Inactivation of Lsd1 triggers senescence in trophoblast stem cells by induction of Sirt4

Coordination of energy metabolism is essential for homeostasis of stem cells, whereas an imbalance in energy homeostasis causes disease and accelerated aging. Here we show that deletion or enzymatic inactivation of lysine-specific demethylase 1 (Lsd1) triggers senescence in trophoblast stem cells (TSCs). Genome-wide transcriptional profiling of TSCs following Lsd1 inhibition shows gene set enrichment of aging and metabolic pathways. Consistently, global metabolomic and phenotypic analyses disclose an unbalanced redox status, decreased glutamine anaplerosis and mitochondrial function. Loss of homeostasis is caused by increased expression of sirtuin 4 (Sirt4), a Lsd1-repressed direct target gene. Accordingly, Sirt4 overexpression in wild-type TSCs recapitulates the senescence phenotype initiated by Lsd1 deletion or inhibition. Inversely, absence of Lsd1 enzymatic activity concomitant with knockdown of Sirt4 reestablishes normal glutamine anaplerosis, redox balance and mitochondrial function. In conclusion, by repression of Sirt4, Lsd1 directs the epigenetic control of TSC immortality via maintenance of metabolic flexibility.

Age-associated pathophysiology is caused by a progressive decline in regenerative capacities, which is unequivocally linked to a decreased potential of stem cells. The capability of stem cells to differentiate into specialized cell types allows for the replacement of damaged or deteriorated tissues. In addition, their capacity to proliferate is essential for selfpreservation. Thus, understanding the molecular mechanisms for self-maintenance of stem cells is an important issue in the prevention of age-related diseases. 1 Cellular senescence is a stress response that accompanies stable exit from the cell cycle without affecting cellular viability. The stress can be diverse such as telomere shortening, oncogene activation, reactive oxygen species (ROS), mitochondrial dysfunction or aberrant epigenetic modification. [2][3][4][5][6][7] The irreversible cease of proliferation in senescent cells is caused by increased expression of cell cycle inhibitors such as cyclin-dependent kinase inhibitor 2a (p16) and 2b (p15) and hypophosphorylation of retinoblastoma 1 (Rb) for the implementation of senescence. 6 Finally, senescent cells are hallmarked by aberrant lysosomal β-galactosidase activity 8 and can be identified by Rb-mediated chromatin-dense foci. 9 Oxidative stress is mostly generated in mitochondria. 10 Stem cells are particular sensitive to increased oxidants 11 suggesting a link between high ROS levels and a decline in stem cell number during aging. 12,13 The ability of cells to antagonize oxidative damage is determined by a controlled balance between the emergence of ROS and the capacity to generate antioxidants. The tripeptide glutathione (GSH) is the most abundant reducing agent within a cell, which can be oxidized to glutathione disulphide (GSSG) and regenerated to GSH at the expense of NADPH. 10 The level of GSH declines during aging and disease resulting in elevated oxidative stress, but the underlying molecular mechanism remains elusive. 14-16 GSH is synthesized from glutamate, cysteine and glycine, in which the conjunction of glutamate with cysteine delineates the rate-limiting step. 10,17 Glutamate is derived from glutamine, the most abundant free amino acid and the main nitrogen donor for synthesis of nucleotides and amino acids. Many cell types including cancer cells rely on glutaminolysis for maintenance of redox balance by generation of GSH and NADH and by replenishing tricarboxylic acid (TCA) cycle (glutamine anaplerosis) for biosynthesis and as energy source. 10 A prominent antagonist of glutamine anaplerosis is sirtuin 4 (Sirt4), a member of the conserved NAD + -dependent protein deacetylase/ADP-ribosyltransferase family implicated in diverse biological processes. 18,19 Sirt4 is localized in the mitochondrial matrix 20 and has been shown to restrain proton leakage, consequently increasing the membrane potential and reducing mitochondrial oxygen consumption. 21 Independently, Sirt4 also controls glutamine anaplerosis by repression of glutamate dehydrogenase and glutamine uptake. [22][23][24] Disclosure of the crosstalk between metabolic signals and a complex network of genetic and epigenetic pathways including histone methylation that direct self-renewal, potency, and differentiation of stem cells has just begun. 25,26 The first identified histone demethylase lysine-specific demethylase 1 (Lsd1; also called Kdm1a) can remove di-and monomethyl groups from histone H3 at lysine 4 (H3K4) 27,28 and lysine 9 (H3K9). 29,30 Ubiquitous deletion, [31][32][33][34][35][36] as well as deletion of Lsd1 in trophectoderm or epiblast leads to early embryonic lethality. 36 Lsd1 has been associated with repression of stem cell genes at the onset of mouse embryonic stem cell (ESC) differentiation, whereas it is dispensable for ESC maintenance. 37 In contrast, in human ESCs LSD1 balances self-renewal and differentiation. 31 In trophoblast stem cells (TSCs), giving rise to the embryonic part of the placenta, overexpressed Lsd1 has been reported to interact with oestrogen-related receptor β, 38 a nuclear hormone receptor essential for placental formation. 39 Deletion of Lsd1 in TSCs skews cell fate and, despite the preservation of stemness marker gene expression in undifferentiated TSCs, leads to increased cell size 36 indicating the requirement for Lsd1 to sustain proper stem cell characteristics. Furthermore, analyses of adipocytes implicate Lsd1 in the regulation of mitochondrial homeostasis. [40][41][42] To date, little is known about the molecular mechanisms linking chromatin modification, metabolism and stem cell maintenance.
Here we show that loss, genetic or chemical inhibition of the histone demethylase Lsd1 triggers senescence in TSCs. Decrease of Lsd1 activity induces deregulated anaplerotic metabolism and loss of redox balance by elevated expression of the direct target gene Sirt4. Consequently, increased Sirt4 levels are sufficient to repress glutamine anaplerosis, disrupt redox homeostasis and induce senescence. Vice versa, reduction of Sirt4 expression in the absence of Lsd1 activity prevents premature senescence in TSCs. In summary, our data establish Lsd1-directed metabolism as a vital hub for the control of senescence in TSCs.

Results
Lsd1 deficiency boosts senescence in TSCs. We previously showed that deletion of Lsd1 (Lsd1 − / − ) in TSCs 36 leads to altered morphology reminiscent of senescent cells, 6 which prompted us to inspect senescence-associated lysosomal hydrolase β-galactosidase activity. We observed a marked increase of β-galactosidase-positive stained cells within 40 days in Lsd1-deleted compared with wild-type (Lsd1 +/+ ) cultures (Figures 1a and b). To examine any requirement for the enzymatic activity of Lsd1, we analyzed the impact of two different Lsd1-specific, nanomolar affinity inhibitors (Lsd1 i-1 ; ORY-1001 (ref. 43); Lsd1 i-2 ; QC6688) on senescence in wild-type TSCs. We observed a marked raise in the number of senescence-associated β-galactosidasepositive cells after 8 days of Lsd1 inhibition (Supplementary Figure S1a). In agreement, proliferation of TSCs incubated with Lsd1 inhibitors decreased in comparison with the control (Supplementary Figure S1b). To verify the impact of the demethylase activity of Lsd1 in controlling senescence, we generated TSCs homozygous for a conditional, enzymatically inactive mutant knockin allele. The endogenous Lsd1 gene is replaced with the mutant (Lsd1 ki/ki ) after induction of CreERT2-mediated recombination. Similar to Lsd1-deficient stem cells, the percentage of cells stained for senescence-associated β-galactosidase activity was robustly increased compared with wild-type TSCs within 40 days (Figures 1a  and b). To corroborate premature senescence in TSCs devoid of Lsd1 activity, we stained cells with DAPI. In contrast to wild-type cells, senescence-associated nuclear foci were clearly visible in Lsd1 − / − and Lsd1 ki/ki TSCs (Figure 1c). Consistently, proliferation was severely impaired in Lsd1deficient and Lsd1 enzymatic inactive mutant TSCs in comparison with wild-type cells after 40 days (Figure 1d).
At the molecular level, comparison of control to Lsd1 − / − and Lsd1 ki/ki cells disclosed increased protein levels of Rb, p15 and p16, while phosphorylated Rb was reduced 8 days after addition of tamoxifen as indicated in Figure 1e. Quantification of p16 and phosphorylated Rb for individual cells by FACS in wild-type and Lsd1-deficient cells verified the altered protein levels as indicated by Supplementary Figure S1c. TSCs cultured in the presence of Lsd1 inhibitors for 24 h already displayed upregulation of cell cycle inhibitors (Supplementary Figure S1d), which was corroborated by quantitative reverse transcription PCR (qRT-PCR; Figure 1f and Supplementary Figure S1e). Furthermore, we depicted the dynamic of the phenotype by quantification of p16 and phosphorylated Rb at the single-cell level. TSCs cultured in the presence of Lsd1 inhibitor for 24 h, 3 and 8 days showed a progressive increase of p16 and a decrease of phosphorylated Rb immunoreactivity in comparison with control cells (Supplementary Figure S1f). Moreover, we also analyzed DNA damage by quantification of phospho γH2ax. The level was increased 8 days after Lsd1 deletion and in TSCs inhibited for 24 h (Supplementary Figures S1g and h). As Lsd1 has been associated with DNA damage response, [44][45][46] we also quantified p53, phosphorylated p53 and p21. Although we did not detect a direct impact of Lsd1 on transcription, we observed increased protein levels for p53, phospho 53 and p21 in accordance with elevated phospho γH2ax levels as indicated in Supplementary Figures S1g and h.
Taken together, our data demonstrate that the loss of Lsd1 demethylase activity, either by genetic deletion, chemical inhibition or replacement of enzymatic-active by enzymatic inactive Lsd1 inevitably provokes senescence in TSCs.
Lsd1-controlled transcription regulates metabolic pathways and aging. To gain first insight into the molecular underpinnings of Lsd1-controlled senescence, we performed transcriptome analysis by RNA-sequencing (RNA-seq). We cultured wild-type TSCs in the presence or absence of a Lsd1-specific inhibitor for 24 h to unravel immediate early differentially expressed genes before senescence-associated β-galactosidase activity is increased (Supplementary Figure S1a). We obtained 1115 differentially expressed genes (≥1.4-fold expression change, ≥ 50 reads and P ≤ 10 − 5 ) with on average 1.9-fold change in transcription ( Figure 2a). To identify direct target genes of Lsd1, we also determined genome-wide Lsd1 chromatin occupancy in wildtype TSCs by chromatin immunoprecipitation using an Lsd1 antibody followed by massive parallel sequencing (ChIPseq). The antibody has previously been validated and specifically recognizes Lsd1. 40 In total, we identified 13 675 high confidence Lsd1 peaks with a false discovery rate (FDR) o0.1%. (Figure 2b). Lsd1-binding sites within the promoter were centred around the transcription start site (Figure 2c).
Intersection of the Lsd1 promoter occupancy and transcriptome data sets revealed 621 differentially expressed direct target genes of Lsd1 in TSCs of which 59% were upregulated (Figure 2d; Supplementary Table S1). Enrichment analyses depicted metabolic processes and mortality/aging as the topranking categories for differentially expressed direct Lsd1 target genes (Figure 2e).

Lsd1 regulates anaplerosis and maintains redox balance.
To gain further insight into the dynamics of Lsd1-regulated metabolism, we quantified annotated metabolites of wild-type TSCs at two different time points by mass spectrometry (LC-MS/MS). We cultured TSCs in the presence or absence of Lsd1 inhibitor for 24 h and with or without tamoxifen for 8 days (Figure 3a and Supplementary Figure S2a). For As ammonia is mostly derived from glutamine anaplerosis, 47 we also measured NH 4 + concentration. The level of ammonia in TSCs was clearly decreased upon inhibition of Lsd1 (Figure 3e), which may explain why the levels of many amino acids were reduced ( Figure 3a).
As the flux into the TCA cycle is directly linked to oxidative phosphorylation, we determined mitochondrial function in the presence or absence of Lsd1 activity after 24 h by quantification of the normalized oxygen consumption rate (OCR). We quantified the basal respiration by subtraction of the nonmitochondrial OCR after disabling the electron transport system by addition of rotenone (R) and antimycin A (A). The basal respiration was decreased in TSCs in the presence of Lsd1 inhibitor (Figure 3f). The reserve capacity depicts the capability of cells to respond to increased energy demand or stress. It is determined by subtracting the basal OCR from the maximal respiration after uncoupling by FCCP (F). The reserve capacity was not affected by Lsd1 inhibition (Figure 3f). Finally, we determined the mitochondrial membrane potential by JC-1 assay. Inhibition of Lsd1 for 24 h induced an increase in the membrane potential in comparison with the control (Figure 3g). Our data show that oxygen consumption was still coupled to emergence of ATP after Lsd1 inhibition. Direct measurement corroborated that the ATP concentration was not altered (Figure 3h). Hence, absence of Lsd1 activity for 24 h induces mitochondrial stress indicated by decreased basal oxygen consumption, increased mitochondrial membrane potential and glycolysis. In accordance with our metabolomics data, this could be caused by a reduced substrate supply from glutamine anaplerosis.
Increased membrane potential with curtailed oxygen consumption suggests increased ROS levels. Accordingly, we quantified H 2 O 2 in the cytoplasm and in mitochondria using ratiometric roGFP2-based redox probes. 48 Inhibition of Lsd1 for 24 h induced a raise in H 2 O 2 in both cellular compartments ( Figure 3i). Our data were confirmed by measurements of overall ROS formation. Using DCF as a dye that becomes fluorescent upon oxidation, we observed an increase in relative signal intensity upon Lsd1 inhibition (Figure 3j). We also evaluated the level of glutathione, the major redox reagent in the cytoplasm and in mitochondria, using ratiometric roGFP2-based redox reporters. 48 Inhibition of Lsd1 for 24 h raised the relative amount of oxidized glutathione (GSSG) in both compartments (Figure 3k). Consistently, reducing power was diminished as indicated by the increased ratio of oxidized to reduced levels of NADP, NAD and decreased GSH concentration (Figures 3l and m). In brief, short-term inhibition of Lsd1 raises ROS-mediated stress, which is exacerbated by diminished reducing capacity. The latter may be based on the decreased glutamine anaplerosis leading to a decline in GSH synthesis.
We next investigated the impact of Lsd1 on mitochondrial function at a later time point 8 days after the induction of Lsd1 deletion. Similar to short-term Lsd1 inhibition, the basal OCR was reduced in Lsd1 − / − TSCs. In contrast, reserve capacity, mitochondrial membrane potential, and ATP levels were reduced displaying that the initial mitochondrial stress caused by the absence of Lsd1 leads to mitochondrial damage over time ( Supplementary Figures S2c-e). H 2 O 2 and ROS were maintained at an elevated level and the amount of reducing metabolites remained low in Lsd1-deficient TSCs (Supplementary Figures S2f-j). In summary, loss of Lsd1 diminishes glutamine anaplerosis essential to maintain redox balance and functional mitochondria.
Sirt4 triggers senescence in TSCs. To uncover the immediate early molecular mechanism triggering senescence in TSCs, we screened differentially expressed direct targets of Lsd1 involved in metabolic pathways, regulation of mitochondrial activity and glutamine anaplerosis. We considered Sirt4 as the most promising candidate. Sirt4, a direct target of Lsd1 and upregulated by inhibition of Lsd1 (Figure 4a), is known to restrain proton leakage 21 and glutaminolysis by repression of glutamine uptake and Glud1 activity. 22,23,49 We validated the increase in Sirt4 expression after inhibition of Lsd1 by qRT-PCR (Figure 4b) and by western blot (Figure 4c). Consequently, we were keen to link the impact of enhanced Sirt4 expression in wild-type TSCs to senescence, glutamine anaplerosis, redox balance and mitochondrial function. Transient transfection of a SIRT4  Figure S3a). Accordingly, the number of senescence-associated β-galactosidase-positive cells was increased ( Figure 4e) and proliferation was curtailed (Figure 4f) similar to the phenotype observed in TSCs cultured in the presence of Lsd1 inhibitor for 8 days (Supplementary Figures S1a and b). In comparison with mock-transfected TSCs, glutamine uptake (Supplementary Figure S3b) and Glud1 activity (Supplementary Figure S3c)  To analyze the impact of decreased glutamine anaplerosis on senescence in TSCs, we inhibited glutaminase (Gls), which converts glutamine to glutamate, by BPTES 50 or compound 968. 51 Oxidation of glutamate to α-ketoglutarate by Glud1 was inhibited by EGCG. 52 Either inhibition enhanced the expression of cell cycle inhibitors p16 and p15 and Rb while the transcription of Lsd1 was not affected (Supplementary Figure S3i). Furthermore, senescence-associated β-galactosidase activity was increased (Supplementary Figure S3j). Thus, direct inhibition of glutamine anaplerosis triggers senescence in TSCs similar to SIRT4 overexpression or Lsd1 inhibition.
Knockdown of Sirt4 prevents redox imbalance in the presence of Lsd1 inhibitor. To prevent the imbalance of the redox potential triggering senescence in TSCs devoid of Lsd1 demethylase activity, we reduced Sirt4 by RNAi. Reduction of endogenous Sirt4 expression by two specific siRNAs in the presence of Lsd1 inhibitor was verified by western blot (Figure 5a). We first analyzed mitochondrial function. Inhibition of Lsd1 in TSCs transfected with an unrelated control siRNA decreased the basal OCR. In contrast, knockdown of Sirt4 by two independent siRNAs returned the basal OCR to the level of control cells (Figure 5b). Similarly, reserve capacity and membrane potential were restored to control levels (Figures 5b and c). Similarly, ROS caused by the inhibition of Lsd1 was curtailed by simultaneous transfection of siRNAs directed against Sirt4 (Figure 5d). Accordingly, the relative levels of NADH and GSH were elevated upon Sirt4 reduction despite the presence of Lsd1 inhibitor (Figures 5e  and f). Similarly, glutamine uptake was augmented by knockdown of Sirt4 in the presence of Lsd1 inhibitor (Figure 5g). Taken together, the imbalance of the redox potential caused by inhibition of Lsd1 can be averted by concomitant reduction of Sirt4 expression.
Lsd1 controls senescence by repression of Sirt4. To finally prove that Lsd1 prevents senescence by repression of Sirt4 in TSCs, we reduced Sirt4 expression in Lsd1 − / − TSCs. Specific knockdown of Sirt4 was verified by western blot (Supplementary Figure S4a). Expression of the cell cycle inhibitors p15 and p16 and Rb were decreased and phosphorylation of Rb increased (Figure 6a and Supplementary Figure S4a). Accordingly, proliferation was enhanced in Lsd1 − / − TSCs upon Sirt4 knockdown (Figure 6b). Importantly, knockdown of Sirt4 in Lsd1-deficient TSCs markedly decreased the percentage of senescence-associated β-galactosidase-positive cells in comparison with the control (Figure 6c and Supplementary Figure S4b).
In summary, we demonstrate that Lsd1 deletion triggers senescence in TSCs, by unravelling a key role of Lsd1 to sustain metabolic flexibility. We propose the following model ( Figure 6d). Lsd1 binds to the promoter of Sirt4 and impedes transcription. Consequently, glutamine anaplerosis and GSH synthesis are not restrained by Sirt4 expression. In parallel, enhanced mitochondrial membrane potential exacerbated by diminished oxygen consumption favouring ROS is averted. Thus, the redox level is kept in balance preventing stressinduced senescence.

Discussion
Our data show that the demethylase activity of Lsd1 governs redox homeostasis and metabolic flexibility to control senescence in TSCs. The increase in Sirt4 transcription after Lsd1 inhibition in conjunction with binding of Lsd1 to the Sirt4 gene locus indicates that the expression of Sirt4 is directly repressed by Lsd1. Overexpression of Sirt4 in TSCs is sufficient to increase the mitochondrial membrane potential accompanied by simultaneous reduction of oxygen consumption, a combination, which favours the emergence of ROS. We suggest that loss of redox and metabolic balance triggers senescence in TSCs devoid of functional Lsd1 (Figure 6d).
The raise in Sirt4 expression in TSCs devoid of Lsd1 activity interferes with glutamine anaplerosis. Surprisingly though, we also do observe increased levels of fructose 6-phosphate, 3-phosphoglyceric acid, pyruvate and citrate after inhibition of Lsd1 proposing an enhanced glycolytic catabolism into the TCA cycle. This observation is counterintuitive, because Sirt4 has been associated with the inhibition of the pyruvate dehydrogenase complex. 53 The effect of Lsd1 inhibition on glucose metabolism is further complicated by decreased levels of serine, glycine and cysteine, which all can be derived from 3-phosphoglycerate. The low levels of these amino acids may be due to the importance of glutamine anaplerosis as the main nitrogen donor for the synthesis of amino acids 54 or to an increased glycolytic flux. 55 Furthermore, we noticed decreased cysteine levels suggesting that the one-carbon cycle is diminished. This is corroborated by a 70% reduction of the methyl donor S-adenosyl methionine. In the future, it will be interesting to further analyze the impact of the histone demethylase Lsd1 on the substrate level for histone and DNA methyltransferases. Lsd1 may direct a potential feedback mechanism, linking metabolism with chromatin methylation.
In the context of senescence triggered by Lsd1 inhibition, it is compelling to note that LSD1 and SIRT4 control proliferation in cancer cells. Enhanced LSD1 expression has been linked to tumour progression [56][57][58][59] and can even prime cells for transformation. 60 Consequently, clinical trials are currently ongoing to implement LSD1 inhibitors for the treatment of refractory small cell lung cancer (NCT02034123) and acute myeloid leukaemia (NCT02177812, NCT02273102 and NCT02261779). Inversely, SIRT4 can function as a tumour suppressor and reduced expression is associated with several cancers. 22,61 Furthermore, Sirt4-deficient mice spontaneously develop lung tumours. 24 Many cancer cells heavily depend on glutaminolysis to couple bioenergetics to redox balance, supporting cell survival. 54,62 The LSD1-directed control of glutamine anaplerosis offers an explanation for the abatement on tumour progression by LSD1 inhibitors in cancers. Excitingly, our data suggest that Lsd1 inhibition also causes increased glycolytic flux at least in TSCs. In the future it will be interesting to analyze the impact of LSD1 on glucose anaplerosis in other stem and cancer (stem) cells. Treatment with LSD1 inhibitors and simultaneous repression of glycolysis may enhance the efficacy of treatment by killing senescent cancer cells but also could increase the range of tumours susceptible to growth inhibition by LSD1 inhibitors. Of note, inhibition of Lsd1 has never been connected with senescence, indicating that the cellular context is important for the control of proliferation by Lsd1. Unravelling the corresponding molecular mechanism and identification of responsive cell types may pave the way for an extension of cancers susceptible to co-treatment with Lsd1 inhibitors and compounds affecting central metabolic pathways. Stem cell maintenance during aging and obliteration of cancer stem cells during tumour progression are connected by chromatin modifications and the control of metabolism. Thus, the histone demethylase Lsd1 is a very attractive target integrating the control of chromatin and metabolism.  Table S2) 48 h before start of treatment with Lsd1 inhibitor or 4OH-tamoxifen and maintained for the duration of the experiment. TSCs transfected with pEF-DEST52-Cherry-SIRT4 or empty vector were selected by FACS. During selection and duration of the experiment, the cells were incubated in stemness conditions (cell culture medium containing 25 ng/ml human recombinant FGF4 (Cell Concepts, Umkirch, Germany, C-10312M), 1 μg/ml Heparin (Sigma, H3149) and 75% feeder-conditioned medium).
Proliferation. Assays were performed using xCelligence system (OLS-OMNI Life Science, Bremen, Germany) system by seeding 1 × 10 5 cells per well in E-8 chambers (OLS-OMNI Life Science) in stemness medium. Cell indices were automatically recorded every 15 min. Relative proliferation velocity was calculated by using the slope.
Western blot analysis and immunofluorescence. Whole-cell lysate was extracted from TSCs as described. 30 Briefly, cells were lysed in SC buffer and protein extract supernatants were resolved using acrylamide gels and electrotransfered to PVDF membranes (Millipore, IPVH00010), according to the manufacturer's instructions. Membranes were incubated in blocking buffer for 1 h at room temperature, decorated with antibodies directed against Lsd1 (Schüle laboratory, Freiburg, Germany, 3544 (ref. Chemiluminescence was detected and images were acquired with a CHEMISMART documentation system and the Chemi-Capt 5000 software (Peqlab, Vilber Lourmat, Eberhardzell, Germany) and Image Quant TL software (General Electric, Munich, Germany). Quantitative analysis for normalization was performed with the Bio-1D software (Vilber Lourmat). The signal was normalized to the internal loading control and expressed relative to the sample control.
RNA extraction, qRT-PCR and RNA-seq. RNA was isolated with Trizol (Thermo Fisher Scientific, 15596), reverse transcription and qPCR were performed as described. 30 Transcription of Ppia, Rplp0 and Hprt was used for normalization and expression was related to the control sample. Experiments were repeated at least with three biological samples. Primers are listed in Supplementary Table S2. RNA-seq analysis was performed according to standard Illumina protocol. Raw sequence files (.fastq files) were aligned to mouse genome (mm10) using TopHat version 2. 65 Counting of the aligned reads was performed using HOMER software (analyzeRNA.pl). Differentially expressed genes were identified with EdgeR software. 66 ChIP-seq. TSCs cultured in stemness conditions were fixed with 1% PFA in 20 mM Tris-HCL pH 8, 0.5 mM EGTA, 1 mM EDTA, 10 mM NaCl for 5 min, quenched with glycine for 5 min and rinsed twice with ice-cold PBS. Cells were harvested in PBS supplemented with Complete EDTA-free protease inhibitors cocktail (Roche, Mannheim, Germany, 11873580001) pelleted and flash-frozen in liquid nitrogen. Chromatin preparation was done as previously described. 67 Fixed cells were resuspended in Farnham Lab buffer and nuclei were extracted using an ultrasound-based nuclei extraction method. Nuclei were pelleted, resuspended in shearing buffer and sonicated for 15 min using Covaris S220 (Brighton, UK, instrument settings: 140 W peak power, 5% duty factor, 200 cycles/burst, at 4°C), to a fragment size distribution of 100-800 bp.
ChIP was performed using the automated platform IP-Star SX-8G Compact (Diagenode, Oxford, UK) and Auto Histone ChIP-seq kit (Diagenode, C01010022) following the 'indirect method' procedure. ChIP reactions were incubated with 2 μg Lsd1 antibody 20752 (ref. 40) for 10 h, followed by 3 h of incubation with beads and 5-min washes at 4°C. Libraries were prepared from immunoprecipitated DNA using NEBNextUltra (Frankfurt, Germany) library preparation kit without size selection. ChIPseq libraries were sequenced using a HiSeq 2500 (Illumina, Cambridge, UK) and mapped to the mm10 reference genome using bowtie 2 (ref. 68). Data were further analyzed using the peak finding algorithm MACS 1.41 (ref. 69) using input as control. All peaks with FDR 41% were excluded from further analysis. The uniquely mapped reads were used to generate the genome-wide intensity profiles, which were visualized using the IGV genome browser. 70 HOMER 71 was used to annotate peaks. The genomic features were defined and calculated using Refseq and HOMER.
Metabolite analyses. TSCs were snap frozen in liquid nitrogen and stored at − 80°C. Sample preparation was performed according to metaSysX standard procedure, a modified protocol from Giavalisco et al. 72 Sample processing and measurement were performed with an ACQUITY RP-UPLC (Waters, Eschborn, Germany) coupled to an Exactive mass spectrometer (Thermo Fisher Scientific) GC-MS measurements were performed on a GC (Agilent Technologies, Waldbronn, Germany) coupled to a Pegasus HT mass spectrometer (Leco, Saint Joseph, MI, USA). UPLC-MS and GC-MS measurements were performed for the analysis of primary and secondary metabolites. After extraction from the chromatograms, the data were processed, aligned, filtered and annotated according to metaSysX standard procedures.
Cellular ROS content was measured by incubation with 5 μM H2DCF-DA (Thermo Fisher Scientific) for 20 min in phenol red-free medium. The cells were washed in PBS buffer, trypsinized and resuspended in 500 μl PBS.
Total glutathione (GSH) levels were tested using intracellular GSH assay kit (Abcam) following the manufacturer's instructions. Fluorescence intensity was visualized by flow cytometry using FL1 channel.
Biochemical assays. Quantification of NAD and NADP was performed according to the manufacturer's instructions (Abcam, ab65348) and (Abcam, ab65349), respectively. Briefly, 2 × 10 6 cells were suspended in respective assay extraction buffer by freeze/thaw cycles. The cells supernatants were deproteinized in 10 kD spin columns. NADPH/NADH levels were measured in the resulting samples by NADP/NAD heat decomposition (60°C for 30 min).
Ammonia content was measured using Assay kit (Abcam, ab83360), according to the manufacturer's instructions. Briefly, 2 × 10 6 cells were homogenized in assay buffer, after centrifugation the supernatants were diluted to fit standard curve and resolved in a microplate reader. Pyruvate control levels were subtracted for each sample.
ATP levels were measured using Assay kit (Abcam, ab83355), according to the manufacturer's instructions. Briefly, 2 × 10 6 cells were suspended by pipetting up and down in assay buffer. The samples were deproteinized by PCA precipitation and diluted at different concentrations to ensure reading within the standard curve and resolved by fluorometric assay.
Glutamate dehydrogenase activity was measured according to the manufacturer's instructions (Sigma, MAK099). Briefly, 1 × 10 6 cells were homogenized in assay buffer. After 10 min of incubation, the insoluble material was removed by centrifugation. The samples were measured by colorimetric assay and the Glud1 activity calculated by the ratio difference between the initial and final 450 nm absorbance.
For quantification of glutamine uptake, 1 × 10 6 TSCs were seeded. The medium was replaced and collected after 24 h, centrifuged and the glutamine levels in the supernatant were measured according to the manufacturer's instructions (Biovision, Wehrheim, Germany, K556).
Plasmids. Expression plasmids for SIRT4 were generated by BP and LR recombination according to the supplier (Thermo Fisher Scientific). The expression plasmid pDest26-SIRT4 (Source Bioscience, Berlin, Germany, IOHx59335) was used for the generation of a zeocin-resistant entry clone and expression plasmids containing mCherry or CFP reporters. The QuickChange II site-directed mutagenesis kit (Agilent Technologies) was used to delete the translational STOP codon of SIRT4 to generate a V5-tagged fusion protein. The redox potential was quantified by transfection of pLCPX-mito-Grx1-roGFP2, pLCPX-Grx1-roGFP2 (ref. 48). For determination of intracellular hydrogen peroxide concentration pLCPX-mito-Orp1-roGFP2 and pLCPX-mito-Orp1-roGFP2 were used. 48 Statistical analysis. If not otherwise stated, significance was calculated using an unpaired, two-tailed Student's t-test. Data are calculated as mean +S.E.M. Experiments were repeated at least three times. Principal component and cluster analysis were used as descriptive methods to confirm or reject discrimination between the sample groups. Overrepresentation of candidate genes in regards to classification was determined by binominal test.