Ubiquitin-dependent rapid degradation conceals a cell-protective function of cytoplasmic SIRT3 against oxidative stress

. translocation of SIRT3ct and induced histone H3 deacetylation and superoxide dismutase 2 expression. The overexpression of SIRT3ct decreased cell death by H 2 O 2 at similar levels achieved by that of SIRT3mt. Knockdown of Sirt3 mRNA increased cell death by amyloid-  (A  ) and the overexpression of SIRT3ct opposed the toxic function of A  in PC12 cells. These results indicated that SIRT3ct participated in cell survival under various stress conditions. of expressing hrGFP with condensed or normal counted from 5 of each treatment, and the percentage of dead cells was represented. Three independent experiments were carried out.


Introduction
SIRT3, a member of the sirtuin family, deacetylates acetylated proteins with NAD + resulting in deacetylated proteins, O-acetyl-ADP-ribose and nicotinamide 1 . Among the proteins of the sirtuin family, SIRT1 and SIRT2 localize both in the cytoplasm and nucleus. SIRT3, SIRT4 and SIRT5 are mitochondrial proteins, whereas SIRT6 and SIRT7 are nuclear proteins 1 . The human SIRT3 protein contains a mitochondrial targeting sequence in its NH2-terminal region and is localized in the mitochondrial matrix 2 . SIRT3 regulates various mitochondrial proteins and participates in stress resistance, oxidative phosphorylation, lipolysis and urea formation. SIRT3 deacetylates and activates superoxide dismutase 2 (SOD2) and isocitrate dehydrogenase 2 in mitochondria, resulting in a decrease of superoxide anion and an increase in glutathione levels 3,4 . SIRT3 has been shown to ameliorate various pathological conditions. In the heart, SIRT3 has a protective role in cardiac ischemia, hypertrophy and fibrosis 5 . In the brain, SIRT3 is involved in the adaptive antioxidant response and prevents neuronal death driven by wakefulness in locus ceruleus 6 and in the mouse model of epilepsy and Huntington's disease 7 . Although most studies show that SIRT3 is exclusively localized in mitochondria, several reports have argued cytoplasmic and/or nuclear localization of SIRT3 [8][9][10][11][12][13] . Scher et al. showed that two isoforms of SIRT3 existed and full length and 4 processed forms of SIRT3 localized to the nucleus and mitochondria, respectively 8 . In the nucleus, SIRT3 could deacetylate histones and was recruited into the mitochondria upon cell stress 8  reported that SIRT3 localized in the mitochondria, nucleus and cytoplasm in the brain 13 .
Thus, there is no doubt about the localization and functions of mitochondrial SIRT3, but that of cytoplasmic or nuclear SIRT3 are still obscure.
In this study, we found that Sirt3 mRNA for cytoplasmic SIRT3 (SIRT3ct) was highly expressed in the brain but the SIRT3ct protein level was much lower than that of SIRT3mt. We found that SIRT3ct was rapidly degraded by a unique ubiquitindependent pathway, which might be regulated by cellular stress. SIRT3ct promoted cell survival against oxidative stress and amyloid- (A). Our study suggested a latent and important role of SIRT3ct, which became manifest under stress conditions in the brain.

Comparison of mitochondrial and cytoplasmic SIRT3.
Five types of mouse Sirt3 mRNAs have been reported in the NCBI Reference Sequence Database and three forms, Sirt3mt, Sirt3ct1 and Sirt3ct2, are assumed to encode active enzymes as shown in Fig. 1a. These Sirt3 mRNAs are generated by alternative splicing of the mouse SIRT3 gene. We screened 6 x 10 5 phage clones of a mouse brain cDNA library and obtained seventeen SIRT3 cDNAs. Among them, two clones were cDNAs for Sirt3mt (NM_001177804.1) that encoded the SIRT3mt protein of 334 amino acids with a putative mitochondrial targeting sequence (Fig. 1a-c). The remaining fifteen cDNA clones were identical to Sirt3ct1 (NM_001127351.1) which encoded SIRT3ct protein of 257 amino acids without a mitochondrial targeting sequence (Fig. 1a,c). We could not detect a cDNA clone for Sirt3ct2 (NM_022433.2) in the brain. The NCBI reference sequences show that human SIRT3 proteins consist of two isoforms depicted as human SIRT3mt and human SIRT3ct (Fig. 1c). The NH2-terminal sequence corresponding to residues 1-25 of human SIRT3mt is required for mitochondrial import of the protein 2 . Amino acid residues 15-35 of mouse SIRT3mt had high similarity to the NH2-terminal mitochondrial targeting sequence of human SIRT3mt (Fig. 1b,c) and contained a cluster of charged arginine residues opposed by a hydrophobic cluster in a helical wheel plot (right side of Fig. 1b), suggesting that this region had an amphipathic- 6 -helix, a feature of mitochondrial targeting sequences 2 . During translocation of mitochondrial proteins from cytoplasm to mitochondria, mitochondrial targeting sequences are cleaved by mitochondrial processing peptidase (MPP). MPP cuts mitochondrial targeting sequences which contain arginine residues at positions -10 and -2 and/or -3 from the cleavage site 14 . The targeting sequences also have a proline residue between the distal and proximal arginine residues 14 . Arginine residues at positions 84, 99, and 100, and proline residue at 88 of human SIRT3mt are consistent with the consensus amino acid sequence of mitochondrial targeting sequences and arginine residues at 99 and 100 of human SIRT3mt are shown to be necessary for the cleavage by MPP, which produces 28 kDa human SIRT3mt in mitochondria 2 . Mouse SIRT3mt had arginine residues at positions of 27, 35, and 36 and a proline residue at position 31, which matched with consensus mitochondrial targeting sequences, suggesting that the amino acid sequence of 1-37 was the mitochondrial targeting sequence of mouse SIRT3mt (Fig. 1b,c).
To examine the subcellular localization of mouse SIRT3mt and SIRT3ct, SIRT3mt and SIRT3ct were tagged with EGFP in their COOH-terminal regions and expressed in COS7 cells, in which mitochondria were stained with MitoTracker Red.
SIRT3mt-EGFP mainly localized in the mitochondria, whereas SIRT3ct-EGFP was 7 found in the cytoplasm (Fig. 1d). SIRT3mt-EGFP was also found in the extramitochondrial cytoplasmic region where SIRT3mt-EGFP showed a dot-like appearance ( Fig. 1d). Interestingly, SIRT3ct-EGFP frequently showed a dot-like appearance in the cytoplasm (Fig. 1d). To further investigate the localization of SIRT3mt and SIRT3ct, the two isoforms were fused with a FLAG-tag at COOH-terminal ends and expressed in COS7 cells, and cells were then homogenized and fractionated into nuclear (Nu), cytoplasmic (Ct) and mitochondrial fractions (Mt). Lamin A/C, cytoplasmic aspartate aminotransferase (cAST) and mitochondrial aspartate aminotransferase (mAST) are nuclear, cytoplasmic and mitochondrial marker proteins, respectively. Immunoblot analysis showed that expression of SIRT3mt-FLAG cDNA gave rise to three bands of SIRT3mt-FLAG proteins (37, 33, and 32 kDa), which were dominantly detected in the mitochondrial fraction and also in the cytoplasmic fraction (Fig. 1e). The intensity of the band of the smallest SIRT3mt (32 kDa) was higher in the mitochondrial fraction than that of the cytoplasmic fraction (Fig. 1e). A small amount of 32 kDa SIRT3mt-FLAG was detected in the nuclear fraction (Fig. 1e). SIRT3ct-FLAG was present in the cytoplasmic fraction and its molecular weight was estimated to be 28 kDa (Fig. 1e). A small amount of SIRT3ct-FLAG was also detected in the nuclear and mitochondrial fractions (Fig. 1e). Although the same DNA amounts of SIRT3ct-FLAG and SIRT3mt- 8 FLAG cDNAs were transfected in COS7 cells, the intensity of SIRT3ct-FLAG bands in immunoblotting was always much weaker than those of SIRT3mt-FLAG (Fig. 1e).
SIRT3 is involved in stress resistance of cells 1,5 . We examined the effect of staurosporine (SP), a non-genotoxic stressor, on the expression and distribution of SIRT3mt-FLAG and SIRT3ct-FLAG. Homogenates of COS7 cells transfected with the same amount of cDNA for SIRT3mt-FLAG or SIRT3ct-FLAG were fractionated into the nuclear fraction (Nu) and the extra-nuclear fraction (En) containing cytoplasm and mitochondria (Fig. 1f). SP treatment enhanced the expression levels of SIRT3mt-FLAG proteins of 37 kDa and 33 kDa in the nuclear and extra-nuclear fractions, whereas the intensity of the smallest SIRT3mt-FLAG of 32 kDa was not affected by SP treatment (Fig. 1f). SP treatment significantly increased the SIRT3ct-FLAG expression level in the cytoplasm (Fig. 1f). A small amount of SIRT3ct-FLAG band was also detected in the nuclear fraction after SP treatment (Fig. 1f).

Expression of SIRT3 in the brain and neural precursor cells.
We examined the expression of Sirt3 mRNA in various organs. Northern blot analysis showed that Sirt3 mRNA was highly expressed in the heart, brain, kidney and embryo ( Fig. 2a) and was constantly expressed in the mouse brain of various developmental stages and ages (Fig. 2b). Developmental change of the expression of Sirt3 mRNA in the 9 fetuses showed that Sirt3 mRNA levels were high in the embryonic day (E) 4.5 to E6.5 and then reduced. In the late embryonic stages, Sirt3 mRNA levels increased again and the highest expression levels were found at E16.5-E18.5 (Fig. 2c). RT-PCR experiments using the fetal and post-natal brain also showed higher expression levels of Sirt3 mRNA at E14.5 and P2 compared with that of E10.5 (Fig. 2d). Immunostaining of SIRT3 in the mouse brain showed that high levels of SIRT3 expression were found in the fetal brain and P2 mouse brain where cells around cerebral ventricles expressed high levels of SIRT3 (Fig. 2e). SIRT1, another NAD + -dependent protein deacetylase, is highly expressed in neural precursor cells (NPCs) and participates in neuronal differentiation 15 . To examine whether NPCs express SIRT3, neurospheres were cultured from the striatum of the E14.5 mouse brain and were immunostained with an anti-SIRT3 antibody. Similar to SIRT1, SIRT3 was highly expressed in NPCs and almost all of the SIRT3 + cells expressed nestin, a marker of NPCs (Fig. 2f). Costaining of SIRT3 with nestin indicated that SIRT3 was diffusely expressed in the cytoplasm but not in the nucleus.
SIRT3 protein expression was also examined in the adult mouse brain.
Immunostaining showed that SIRT3 was highly expressed in some subsets of neural cells in such as the hippocampus, cerebral cortex, cerebellum and in the ependyma (Fig.   10 2g). To confirm the expression of SIRT3 in neurons, cortical neurons were isolated from newborn mouse brains and cultured. The cells were immunostained with an anti-SIRT3 antibody and an anti-III-tubulin antibody, a marker of neural cells.
Mitochondria were stained with MitoTracker Red. SIRT3 distributed diffusely in cortical neurons, in which SIRT3 was detected not only in the mitochondria but also in the cytoplasm (Fig. 2h). Cellular protrusions were also stained by the SIRT3 antibody ( Fig. 2h).

Knockdown of SIRT3 disturbed differentiation of neural precursor cells.
SIRT3 is highly expressed in the subventricular zone of the brain and in cultured NPCs ( Fig. 2e,f). To examine the distribution of SIRT3 in NPCs, we dissociated NPCs of neurospheres and immunostained them with antibodies against SIRT3 and cytochrome C, a marker of mitochondria. Diffuse cytoplasmic immunostaining of SIRT3 suggested that SIRT3 protein localized in both the cytoplasm and mitochondria (Fig. 3a). In the proliferation condition, neurospheres were round and highly expressed SIRT3 and SIRT1 (Fig. 3b). Expression of βIII-tubulin, a marker of neuronal cells, was not detected (Fig. 3b). Neurospheres can differentiate into neurons, oligodendrocytes and astrocytes under differentiation condition. When neurospheres were cultured in the differentiation medium for 5 days, they became flat and differentiated cells were 11 scattered around spheres. Under differentiation condition, expression of SIRT3 and SIRT1 was suppressed and βIII-tubulin + cells appeared (Fig. 3b), suggesting that SIRT3 might have some effect on undifferentiated neurospheres.
Because SIRT3 is involved in a cell protective mechanism against oxidative stress, we examined the effect of this stress on the differentiation of neurospheres using antimycin A (AA), an inhibitor of mitochondrial respiratory chain complex III and a ROS inducer. Ten nM or 100 nM AA was added to the differentiation medium and NPCs were cultured in it for 5 days. In the absence of AA, more than 90% of the neurospheres were spreading after 5 days in the differentiation medium indicating that differentiated cells migrated and extended protrusions. AA treatment dose-dependently inhibited spreading of neurospheres after differentiation (Fig. 3c). As shown in Fig. 3d, immunostaining of differentiated neurospheres gave rise to βIII-tubulin + , O4 + and GFAP + cells, indicating that spheres differentiated into neurons, oligodendrocytes, and astrocytes, respectively ( Fig. 3d; AA (-)). Treatment of NPCs with 10 nM AA in the differentiation medium for 5 days faintly affected cell viability but reduced the number of βIII-tubulin + cells and O4 + cells after differentiation (Fig. 3d). Ten nM AA also suppressed extension of protrusions of βIII-tubulin + cells and O4 + cells (Fig. 3d).
However, the number and shape of GFAP + cells were not affected by 10 nM AA (

Ubiquitin-dependent degeneration of SIRT3ct protein.
SIRT3ct protein expression levels were always weaker than those of SIRT3mt protein in the Western blot experiments (Fig. 1e,f). Low levels of SIRT3ct protein expression may be mediated by accelerated degradation of the protein in cells. Anisomycin is an inhibitor of protein synthesis. When protein synthesis was inhibited by anisomycin in COS7 cells 48 h after transfection of SIRT3ct-FLAG cDNA, the SIRT3ct-FLAG protein could not be detected after 1 h incubation with the drug. In contrast, SIRT3mt expression was not affected even after 2 h incubation with anisomycin (Fig. 4a). The SIRT3ct protein might be fragile and rapidly degraded by cellular proteases. We examined the effects of various protease inhibitors on SIRT3ct-FLAG protein levels in 13 HEK293T cells. Treatment of cells with lactacystin, a proteasome inhibitor, increased the intensity of the SIRT3ct-FLAG band (Fig. 4b), whereas phenylmethylsulfonyl fluoride (PMSF) and leupeptin, inhibitors of serine, cysteine and/or threonine proteases, could not prevent degradation of SIRT3ct-FLAG (Fig. 4b). Pepstatin A, an inhibitor of acid proteases, failed to protect SIRT3ct-FLAG (Fig. 4b). A potent proteasome inhibitor MG132 significantly increased the expression level of SIRT3ct-FLAG but faintly affected the SIRT3mt-FLAG level (Fig. 4c). Since lactacystin and MG132 are proteasome inhibitors, SIRT3ct-FLAG may be degraded by a ubiquitin-dependent pathway. COS7 cells transfected with SIRT3ct-EGFP cDNA were immunostained with an anti-ubiquitin antibody. As shown in Fig. 4d, colocalization of SIRT3ct-EGFP with ubiquitin suggested that SIRT3ct was polyubiquitinated in cells. To detect ubiquitination of SIRT3ct, SIRT3ct-FLAG expressed in COS7 cells was immunoprecipitated with an anti-FLAG antibody and then analysed by an anti-ubiquitin antibody. Multiple bands larger than 28 kDa of SIRT3ct-FLAG were detected, when cells were treated with MG132 (Fig. 4e). Thus, SIRT3ct-FLAG was ubiquitinated. The expression levels of SIRT3ct-FLAG and SIRT3ct-EGFP were faint in the absence of MG132 and were significantly enhanced by the treatment of cells with MG132 ( Fig.   4g). SIRT3ct-FLAG and SIRT3ct-EGFP were proteins in which FLAG peptide and 14 EGFP were fused in the COOH-terminal end of SIRT3ct. Ubiquitin-dependent protein degradation was mediated by the ubiquitination of lysine residues or the NH2-terminal methionine residue of target proteins 16 . Because ubiquitination usually occurs in the amino group of lysine residues, we changed all of the 8 lysine residues of SIRT3ct to arginine residues in SIRT3ct-8KR-FLAG by site-directed mutagenesis ( Supplementary   Fig. 1). The expression level of SIRT3ct-8KR-FLAG was higher than that of SIRT3ct-FLAG, but it was prominently increased by MG132 treatment, indicating that SIRT3ct-8KR-FLAG was still degraded by a ubiquitin-proteasome system (Fig. 4f). Because ubiquitination of the NH2-terminal methionine residue of ERK3 is inhibited by a fusion of EGFP at the NH2-terminal end 17 , we examined the effect of EGFP fusion to SIRT3ct-FLAG on the stability of SIRT3ct. EGFP-SIRT3ct-FLAG was the protein to which EGFP was fused at the NH2-terminal end of SIRT3ct-FLAG. We found that the expression of EGFP-SIRT3ct-FLAG was easily detected in the absence of MG132 (Fig.   4g), suggesting that the NH2-terminal methionine of SIRT3ct was a ubiquitination site.
Treatment of cells expressing EGFP-SIRT3ct-FLAG with MG132 slightly enhanced the expression level of the protein compared with that in non-treated cells (Fig. 4g). When EGFP was introduced in the NH2-terminal region of SIRT3ct-8KR-FLAG, the intensity of the band of EGFP-SIRT3ct-8KR-FLAG was not enhanced by MG132 (Fig. 4h). In 15 contrast, the expression level of EGFP-SIRT3ct-FLAG was slightly increased by MG132 treatment (Fig. 4h). These results suggested that the ubiquitination of the NH2terminal methionine residue mainly participated in the degradation of SIRT3ct and those in lysine residues of SIRT3ct partially contributed to SIRT3ct degradation.

SIRT3ct as well as SIRT3mt protected cells against oxidative stress.
SIRT3 plays a role in the suppression of reactive oxygen species (ROS) 1,5 . Stress inducer SP enhanced SIRT3ct expression in the cytoplasm and nucleus (Fig. 1f). We examined the effect of H2O2 on the expression and subcellular localization of SIRT3 in pheochromocytoma PC12 cells. Sirt3 mRNA expression levels were increased by H2O2 treatment (Fig. 5a). Immunostaining of endogenous SIRT3 in PC12 cells showed increased expression levels of SIRT3 by H2O2 treatment (Fig. 5b).
FOXO4, a transcription factor, has a cell protective role in response to oxidative stress and induces ROS-detoxifying enzymes such as SOD2 18 . Treatment of cells with H2O2 enhanced the expression of SIRT3 and induced nuclear FOXO4 expression. Moreover, some cells showed nuclear costaining of SIRT3 and FOXO4 (Fig. 5b). Treatment of cells with AA also enhanced the expression of SIRT3 (Fig. 5b).
Expression of SIRT3ct-EGFP in COS7 cells showed cytoplasmic localization of SIRT3ct-EGFP, where SIRT3ct-EGFP presented a dot-like appearance (Fig. 5c, bottom 16 left panel). Compared to that of untreated cells, treatment of cells with H2O2 significantly enhanced the fluorescence intensity of SIRT3ct-EGFP in the cytoplasm of cells where SIRT3ct-EGFP was uniformly distributed (Fig. 5c, bottom right panel). We found that SIRT3ct-EGFP fluorescence was also detected in the nuclei of cells treated with H2O2 (Fig. 5c, bottom right panel). H2O2 treatment slightly enhanced the fluorescence level of SIRT3mt-EGFP in COS7 cells and was also induced nuclear localization of SIRT3mt-EGFP (Fig. 5c, upper panels). Nuclear localization and an increase of SIRT3 immunostaining by H2O2 were detected in the primary cultured neuronal cells isolated from the newborn mouse brain (Fig. 5d). More than 40% of cells showed nuclear SIRT3 immunostaining by H2O2 stimulation (Fig. 5d). Since SIRT3 has been reported to deacetylate histones 8 , it may also be possible that nuclear localization of SIRT3ct and SIRT3mt by H2O2 induces deacetylation of histone H3. Accordingly, in the presence of H2O2, overexpression of SIRT3ct-FLAG or SIRT3mt-FLAG significantly induced the deacetylation of histone H3 in COS7 cells (Fig. 5e).
Overexpression of SIRT3 increases SOD2 expression level by deacetylation and activation of FOXO 19 . Using PC12 cells, we found that SOD2 immunostaining was enhanced by SIRT3ct-EGFP expression (Fig. 5f). These observations suggested that SIRT3ct stability was increased by oxidative stress, and that SIRT3ct translocated into the nucleus and modulated gene expression.
SIRT3 has been shown to decrease ROS levels and inhibit cell death by ROS 1,5 . We examined the effect of the overexpression of SIRT3ct on H2O2-induced cell death and compared the result with that of SIRT3mt in COS7 cells. To identify transfected cells, we co-expressed the GFP protein with SIRT3ct or SIRT3mt. Using a pIRES-hrGFP vector, GFP was expressed separately from SIRT3 to eliminate an artificial effect that could result from the fusion of GFP to SIRT3. We found that the overexpression of SIRT3ct significantly reduced the number of dead cells under H2O2 treatment (Fig. 5g). Similar to the anti-apoptotic effect of SIRT3ct, the overexpression of SIRT3mt also decreased cell death induced by H2O2 (Fig. 5g). The potency of the cell-survival effect of SIRT3ct and that of SIRT3mt was similar under the present condition. Both SIRT3ct and SIRT3mt could participate in a cellular protective function under an oxidative stress condition.

SIRT3ct overexpression in PC12 cells.
Extracellular accumulation of A peptides is a pathological hallmark of Alzheimer's disease 20 . A peptides induce neuronal cell death via various pathways including 18 increase of ROS levels 20 . A peptide 25-35 (A25-35) is shown to increase ROS level and cell death in PC12 cells 21 . To examine whether SIRT3 participates in a cell protective role against A peptides, we examined the effect of SIRT3 knockdown on cell death induced by A25-35 treatment in PC12 cells. Both Sirt3-siRNA1 and Sirt3-siRNA2 significantly reduced the expression levels of Sirt3 mRNA and the SIRT3 protein in PC12 cells (Fig. 6 a,b). Knockdown of SIRT3 with Sirt3-siRNA2 alone increased the number of dead cells, which were significantly increased by Aβ25-35 treatment (Fig. 6c). In the presence of Aβ25-35, SIRT3 knockdown increased the number of dead cells by 2.5-fold compared with those transfected with control-siRNA (Fig. 6c).
To examine whether overexpression of SIRT3ct might rescue PC12 cells exposed to  (Fig. 6d). 19 Our experiments using cDNA cloning suggested that Sirt3ct mRNA was highly expressed compared to Sirt3mt mRNA in the brain, but SIRT3ct protein expression was much less than that of SIRT3mt. We found that SIRT3ct was the target of ubiquitindependent degradation and that the majority of SIRT3ct was degraded through ubiquitination of the NH2-terminal methionine residue. Proteins such as MyoD, p21 and ERK3 are degraded by the NH2-terminal ubiquitination, which can escape through a fusion of large tags or EGFP at the NH2-terminal end 16,17,22 . Fusion of EGFP into SIRT3ct at the NH2-terminal region inhibited SIRT3ct degradation (Fig. 4f-h). Akimov SIRT3mt was resistant against proteasome degradation, since proteasome inhibitor MG132 had little effect on the stability of SIRT3mt (Fig. 4c). Because the second amino acid residue of mouse and human SIRT3mt was alanine (Fig. 1b), the amino 20 terminal end of SIRT3mt might have been acetylated and able to escape from protein degradation. Although many proteins are ubiquitinated at the NH2-terminal end, the stability of these proteins is usually not affected by the ubiquitination 23 . Similar to p21, ERK3 and MyoD, SIRT3ct was an exception to the rule. p21 and ERK3 are degraded by the NH2-terminal ubiquitination but not by the ubiquitination of lysine residues, and only MyoD is degraded by both the ubiquitination at NH2-terminal methionine and lysine residues 16,17,22 . Since MyoD is a short-lived protein with a half-life of 30-60 min 24 , ubiquitin-dependent degradation through NH2-terminal methionine and lysine residues of SIRT3ct may result in the short lifespan of SIRT3ct (Fig. 4a). At present, it is unknown whether the short half-life of SIRT3ct has any physiological significance.

Discussion
SIRT3ct was promptly degraded in a control condition but we showed that a nongenotoxic stressor SP enhanced the expression of SIRT3ct in the cytoplasm and in the nucleus (Fig. 1f). The immunofluorescence level of SIRT3ct-EGFP was enhanced by H2O2 treatment (Fig. 5c) and immunostaining also showed an increased level of endogenous SIRT3 by oxidative stress in PC12 cells and in cultured neural cells (Fig. 5 b,d). Although Sirt3 mRNA expression was enhanced by oxidative stress (Fig. 5a), enhancement of the SIRT3ct level expressed with SIRT3 cDNA by H2O2 treatment (Fig. 5c) suggested that ubiquitin-dependent degradation of SIRT3ct might be regulated 21 by cellular stress. However, further study is needed to clarify the effect of stress signals on SIRT3ct stability. SIRT3ct reduced cell death induced by oxidative stress, in which the cell survival effect of SIRT3ct was comparable with that of SIRT3mt in COS7 cells (Fig. 5g). These experiments showed that SIRT3ct could participate in cell survival under stress condition, which was consistent with the role of SIRT3mt on cellular  (Fig. 1e,f and 4c).
SIRT3ct and SIRT3mt dominantly localized in cytoplasm and mitochondria, respectively. However both SIRT3ct and SIRT3mt could localize in the nucleus under stress condition (Fig. 1 e,f). Overexpression of SIRT3mt promoted the deacetylation of histone H3 to the similar levels to that of SIRT3ct (Fig. 5e), indicating that SIRT3mt was also recruited to the nucleus. SIRT3 may be a nucleocytoplasmic shuttling protein similar to SIRT1, which has the nuclear localization signal (NLS) and nuclear export signal (NES) in its amino acid sequence 25 . Proteins with molecules smaller than 25-40 kDa passively diffuse across the nuclear membranes 26 . Because molecular weights of nuclear SIRT3mt and nuclear SIRT3ct were estimated to be 32 kDa and 28 kDa, respectively, SIRT3 may be passively transported into the nucleus. Alternatively, efficient transport and export of a protein into the nucleus need NLS and NES in its amino acid sequences 26 . In the COOH-terminal region of mouse SIRT3, we found a cluster of hydrophobic amino acids, i.e. PRLLINRDLVGP, and a stretch of negatively charged amino acids, i.e. PRRKDVV, which showed some similarity to those of NES and NLS 27 , respectively. Although similar sequences of these putative NES and NLS 23 were found in human SIRT3, further studies are necessary to elucidate the regulation of subcellular localization of SIRT3.
We showed that differentiation of NPCs was severely affected by AA or SIRT3 knockdown (Fig. 3c-e). SIRT3 is necessary for the differentiation of osteoblasts. SIRT3 deficient mice show osteopenia with osteoblast dysfunction, whereas overexpression of SIRT3 or SOD2 improves the differentiation capability of primary osteoblasts derived from SIRT3-deficient mice 28 . Because SOD2 knockdown also suppresses osteoblast differentiation 28 , oxidative stress may be involved in the disturbance of osteoblast differentiation by SIRT3 knockdown. Overexpression of SIRT3 in fact induces SOD2 via FOXO activation 19 and SIRT3 has been shown to deacetylate and activate SOD2 3 .
Weir et al. has reported that overexpression of SIRT3mt in primary cultured hippocampal neurons protects the neurons against ROS induced by AA 13 . We showed that overexpression of SIRT3ct reduced cell death induced by A in PC12 cells (Fig.   6d). Sirt3 mRNA levels of brain increase in model mice and human patients of Alzheimer's disease 13 and ROS levels significantly correlate with cognitive behaviour in model mice with Alzheimer's disease 29,30 . These results suggest that both SIRT3ct and SIRT3mt are involved in neuroprotection against A toxicity in Alzheimer's disease via ROS reduction. Since Sirt3 mRNA and SIRT3 protein were constantly and 24 highly expressed in the brains of various ages (Fig. 2), SIRT3ct as well as SIRT3mt might contribute to maintain brain homeostasis throughout the lifespan of both healthy and diseased brains.

Materials and Methods
Animals. Slc:ddY and C57BL/6J mice were purchased from Nihon SLC (Hamamatsu, Japan) and Sankyo Labo Service (Tokyo, Japan), respectively. containing EGF (20 ng/ml) and bFGF (20 ng/ml) as described previously 15 . For differentiation, neurospheres were plated onto poly-L-ornithine-coated microcoverslips and incubated for 5 days in a growth factor-free MHM medium containing 1% FBS (differentiation medium). In some experiments, AA was added in the differentiation medium and spheres were incubated for 5 days. To estimate differentiation, the numbers of round spheres and flat-shaped differentiated neurospheres with migrating cells were counted in ten visual fields and compared. Mouse cortical neurons were cultured from the embryonic Scl:ddY mouse brain as described by Brewer et al. 34 . Cortical neurons were plated on poly-D-lysine-coated glass discs and cultured with a neurobasal medium containing 2% B-27 Plus Supplement (Thermo Fisher Scientific) for 5 days. For the knockdown of SIRT3 in NPCs, the Sirt3-shRNA lentivirus or control-shRNA lentivirus were used to infect NPCs one day before differentiation 33,35 .

Treatment of cells with staurosporine, H2O2 and Aβ. COS7 cells was transfected
with SIRT3mt-FLAG or SIRT3ct-FLAG cDNA and incubated for 24 h and were then Statistics. Data were expressed as means ± SEMs. Comparisons between multiple groups were made using one-way analysis of variance followed by a post hoc Tukey test. The difference was considered significant if p was <0.05.

Data Availability
The datasets generated and/or analysed during the current study are available from the corresponding authors on reasonable request. 36 The Animal Welfare Guidelines of Sapporo Medical University were followed in all of the studies and experiments using mice. Approval for the experiments was obtained from the Ethical Committee for the Care and Use of Laboratory Animals of Sapporo Medical University (approval numbers: 11-017 and 11-046).
Competing interests. The authors declare no competing interests.