SIRT6 safeguards human mesenchymal stem cells from oxidative stress by coactivating NRF2

SIRT6 belongs to the mammalian homologs of Sir2 histone NAD+-dependent deacylase family. In rodents, SIRT6 deficiency leads to aging-associated degeneration of mesodermal tissues. It remains unknown whether human SIRT6 has a direct role in maintaining the homeostasis of mesodermal tissues. To this end, we generated SIRT6 knockout human mesenchymal stem cells (hMSCs) by targeted gene editing. SIRT6-deficient hMSCs exhibited accelerated functional decay, a feature distinct from typical premature cellular senescence. Rather than compromised chromosomal stability, SIRT6-null hMSCs were predominately characterized by dysregulated redox metabolism and increased sensitivity to the oxidative stress. In addition, we found SIRT6 in a protein complex with both nuclear factor erythroid 2-related factor 2 (NRF2) and RNA polymerase II, which was required for the transactivation of NRF2-regulated antioxidant genes, including heme oxygenase 1 (HO-1). Overexpression of HO-1 in SIRT6-null hMSCs rescued premature cellular attrition. Our study uncovers a novel function of SIRT6 in maintaining hMSC homeostasis by serving as a NRF2 coactivator, which represents a new layer of regulation of oxidative stress-associated stem cell decay.


Introduction
Stem cell exhaustion is considered as one of the hall-marks of aging [1,2]. Mesenchymal stem cells (MSCs) are adult stem cells with the potential of differentiation into mesodermal lineages, such as osteoblasts, chondrocytes, and adipocytes [3], and play an important role in tissue regeneration [4]. Aged human and animals exhibit decreased number or diminished proliferative potential of MSCs [5,6]. Accelerated attrition of MSC pool has been observed in human stem cell and mouse models for human premature aging disorders, including Werner syndrome (WS) and Hutchinson Gilford progeria syndrome [7][8][9][10]. Transplantation of mesoderm-derived stem npg www.cell-research.com | Cell Research cells from young mice increases the lifespan and fitness of progeroid mice [11,12]. While MSC exhaustion is known as a culprit for aging-associated degeneration of mesodermal tissues [1], the underlying genetic pathways employed by human MSCs (hMSCs) to maintain homeostasis have not been fully elucidated.
A number of intracellular and intercellular factors are known to maintain stem cell homeostasis [13,14] and one of such factors is reactive oxygen species (ROS) [5]. ROS is normally maintained at a balanced level by cellular antioxidant systems [15]. Nuclear factor erythroid 2-related factor 2 (NRF2) is a critical redox sensor and is one of the master regulators of antioxidant responses. NRF2 binds to the antioxidant response elements (AREs) and activates the transcription of a number of antioxidant genes such as heme oxygenase 1 (HO-1) that is known for counteracting ROS [16]. The decline in NRF2-ARE activity is observed in aged cells, and may account for the oxidative stress-associated tissue degeneration [17,18]. The key factors regulating NRF2 antioxidant pathway in hMSCs, however, remain to be identified.
SIRT6 is one of the seven mammalian sirtuin homologues of yeast Sir2 longevity protein [19]. Sirt6-deficient mice suffered from multiple acute premature aging syndromes, which ultimately led to premature death within 1 month after birth [20]. SIRT6 contains a domain with NAD + -dependent deacetylase activity [21,22], which accounts for transcriptional repression of target genes such as NF-κB, HIF1α, and c-JUN by deacetylating histone H3 lysine 9 (H3K9) at the gene promoters [23][24][25]. In addition, silencing SIRT6 in fibroblasts leads to the accelerated cellular senescence possibly due to attenuated deacetylation of H3K9 at the telomere region [26]. SIRT6 is also involved in DNA damage repair by deacetylation of H3K56 at the damaged DNA region [27][28][29]. Recently, evidence has emerged that SIRT6 protects cells from oxidative stress-associated DNA damage [21], raising a possibility that SIRT6 could be involved in redox-related cellular homeostasis regulation.
Here we demonstrate that SIRT6 is a positive modulator of NRF2-HO-1 antioxidant pathway in hMSCs, a novel mechanism implicated in safeguarding stem cells from oxidative stress-associated functional decay.

SIRT6-depleted hMSCs exhibit accelerated functional decay
To study the role of SIRT6 in hMSCs, we first generated SIRT6-deficient human embryonic stem cells (hESCs). The exon 1 of SIRT6 gene was removed in hESCs by a transcription activator-like effector nuclease (TALEN)-mediated homologous recombination strategy [30][31][32] (Figure 1A). Biallelic loss of SIRT6 in SIRT6 −/− hESCs was confirmed by genomic PCR and Southern blotting (Supplementary information, Figure S1A and S1B). Loss of mRNA and protein in SIRT6 −/− hESCs was verified by reverse transcription-PCR (RT-PCR) and western blotting, respectively ( Figure 1B). Immunofluorescence staining demonstrated a punctate staining pattern of SIRT6 in the nuclei of wild-type (WT) hESCs, which was absent in SIRT6 −/− hESCs ( Figure 1C). The SIRT6-deficient hESCs were cultured for more than 50 passages without discernible morphological abnormality and still exhibited defining characteristics of pluripotency, including the expression of pluripotency markers and the in vivo differentiation potential towards three germ layer lineages (Supplementary information, Figure S1C and S1D).
Given that tissue stem cell exhaustion has been recently recognized as one of the hallmarks of aging [10], and the Sirt6-deficient mice exhibit features of premature aging in mesodermal tissues [20], we hypothesized that SIRT6 deficiency would result in accelerated attrition of hMSC pool. We thus differentiated the SIRT6 −/− and WT hESCs into hMSCs. SIRT6 −/− and WT hMSCs expressed typical hMSC markers such as CD73, CD90, and CD105 ( Figure 1D), and were found negative for hMSC irrelevant antigens CD45, CD34 and CD43 (Supplementary information, Figure S1E) [10,31]. RT-qPCR and western blotting results confirmed the loss of both SIRT6 mRNA and protein in SIRT6 −/− hMSC ( Figure 1E and Supplementary information, Figure S1F). Immunofluorescence indicated a relatively diffused localization pattern of SIRT6 in the nuclei of WT hMSCs, which was absent in the SIRT6 −/− hMSCs ( Figure 1F).
As the Sirt6-deficient mice exhibited bone abnormalities [20], we wondered whether the differentiation potential was compromised in SIRT6 −/− hMSCs. As expected, compared with WT hMSCs, SIRT6 −/− hMSCs exhibited substantially impaired differentiation towards osteoblasts or chondrocytes (Supplementary information, Figure  S1G). Next, we investigated the link between SIRT6 and hMSC aging. Although serial passaging of SIRT6 −/− hM-SCs led to several observations implicated in premature senescence such as early onset of senescence-associated (SA)-β-Gal activity ( Figure 1G), upregulation of P16 and P21 proteins [33] (Figure 1H), and progressive impairment in cell proliferation ability (Supplementary information, Figure S1H), SIRT6 −/− hMSCs also exhibited features distinct from hMSCs derived from premature aging patients [10]. For instance, different from WS-specific hMSCs [10], decreased transcripts from centromeric repetitive sequences [34], and a mild but Protein extracts from wild-type (WT, SIRT6 +/+ ) and SIRT6-deficient (SIRT6 −/− ) hESCs were analyzed by western blotting using an anti-SIRT6 antibody. β-actin was used as the loading control. Right panel: RT-PCR analysis of SIRT6 mRNA in hESCs. A pair of PCR primers spanning the junction region of SIRT6 mRNA exon 1 and exon 2 was used. 18S rRNA was used as the loading control. (C) Bright-field and SIRT6 immunofluorescence micrographs of WT and SIRT6-deficient hESCs. DNA was stained by Hoechst 33342. Bright-field scale bar, 200 µm; immunofluorescence scale bar, 20 µm; zoom-field immunofluorescence scale bar, 10 µm. (D) Bright-field micrographs and FACS analysis of the surface markers CD105, CD73, and CD90 in WT and SIRT6-deficient hMSCs. Scale bar, 100 µm. (E) Western blotting analysis of SIRT6 protein in hMSCs. Protein extracts from WT and SIRT6-deficient hMSCs were analyzed by western blotting using anti-SIRT6 antibody. β-actin was used as the loading control. (F) Immunofluorescence analysis showing the absence of SIRT6 protein in the nuclei of SIRT6-deficient hMSCs. Scale bar, 10 µm. (G) SA-β-GAL staining from passage 6-8 showing an accelerated senescence in SIRT6-deficient hMSCs. Percentages of SA-β-GAL-positive cells were calculated. Data were presented as mean ± SEM, n = 5, NS, not significant, **P < 0.01. (H) Western blotting analysis of P16 and P21 protein in hMSCs. Protein extracts from WT and SIRT6-deficient hMSCs at late passage (LP, passage 9) were analyzed by western blotting. β-actin was used as the loading control. (I) Analysis of luciferase activity in the TA muscles of immunodeficient mice by in vivo imaging system (IVIS) demonstrating premature attrition of SIRT6-deficient hMSCs after implantation. WT (1 × 10 6 , left) and SIRT6-deficient (1 × 10 6 , right) hMSCs (passage 6) pretransduced with luciferase were implanted into the muscles of mice. Luciferase activities were imaged and quantified 1 week after implantation. Data were presented as mean ± SEM, n = 4, ***P < 0.001.  Figure  S1I and S1J). We ruled out the possibility that the premature aging phenotypes observed in SIRT6 −/− hMSCs resulted from chromosomal instability as reported in mouse study [20], as SIRT6 −/− hMSCs exhibited normal karyotypes and minimal genomic mutational load determined by genome-wide copy number variation (CNV) analysis (Supplementary information, Figure S2A and S2B). Moreover, no significant increase in DNA damage response marker γH2AX was observed in SIRT6 −/− hMSCs (Supplementary information, Figure S2C and S2D). Finally, we investigated whether SIRT6 −/− hMSCs suffered from accelerated functional decay in an in vivo niche by transplanting WT and SIRT6 −/− hMSCs into tibialis anterior (TA) muscle of the immunodeficient mice [10]. Compared with WT hMSCs, an accelerated in vivo decay of SIRT6 −/− hMSCs was observed after engraftment ( Figure 1I, Supplementary information, Figure  S2E and S2F). The attrition of SIRT6 −/− hMSCs could be alleviated by lentiviral vector-mediated reconstitution of exogenous SIRT6 (Supplementary information, Figure  S2G). Altogether, these results support that SIRT6 has a key role in preventing hMSCs from premature attrition.

SIRT6 positively regulates NRF2-mediated HO-1 expression in hMSCs
To uncover the molecular mechanism underlying how SIRT6 regulates redox homeostasis in hMSCs, we performed genome-wide RNA sequencing (RNAseq) and H3K4me3 chromatin immunoprecipitation (ChIP) sequencing (ChIP-seq) analyses. We identified 119 upregulated genes and 246 downregulated genes in SIRT6 −/− hMSCs relative to WT hMSCs ( Figure 3A and Supplementary information, Figure S3A-S3C and Tables S1-S2). We analyzed biological pathways that were potentially altered due to SIRT6 deficiency and found that SIRT6 −/− hMSCs had specific downregulation in genes associated with the GO terms such as "skeletal system development" (Figure 3B and Supplementary information, Figure S3D), which is in agreement with observed phenotypic defects in osteogenesis and chondrogenesis (Supplementary information, Figure S1G). Interestingly, the most significant GO term for downregulated genes was "response to oxygen levels" (P = 4.8E−8), which included a list of NRF2-regulated antioxidant genes [16,40] ( Figure 3C and 3D and Supplementary information, Figure S3D and S3E and Table S3). These genes were markedly downregulated in SIRT6 −/− hMSCs ( Figure 3D), which was associated with decreased levels of active chromatin mark H3K4me3 at their promoters ( Figure 3E and Supplementary information, Figure S3F and S3G). In contrast to previous studies in mouse cells [23][24][25]41], we did not observe global upregulation of HIF1α, c-MYC, c-JUN, and NF-κB target genes in SIRT6 −/− hMSCs (Supplementary information, Figures S4A and  S4B). In addition, SIRT6 deficiency-induced coordinated downregulation of NRF2 target genes appears to be specific to hMSCs, as we did not observe this in Among the NRF2-responsive genes downregulated in SIRT6 −/− hMSCs, we chose to focus on HO-1 (also referred to as HMOX1), which encodes a potent antioxidative and cytoprotective factor [42][43][44]. We found that HO-1 was quickly induced by PX-12 treatment with Apoptotic cell percentage in vehicle-treated WT hM-SCs was normalized to 1. Data were presented as mean ± SEM, n = 3, NS, not significant, *P < 0.05. (C, D) Cellular reactive oxygen species (ROS) and 8-oxodG levels were determined by staining with H2DCFDA probe (C) and an anti-8-oxodG antibody (D), respectively, and measured by FACS. (E) SIRT6-deficient hMSCs were pretreated with H 2 O (control) or 1 mM NAC for 1 week, and then were treated with vehicle (DMSO) or 50 µM PX-12 for 24 h. Cellular apoptosis was measured by Annexin V-PI staining. Data were presented as mean ± SEM, n = 3, **P < 0.01. (F) Overexpression of WT SIRT6 (SIRT6 (WT)), not SIRT6 H133Y mutant (SIRT6 (HY)), in SIRT6-deficient hMSCs partially restored cellular ROS to normal levels. A luciferase (Control)-expressed vector was used as control. (G, H) SIRT6-deficient hMSCs were transduced with SIRT6 (WT), SIRT6 (HY), or Control vector, and then cells were treated with vehicle (DMSO) or 50 µM PX-12 for 24 h. Cell viability (G) and cellular apoptosis (H) were measured by MTS assay and Annexin V-PI staining, respectively. Data in G and H were presented as mean ± SEM, n = 6, *P < 0.05, **P < 0.01, ***P < 0.001.
mRNA level peaking at 4 h and protein level peaking at 8 h in WT hMSCs ( Figure 3F and 3G), indicating that HO-1 is an early response gene induced by oxidative stress in hMSCs. In contrast, HO-1 expression level in SIRT6 −/− hMSCs was insensitive to PX-12 treatment ( Figure 3F and 3G), which was partially restored by overexpression of WT SIRT6, but not the H133Y mutant ( Figure 3H). These data indicate that SIRT6 is required for the induction of HO-1 in response to oxidative stress.
We next determined whether SIRT6 is directly in-npg www.cell-research.com | Cell Research volved in the NRF2-mediated transcription. We transfected SIRT6 −/− and WT hMSCs with an ARE-driven luciferase reporter. Compared with WT, SIRT6 deficiency resulted in a significantly reduced luciferase activity in hMSCs ( Figure 4A). Overexpression of WT SIRT6, but not its H133Y mutant, enhanced NRF2-mediated gene expression in primary hMSCs ( Figure 4B and 4C), suggesting that SIRT6 positively regulated NRF2-mediated transcription in a deacetylase-dependent manner. There was no discernible difference in nuclear NRF2 levels between WT and SIRT6 −/− hMSCs (Supplementary information, Figure S5D). ChIP-qPCR analysis also indicated that SIRT6 deficiency did not affect the amount of NRF2 associated with the classic ARE of endogenous HO-1 enhancer (Supplementary information, Figure S5E). These observations raised an intriguing possibility that SIRT6 directly coactivated NRF2-mediated transcription. To test this, we used the GAL4(DBD)-NRF2/(UAS) 5 -TA-TA-luciferase system. We found that the transactivation ability of NRF2 was positively regulated by SIRT6 in a deacetylase-dependent manner ( Figure 4D and 4E). Using this system, we also identified that NRF2 was able to mediate recruitment of SIRT6 to the promoter DNA, the (UAS) 5 , GAL4-binding site by ChIP-qPCR ( Figure  4F). Co-immunoprecipitation (Co-IP) and GST pulldown analyses showed that SIRT6 associated with NRF2 in a protein complex ( Figure 4G, 4H and Supplementary information, Figure S5F and S5G).We next examined whether SIRT6 serves as a bridge between NRF2 and the basal transcriptional machinery. Co-IP experiment showed that SIRT6 formed a protein complex with RNA polymerase II (RNAP II) and TAF II-p135, two components of RNAP II complex ( Figure 5A). ChIP-qPCR analysis further identified that both SIRT6 and RNAP II were recruited to the HO-1 promoter in WT hMSCs ( Figure 5B and 5C). SIRT6 deficiency led to reduction of HO-1 promoter-bound RNAP II ( Figure 5C). These data suggest that SIRT6 could help assemble NRF2-RNAP II transcription complex at the HO-1 promoter.

Deacetylation of H3K56 by SIRT6 accounts for recruitment of RNAP II to HO-1 promoter
In murine studies, SIRT6 was shown to act as a transcriptional repressor by deacetylating H3K9 at the gene promoters [23][24][25]. As SIRT6 deficiency did not result in upregulation of the H3K9Ac level, it is thus unlikely that H3K9Ac is the major substrate of SIRT6 in hMSCs ( Figure 5D and Supplementary information, Figure S6A). In contrast, the level of H3K56Ac, another substrate of SIRT6 [28,45], was dramatically increased in SIRT6 −/− hMSCs ( Figure 5D, 5E and Supplementary information, Figure S6A). In addition, ChIP-qPCR showed more enrichment of acetylated H3K56 at the promoters of NRF2-responsive genes including HO-1, AKR1C1, and PTGS2 in SIRT6 −/− than in WT hMSCs ( Figure 5F and Supplementary information, Figure S6B). We observed a reduced H3K56 acetylation level at HO-1 promoter when WT SIRT6 but not the H133Y mutant was reintroduced into SIRT6 −/− hMSCs ( Figure 5G and Supplementary information, Figure S6C). In addition, reintroduction of WT SIRT6 instead of the H133Y mutant increased the recruitment of RNAP II to the HO-1 promoter in SIRT6 −/− hMSCs (Supplementary information, Figure S6D), suggesting a deacetylase activity-dependent recruitment of RNAP II by SIRT6. Next, we investigated whether deacetylation of H3K56 by SIRT6 at HO-1 promoter is causally linked to the recruitment of RNAP II. Ectopic expression of H3K56Q, an H3K56 acetylation-mimic mutant [28], in WT hMSCs inhibited the recruitment of RNAP II to the HO-1 promoter (Supplementary information, Figure S6E). On the other hand, ectopic expression of H3K56R, an acetylation-defective mutant [46], in SIRT6 −/− hMSCs promoted occupancy of RNAP II at HO-1 promoter (Supplementary information, Figure S6E). Altogether, these data indicated that SIRT6 is responsible for the deacetylation of H3K56Ac at the HO-1 promoter in hMSCs, which is important for the recruitment of RNAP II transcriptional machinery.

SIRT6-NRF2-HO-1 pathway accounts for anti-oxidative response in hMSCs
As SIRT6 −/− hMSCs exhibited increased basal ROS production and increased sensitivity to oxidative stress, next we tested whether the SIRT6-NRF2-HO-1 axis is linked to hMSCs' anti-oxidative response. We found that overexpression of HO-1 resulted in alleviation of ROS levels in SIRT6 −/− hMSCs ( Figure 6A). In addition, we observed a decrease in the susceptibility of SIRT6 −/− hMSCs to PX-12-induced cytotoxicity upon HO-1 overexpression ( Figure 6B and 6C). Finally, we investigated whether SIRT6-NRF2-HO-1 axis functions in the regulation of hMSC redox homeostasis in an in vivo context. To this end, SIRT6 −/− hMSCs were transduced with a lentiviral HO-1 expression vector and then transplanted into the TA muscles of immunodeficient mice ( Figure 6D and Supplementary information, Figure S6F). As expected, HO-1 overexpression repressed accelerated cell attrition of SIRT6-deficient hMSCs in an in vivo microenvironment ( Figure 6D).
insults; (2) SIRT6-depleted hMSCs show accelerated cellular senescence; (3) SIRT6 deficiency results in impairment of differentiation potential of hMSCs into bone and cartilage. These new findings establish a previously unappreciated connection between SIRT6, oxidative stress, and human stem cell exhaustion, and indicate that SIRT6-NRF2 module may be the key molecular hub to protect the mesodermal tissues from aging-associated degeneration.
Mechanistically, SIRT6 is identified as a positive regu-lator of NRF2-ARE antioxidant pathway in hMSCs (Figure 6E). We propose a model where SIRT6 plays dual roles in NRF2-mediated HO-1 transcription: (1) SIRT6 stabilizes the protein complex comprised of NRF2 and basal transcription machinery, potentially as an adaptor protein; (2) SIRT6 also exerts its histone deacetylase activity specifically at the promoter of HO-1. The two functions of SIRT6 are potentially linked as SIRT6-mediated deacetylation of H3K56 may be a prerequisite for stabilizing the NRF2-SIRT6-RNAP II complex at chro-mosomal levels.
To the best of our knowledge, this study provides the first evidence that SIRT6 can serve as a transcriptional coactivator. Recent murine studies showed that SIRT6 could transrepress the activities of HIF1α, c-JUN, and NF-κB via H3K9 deacetylation [23][24][25]. In hMSCs, however, we found that acetylated H3K9 is not a preferred substrate for SIRT6. Likewise, we did not observe global upregulation of the target genes of HIF1α, c-MYC, c-JUN, and NF-κB [23][24][25]41] in SIRT6-depleted hMSCs. Instead, deficiency of SIRT6 in hMSCs results in an increase in acetylated H3K56 levels, and SIRT6-mediated deacetylation of H3K56 is a key event facilitating NRF2-dependent gene expression. In addition, in line with a role of SIRT6 as a NRF2 coactivator, SIRT6 interacts with RNAP II complex and the depletion of SIRT6 in hMSCs diminishes the enrichment of RNAP II complex at the HO-1 promoter. Of note, a recent genome-wide study in hESCs revealed that SIRT6 co-localizes with RNAP II at gene promoter regions [47], which together with our finding raises an interesting possibility that other SIRT6-mediated transcriptional activation events may also exist. In addition, it should be noted that SIRT6-mediated positive regulation of NRF2 activity appears to be specific to hMSCs. Indeed, we did not observe global downregulation of NRF2 target genes in SIRT6-deficient hVECs. In these contexts, more complex regulations such as cell-type-specific protein-protein interactions, deacetylation-independent activities of SIRT6, as well as potential compensations from other histone deacetylases may be involved.
The implication of SIRT6-regulated H3K56 acetylation in various cellular events has not been fully investigated. A very recent study indicated that human SIRT6 acts as a scaffold for recruiting SNF2H, a chromatin remodeler, and other DNA repair factors to DNA damage foci, an event dependent on SIRT6's deacetylase activity at histone H3K56 [28]. It is likely that SIRT6-mediated low acetylation level of H3K56 at DNA damage site is important for an open local chromatin configuration necessary for DNA repair. Along this line of thinking, one could reason that transcriptional events are also aided by a permissible chromatin configuration, i.e., chromatin remodeling-dependent factor recruitment. Therefore, it is reasonable that SIRT6/NRF2-dependent HO-1 transcription may employ a similar H3K56 deacetylation-based mechanism to recruit RNAP II complex. How the increase in local H3K56Ac levels potentially regulates chromosomal configuration and recruitment of transcriptional machineries warrants further investigation.
The findings that SIRT6-NRF2-HO-1 axis acts as a regulator for stem cell redox homeostasis and SIRT6 functions as a transcriptional coactivator undoubtedly add a new layer of knowledge to SIRT6 biology. Safeguarding redox homeostasis by SIRT6 in hMSCs principally could contribute to mechanistic explanations for various SIRT6-related biological processes, including genomic instability, cellular senescence, cellular transformation, and metabolic dysregulation. Our study provides first evidence that SIRT6 is a key gatekeeper for human adult stem cell homeostasis, highlighting that SIRT6-NRF2 pathway may be a novel target for preventing aging-associated stem cell attrition, and hopefully, for treating aging-associated disorders in the future.
Generation of SIRT6 −/− hESCs TALEN-mediated gene targeting was performed as previously described [31]. In brief, a donor plasmid was constructed by the combination of 1.0-2.0-kb homology arms and drug resistance cassettes (neo). H9 hESCs (5.0 × 10 6 ) were electroporated with a pair of TALEN vectors (Addgene, TAL2454 (Plasmid #36843) and TAL2455 (Plasmid #36844)) and the donor plasmid, and subsequently cultured on MEF feeder. G418 (100 µg/ml, Invitrogen) was then added to the medium to initiate positive selection 2-4 days after electroporation. After about 2 weeks' selection, G418-resistant clones were picked and transferred to a 96-well plate for further characterization and expansion. Gene-targeted clones were determined by genomic PCR with the primers listed in Supplementary information, Table S4 using PrimeSTAR DNA Polymerase (TAKARA). Long PCR cycling included a 1-min initial denaturation at 98 °C, 35 cycles of 10-s denaturation at 98 °C and a 0.5-3-min annealing and extension at 68 °C plus a final extension at 72 °C for 7 min.

Southern blotting
Genomic DNA was extracted following a described strategy [48]. 20 µg genomic DNA of each samples was digested with Hind III and Xho I (NEB) overnight and subjected to electrophoresis at 50 V on a 0.8% agarose gel (SeaKem Gold agarose, Lonza) for 3-4 h. The gel was subsequently incubated in 0.25 M HCl for 3-5 min followed by 2× 15-min incubation in denaturation buffer (0.5 M NaOH, 1.5 M NaCl) and 2× 15-min incubation in neutralization buffer (0.5 M Tris-HCl, pH 7.5; 1.5 M NaCl). The DNA was then blotted overnight onto a nylon membrane (GE Healthcare Life Sciences) by capillary transfer in 20× SSC buffer. The membrane was npg www.cell-research.com | Cell Research then ultraviolet crosslinked. The 5′ and 3′ probes were amplified from genomic DNA using a DIG-label kit (Roche) with the primers in Supplementary information, Table S4, following the manufacturer's protocol. The probes were labeled with DIG and Southern hybridization was performed following the standard protocol.

Excision of the neomycin-resistance cassette
To remove the neomycin-resistance cassette, SIRT6 gene-targeted hESCs were electroporated with pCAG-FLpo-2A-puro vector and then cultured on MEF feeder. Three days after transfection, puromycin (1 µg/ml; Invitrogen) was used to enrich puro-resistant cells. Puromycin was withdrawn after 48 h. Ten days later, the emerging colonies were picked and expanded in 96-well plates. Removal of the neomycin-resistance cassette was verified by PCR using PrimeSTAR DNA Polymerase (TAKARA) with the related primers (Supplementary information, Table S4).

Directed differentiation of hESCs into hVECs
Differentiation was performed as previously described [49], and the generated hVECs were characterized by the expression of hVEC-specific markers VE-cadherin, CD31, and VWF, as well as the activity of AC-LDL uptake.
Clonal expansion assay 2 000 cells were seeded in each well of 12-well plates, cultured for 2 weeks, and stained with 0.2% crystal violet. Cell numbers were counted using light microscope in randomly selected fields. Each experiment was performed in triplicate.

Cell viability analysis
hMSCs (2 × 10 4 ) were seeded onto 0.1% gelatin-coated 96-well plates (Corning, Costar), and upon reaching 90% confluence cells were treated with various stressors for 24 h. Cell viability was measured using MTS approach according to the recommended protocol for CellTiter 96 AQueous One Solution Cell Proliferation Assay (Promega). Stressors used in this assay include DMSO

Cell death analysis
Cell death was determined using Cytotox 96 non-radioactive cytotoxicity assay (Promega). Cells were treated with DMSO or 50 µM PX-12 for 24 h, and then the LDH level in culture medium was assayed. Each experiment was performed in biological triplicate.

Western blotting
Protein quantification was performed using a BCA Kit. Protein lysate was subjected to SDS-PAGE and subsequently electrotransferred to a polyvinylidene fluoride membrane (Millipore). Western blotting was performed as previously described [8,48,51].

Flow cytometry analysis
For cell cycle analysis, cells were pulsed with 10 µM EdU for 2 h, and collected using TrypLE Express (Invitrogen). Collected cells were stained with Alexa Fluor 647 dye azide and propidium iodide according to the manufacturer's instruction for Click-iT EdU Alexa Fluor 647 Flow Cytometry Assay Kit (Molecular Probes, c10419). Cell cycle profile was quantified by a FACS machine (BD LSRFortesa). For cell apoptosis analysis, cells were collected freshly and stained with Annexin V-EGFP and PI, and then apoptotic cells were quantified by FACS. For ROS measurement, cells were collected, loaded with 1 µM H2DCFDA for 30 min using ROS Detection Reagents (Molecular Probes, C6827), and quantified by FACS.

Co-IP assay
For exogenous Co-IP, cells were lysed in NP-40 lysis buffer (1% NP-40, 150 mM NaCl, 10% glycerol, 1 mM EGTA, 1 mM EDTA, 20 mM Tris, pH 7.5, and complete protease inhibitor cocktail (Roche)), and for endogenous Co-IP, cells were lysed in CHAPS buffer (120 mM NaCl, 0.3% CHAPS, 1 mM EDTA, 40 mM HEPES, pH 7.5, and complete protease inhibitor cocktail (Roche)). Cells were lysed on ice for 30 min, and then centrifuged at 12 000 rpm. at 4 °C for 20 min. For endogenous Co-IP, lysates (1 mg protein) were pre-cleared with 20 µl of Protein A/G-PLUS Agarose beads (Santa Cruz) for 2-4 h, and then the supernatants were collected by centrifugation at 3 000 rpm at 4 °C for 3 min. The supernatants mixed with the indicated antibodies and beads were rotated overnight at 4 °C. After being washed with NP-40 buffer or CHAPS buffer for three times, the immunocomplexes were eluted by boiling in 1× SDS-loading buffer for 10 min.

RNA-seq library construction
One million cells were applied to extract total RNA using the RNeasy Mini Kit (Qiagen) according to the manuals. After quantification of RNA by Fragment Analyzer (Advanced Analytical), 1.5 µg of total RNA was used to construct sequencing libraries by TruSeq RNA Sample Preparation Kit (Illumina) following the manufacturer's standard protocol.

cells via DNeasy
Blood & Tissue Kit (Qiagen). After being sheared into fragments ranging from 150 to 200 bp, the DNA was constructed into sequencing libraries using NEBNext DNA Library Prep Reagent Set for Illumina (NEB).

ChIP-seq
ChIP-seq was performed according to a previous protocol with some modifications [54]. Briefly, cells were crosslinked and lysed, and then the released chromatin was sheared into fragments. To obtain fragments of interest, samples were incubated with anti-H3K4me3 antibody (Abcam, ab8580) overnight. Inputs served as negative controls. After the DNA was de-crosslinked and extracted, the ChIP-seq libraries were generated using NEBNext DNA Library Prep Reagent Set for Illumina (NEB) according to manuals.

ChIP-seq data analysis
Clean reads were mapped to hg19 genome by BWA. Then, peaks were called by MACS2 [55] by default parameters using input as control. Density of the promoter region was defined as the non-duplicate reads located in the regions between 2-kb upstream of the TSS (transcription start site) and 2-kb downstream of the TSS and then were normalized by the total non-duplicate reads that were mapped to the hg19 genome.

RNA-seq data analysis
Low-quality reads (N bases >10% of a read, >50% of the bases in a read with phred 33 < 5, or phred score of the first 5 bases < 20) were discarded. Then reads were mapped to the human reference genome hg19 (from UCSC) by Tophat [56] and Cufflinks [57]. Transcript expression and differentially expressed genes were analyzed as previously reported [58].

CNV analysis
Clean reads of genomic libraries were first mapped to hg19 genome by BWA [59]. Then, the reads in 0.5-M window along the genome were normalized by the total bases mapped to the reference and the number of non-N bases in 0.5-M genome window size.

RT-qPCR
Total RNA was extracted from cells according to the instruction of TRIzol Reagent (Invitrogen). 1-2 µg total RNA was used for cDNA synthesis with reverse transcription Master Mix (Promega). Quantitative real-time PCR was carried out using iTaq Universal SYBR Green Supermix (Bio-Rad) on a CFX384 Real-Time PCR system (Bio-Rad). All data were normalized by 18S rRNA transcript and calculated using the ∆∆Cq method. All RT-qPCR primer pairs are listed in Supplementary information, Table S4.

Animal experiments
For mouse experiments, 6-8-week-old male NOD-SCID mice and immune-deficient CD-1 nude mice were used. Animal experiments performed in this study were approved by the Chinese Academy of Science Institutional Animal Care and Use Committee. For teratoma formation assay, NOD-SCID mice were injected s.c. with 3 × 10 6 WT and SIRT6-deficient hESCs in a Matrigel/mTeSR solution, respectively. Teratomas were collected when reaching a size of around 10 mm in diameter, and subjected to immunostaining. For hMSC transplantation assay, 10 6 WT and SIRT6-deficient hMSCs pretransduced with a lentiviral vector encoding luciferase were injected to TA muscle of CD-1 nude mice. Mice were imaged in vivo with IVIS spectrum imaging system (XENOGEN, Caliper) for luciferase activity. After imaging, TA muscle of mice was subjected to immunofluorescence analysis.

Statistical analysis
Results were presented as mean ± SEM. Two tailed Student's t-test was conducted using Graph-Pad Prism Software. P values < 0.05 were considered statistically significant (*).

Accession numbers
All of the RNA-seq and ChIP-seq data have been deposited in GEO under the accession number GSE64642. Whole-genome sequencing data for CNV analysis were deposited in SRA under the accession number SRP059859.