization, we analyzed subcellular localizations of lysine 680 residue mutants of PER2 (Figure 5). Non-acetyl mimetic PER2^K680R mutant only slightly increased in the cytoplasmic population of PER2, compared with the cytoplasmic population of PER2^WT (p = 0.010). Moreover, acetyl mimetic PER2^K680Q had no effect on PER2 subcellular localization. These mutant PER2 experiments suggest that K680 of PER2 is not the responsible lysine to regulate subcellular localization by SIRT1, although we cannot deny the possibility that acetyl mimetic mutant of PER2^K680Q does not completely substitute the functions of acetyl-K680 of PER2.

In this study, we found that low intracellular NAD⁺, which was induced by the pharmacological inhibition of NAMPT, promoted PER2 subcellular localization from the nucleus to the cytoplasm and increased and decreased the expressions of E-box- and RORE-regulated circadian genes, respectively, leading to the attenuated Bmal1-luc circadian oscillation. Furthermore, we demonstrated that SIRT1 was responsible for NAD⁺-dependent PER2 subcellular regulation, and suggested that lysine 680 residue on PER2 is not responsible for SIRT1-regulated PER2 subcellular localization.

Since several lines of evidence indicate that a decline in NAD⁺ is a hallmark of senescence/aging (Braidy et al., 2011; Yoshino et al., 2011; Canto et al., 2012; Massudi et al., 2012; Camacho-Pereira et al., 2016; Zhang et al., 2016; Khaidizar et al., 2017), in this study, we used NAMPT inhibitors, FK866 and GMX1778, to reduce intracellular NAD⁺, which, we expect, mimics cellular senescence. However, many other physiological events and metabolites such as AMP/ATP ratio and polyamines are known to be changed with senescence/aging (James et al., 2015; Zwighaft et al., 2015). An increase in the AMP/ATP ratio promotes AMP-activated protein kinase (AMPK), which acts as a sensor of the reduced energetic state and further activates catabolic pathways while inhibiting anabolic ones (Hardie, 2003; Garcia and Shaw, 2017). Meanwhile, it has been reported that mTOR, which is an intracellular nutrient sensor for high cellular energy state and associated with autophagy, is also upregulated during senescence (Herranz et al., 2015; Laberge et al., 2015; Nacarelli and Sell, 2017). These signaling molecules have been reported to be involved in the circadian clock. AMPK is a rhythmically expressed kinase and phosphorylates CK1ε, resulting in enhanced phosphorylation and degradation of PER2 (Um et al., 2007; Sahar and Sassone-Corsi, 2012) and CRY1 (Lamia et al., 2011; Sahar and Sassone-Corsi, 2012; Jordan and Lamia, 2013). mTOR perturbation, such as RNAi knockdown or mTOR inhibitors, alters circadian rhythms in fibroblast, SCN, and animal behaviors (Zhang et al., 2009; Ramanathan et al., 2018). Several lines of evidence mentioned here imply that the altered aforementioned signaling pathways during senescence may affect circadian clock properties, although it is largely unknown whether these pathways alter circadian clock properties in senescent cells. We recently demonstrated that altered circadian properties such as period extension and phase delay occur in senescent human cells (Ahmed et al., 2019, 2021). Therefore, we assume that these circadian alterations in senescent cells will be useful indexes to evaluate whether low NAD⁺ or other conditions are sufficient for the circadian clock to mimic senescent condition.

Real-time luciferase monitoring assay revealed that the attenuated amplitude of Bmal1-luc oscillations was observed by either NAMPT inhibitors, FK866 and GMX1778; Sirtuins and PARPs broad inhibitor, NAM; SIRT1-specific inhibitor, EX-527; or Sirt1 knockdown. However, the delayed phase of Bmal1-luc oscillations was not observed by SIRT1-specific inhibitor, EX-527, or Sirt1 knockdown. These results indicate that SIRT1 deacetylation activity is involved in amplitude regulation, but not in phase regulation. This is supported by the report that circadian profiles of Dbp and Per2 in Sirt1^–/– cells showed higher amplitude and no delayed phase (Nakahata et al., 2008). In contrast, delayed phases were observed with FK866, GMX1778, or NAM treatment, suggesting that NAD⁺-dependent enzymes other than SIRT1 are involved in the phase regulation. The investigation to address which enzyme is associated with phase regulation will be needed to understand the mechanisms of how circadian clocks work in senescent cells. Actually, we have recently shown that senescent cells possess altered circadian clocks with a prolonged period and delayed phase (Ahmed et al., 2019, 2021).

In this study, we investigated whether acetylation of lysine 680 residue on PER2 is responsible for PER2 subcellular localization; however, neither acetyl mimetic mutant (K680Q) nor non-acetyl mimetic mutant (K680R) of PER2 changed subcellular localization such as low NAD⁺ or SIRT1 inhibitor-treated conditions. Levine and coworkers recently reported that PER2 is acetylated on K680, which is deacetylated by SRIT1; furthermore, FK866 treatment increases PER2 phosphorylation at numerous sites (Levine et al., 2020). This suggests that acetylation of K680 or other lysine residues of PER2 might be a trigger for subsequent phosphorylation(s), which might determine PER2 subcellular localization. Our current results cannot deny the possibility that acetyl mimetic mutant PER2^K680Q may not trigger subsequent phosphorylation(s) because of differences of three-dimensional structures triggered by the acetylated form of lysine residue and substituted glutamine residue. Further experiments including structural biological experiments will be needed to elucidate which lysine residue is responsible for subcellular localization of PER2 regulated by SIRT1.

Our current results suggest that PER/CRY-dependent repression against E-box-regulated genes might be dependent on the NAD⁺/SIRT1 axis. Intriguingly, we have reported that SIRT1 deacetylates lysine 9/14 residues of histone H3 on the promoters of E-box-regulated genes to repress their expressions epigenetically (Nakahata et al., 2008). These results imply that the NAD⁺/SIRT1 axis coordinates the transcriptional repression of E-box-regulated genes by different ways. In addition to the coordinated mechanisms for E-box-regulated genes, RORE-regulated genes might also be coordinated by the NAD⁺/SIRT1 axis. It has been reported that NAD⁺ activates Bmal1 transcription by SIRT1 through the deacetylation of PGC1α (Chang and Guarente, 2014), while we demonstrated in this study that the low NAD⁺ condition increased Reb-erbα/β which in turn might repress Bmal1 gene expression. These coordinated Bmal1 gene regulations by the NAD⁺/SIRT1 axis presumably regulate the amplitude of Bmal1 circadian oscillation, namely the low NAD⁺ condition or decreased SIRT1 activity attenuates the amplitude of Bmal1 circadian oscillation. Therefore, these coordinated transcriptional regulations of E-box- and RORE-regulated circadian genes by the NAD⁺/SIRT1 axis might interlock circadian negative feedback loops.

In addition to our finding that low NAD⁺ retains PER2 in the cytoplasm, some studies have demonstrated that the NAD⁺/SIRT1 axis regulates PER2 properties; PER2 in Sirt1^–/– cells is more stable (Asher et al., 2008) and localizes predominantly in the nucleus (Levine et al., 2020), while PER2 in nicotinamide-treated cells localizes predominantly in the cytoplasm (Miki et al., 2012). Our pharmacologically conducted results were consistent with the results pharmacologically performed by Miki and colleagues, but not with the results performed genetically. These conflicting results by pharmacological (acute) and genetical (chronic) approaches suggest the possibility that responses of molecular clock are different at early and chronic senescence/aging phases. This possibility is worth verifying so further experiments using different stages of senescent cells/aging animals will be needed to reveal this possibility.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

AA performed and analyzed the experiments. YN designed the research and wrote the manuscript. TS, YuF, HY, and YoF provided the reagents. KS assisted with the experiments and reviewed the manuscript. TM and YB designed the research and reviewed the manuscript. All authors contributed to the article and approved the submitted version.

TS and YF were employed by the Mitsubishi Corporation Life Sciences Limited. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.