LSD1-mediated demethylation of OCT4 safeguards pluripotent stem cells by maintaining the transcription of PORE-motif-containing genes

Reversible lysine methylation is essential for regulating histones and emerges to critically regulate non-histone proteins as well. Here we show that the master transcription factor OCT4 in pluripotent stem cells (PSCs) was methylated at multiple lysine residues. LSD1 that is highly expressed in PSCs can directly interact with and demethylate OCT4 at lysine 222 (K222) in the flexible linker region. Reduced LSD1 activity led to the methylation of OCT4-K222 that diminished the differentiation potential of PSCs while facilitating proteasome-independent degradation of OCT4 proteins. Furthermore, site-specifically replacing K222 with phenylalanine to mimic the constitutively methylated lysine promoted the ‘locked-in’ mode engagement of the OCT4 PORE-homodimers that tightly bind to and block the transcription of multiple PORE-motif-containing target genes regulating cell fate determination and cell junction organization, and thereby reducing the pluripotency of PSCs. Thus, LSD1-mediated demethylation of OCT4 plays a crucial role in restricting the ‘locked-in’ mode binding of OCT4 PORE-homodimers to the PORE-motif-containing genes and thereby maintaining their transcription to safeguard the pluripotency of PSCs.


K222 site of human OCT4 protein can be methylated in vitro and in vivo.
To systematically decipher and compare the PTM profiles of human OCT4 protein, we utilized a cell-free PTM system by incubating bacterially-expressed recombinant OCT4 proteins with the pluripotent NCCIT or differentiated U87 whole cell lysates, followed by purifying the lysate-reacted OCT4 proteins and analyzing their PTMs by mass spectrometry (MS) (Fig. 1A, Figure S1) 15,16 . After incubation with NCCIT cell lysates for 0 to 180 min, the recombinant OCT4 protein samples were subjected to SDS-PAGE and probed by immunoblotting with a pan-mono/di-methylated lysine antibody. Remarkably, recombinant OCT4 proteins were methylated in vitro by the NCCIT lysates in a time-dependent manner and the methylation reaction reached a saturated level at about 60 min of incubation (Fig. 1B), indicating the presence of mono-/di-methylated lysine residues in OCT4 proteins in the context of PSCs. Moreover, endogenous OCT4 proteins immunoprecipitated from NCCIT whole cell lysates can be recognized by an anti-mono-/di-methylated lysine [anti-pan K(me/me2)] antibody (Fig. 1D) and an anti-OCT4-pT235 antibody (Fig. 1E), further confirming the presence of methylated OCT4 proteins in PSCs.
Further MS analyses revealed that multiple lysine and arginine residues in the OCT4 protein can be monomethylated or di-methylated (Fig. 1C). Notably, mono-methylated K199 and R232, and di-methylated K231 and R232 were preferentially detected in NCCIT-incubated OCT4 proteins, while mono-methylated K128 and K222 were only detected in U87-incubated OCT4 proteins ( Figure S1 and S2). In the same set of experiments, other PTMs such as phosphorylation, acetylation and ubiquitination were also detected on multiple residues (Fig. 1C), consistent with the results obtained in previous studies. Comparing with other major PTMs, methylation is probably the least studied PTM for non-histone proteins including OCT4, and therefore, we decided to take a closer look at the methylated residues in the OCT4 protein. Given the crucial role of the linker region in regulating the configuration of OCT4 and its binding mode with various target DNA motifs as mentioned above, it was of particular interest to us that the highly conserved residue K222 is the only positively-charged residue within the entire linker region ( Fig. 1F and S4), and differentially methylated in the cellular context of PSCs (NCCIT) vs somatic cells (U87, 293 T) (Fig. 1C, Figure S1, S2, S3 and Table S1).

LSD1 is a bona fide demethylase for OCT4-K222.
In a new set of experiment, we systematically compared the methylation profiles of recombinant OCT4 proteins incubated with the whole cell lysates from PSCs (H1 hESC, NCCIT hECC) and somatic cancer cells (U87), with the untreated recombinant OCT4 proteins as control. Consistent with the above results, OCT4-K222 was demethylated in the cellular contexts of both ESCs and ECCs, while OCT4-K222 in U87 cells was mono-methylated and untreated recombinant OCT4 was dimethylated ( Fig. 2A, S1 and Table S1). Given OCT4-K222 was demethylated or methylated at very low levels in PSCs, we next searched for the responsible demethylase(s). As LSD1 was known to occupy the promoters of a subset of developmental genes that were co-occupied by OCT4 and NANOG in human ESCs 17 , it is a likely candidate that demethylates OCT4 protein. Consistent with this assumption, LSD1 was found highly expressed in H1 and NCCIT cells but barely detectable in U87 cells (Fig. 2B). Besides U87, another neuronal cell line U251 also had very low levels of LSD1 protein (Fig. 2C) and LSD1 mRNA (Fig. 2D), and there was no correlation between LSD1 and OCT4 expression at mRNA level ( Fig. 2D vs E) and protein level (Fig. 2C). When H1 or H9 hESCs were differentiated into neural stem cells (NSCs), there was a significant reduction of the LSD1 protein levels (Figs. 2F,G). Again, there was no correlation between LSD1 and OCT4 protein levels during neural differentiation (Figs. 2F,G). Taken together, we demonstrated that a low level of LSD1 expression in neuronal cells was associated with methylated OCT4-K222, while a high level of LSD1 in PSCs was associated with demethylated OCT4-K222, consistent with the hypothesis that LSD1 is a major demethylase for OCT4-K222. However,   www.nature.com/scientificreports/ even in the presence of LSD1, there are most likely other factors that can affect the probability and efficiency of OCT4-K222 demethylation, and therefore the LSD1 expression level may not always be correlated with the level of methylated OCT4-K222 in various cellular contexts. Next, we examined the potential physical interaction between LSD1 and OCT4 proteins via Co-IP and colocalization study. The results showed that LSD1 interacted with OCT4 endogenously in NCCIT cells (Fig. 2H) or when ectopically-expressed in 293 T cells (Fig. 2I), and they were partially co-localized in the nucleus of NCCIT cells (Fig. 2J). To determine if LSD1 is able to directly and site-specifically demethylate OCT4-K222, we incubated recombinant OCT4 proteins with recombinant LSD1 proteins in the presence of U87 cell lysate (to provide necessary reaction substrates and co-factors). MS analysis demonstrated that recombinant LSD1 proteins significantly reduced the peak of the OCT4 peptide spanning the mono-methylated K222 (Fig. 2K), suggesting it is a bona fide demethylase for OCT4-K222. As expected, when the U87 cell lysate was omitted, recombinant LSD1 proteins were unable to demethylate the di-methylated OCT4-K222 (Fig. 2L).
K222 methylation promotes proteasome-independent degradation of OCT4. Finely-tuned OCT4 protein levels are largely controlled by proteasome-dependent degradation pathway 18,19 . To assess the potential effect of LSD1-mediated demethylation on OCT4 protein turnover, a lentiviral-based shRNA was employed to reduce the endogenous LSD1 protein level in NCCIT cells that were pre-treated with MG-132, a specific and cell-permeable proteasome inhibitor, for 1 h prior to treatment with cycloheximide (CHX) to block synthesis of new proteins. CHX treatment revealed that OCT4 proteins in control cells infected with scramble shRNA had a half-life of approximately 4 h which was increased to 6 h by LSD1 shRNA (Fig. 3A). Strikingly, MG-132 failed to block the degradation of OCT4 proteins in LSD1-silenced cells despite it largely blocked OCT4 degradation in control cells (Fig. 3A), indicating that a lack of LSD1 can promote a proteasome-independent pathway for OCT4 protein degradation. Likewise, when tranylcypromine (2-PCPA) HCl, a specific LSD1 demethylase activity inhibitor, was used to treat NCCIT cells pre-treated with MG-132 and CHX, MG-132 also failed to block the degradation of OCT4 proteins (Fig. 3B). These results indicated that a sufficient LSD1 protein level is required for maintaining the dominant proteasome-dependent OCT4 degradation in PSCs.
To determine the kinetics of OCT4 protein turnover in a longer time frame, NCCIT cells pre-treated with 2-PCPA were exposed to CHX for 0 to 24 h. 2-PCPA treatment slowed down the turnover of both OCT4 and di-methylated H3K9/H3K4 proteins (Fig. 3C,D). To evaluate the effect of site-specific methylation of K222 on the turnover of the OCT4 protein, two site-specific mutants were constructed with the FLAG-OCT4-K222F to mimic the constitutively methylated K222, and the FLAG-OCT4-K222A to represent methylation-deficient K222. These two FLAG-tagged OCT4 mutants along with the FLAG-tagged wild type (WT) OCT4 were incorporated into a shOCT4-containing vector, respectively, and introduced into U87 cells by transfection. CHX treatment showed that, compared with the WT OCT4 proteins, OCT4-K222F proteins were more stabilized while OCT4-K222A proteins were degraded more rapidly (Fig. 3E,F). Like WT OCT4, both OCT4 mutants exhibited a predominant nuclear localization when ectopically-expressed in U87, NCCIT, 293 T ( Figure S5), or HeLa ( Figure S6) cells.
As for most proteins, lysosome/autophagy-dependent degradation pathway is the major proteasome-independent degradation pathway that can be blocked by chloroquine 15,20,21 , we pre-treated NCCIT cells with chloroquine combined with MG132 or 2-PCPA followed by treatment with CHX for 0 to 24 h. Chloroquine alone speeded up the degradation of OCT4 which can be prevented by MG-132 ( Fig. 3G), indicating that there is likely an interplay between the lysosome/autophagy-dependent and the proteasome-dependent degradation of OCT4, and blocking degradation via the former pathway may promote that via the latter one. Importantly, 2-PCPA significantly slowed down the expedited degradation of OCT4 by chloroquine (Fig. 3H,I and S7), further confirming that LSD1-mediated demethylation facilitates the proteasome-dependent degradation of OCT4 protein. Collectively, the enhanced OCT4-K222 methylation derived from either LSD1 silencing or inhibition of LSD1 demethylase activity increased the stability of OCT4 proteins and promoted their degradation via the lysosome/autophagy-dependent pathway.

OCT4-K222F promotes the 'locked-in' mode engagement of OCT4 PORE-homodimers and represses their transcriptional activities in vitro.
As stated above, OCT4 proteins can form monomers, PORE-and MORE-homodimers, or OCT4/SOX2 heterodimers to regulate the transcription of target genes harboring the MONO, PORE, MORE or SORE motifs, respectively 9,11,12 . A previous structural analysis indicated that positively-charged K222 (numbered as K85 in the structure) can form an intramolecular salt bridge with negatively-charged D166 (numbered D29 in the structure) that may help to stabilize the POREmode OCT4 homodimers (Fig. 4A) 12 . To evaluate the impact of K222 methylation on the configuration and transcriptional activity of OCT4, we first conducted the EMSA experiments to determine the in vitro binding among recombinant WT OCT4 protein and its variants (K222A or KA, K222R or KR, K222F or KF, and K222D or KD) with DNA probes containing various OCT4-binding motifs. Among all the tested OCT4 variants, the K222D, when present as monomers, had the lowest binding affinity to NANOG-and OCT4-SORE motifs (Fig. 4B,C and S8). When present as homodimers or heterodimers, K222D proteins minimally bound to the PORE motif ( Fig. 4C) but had no obvious impact on binding to the SORE or MORE motifs (Fig. 4B,C). In contrast, K222F homodimers significantly increased the binding to the PORE motif compared to WT and other OCT4 variants (Fig. 4C). These in vitro data strongly suggested that K222 is the crucial residue controlling the configuration of OCT4 homodimers, and constitutively methylated K222 can greatly promote the formation of PORE-mode OCT4 homodimers and strengthen their binding to the PORE genes.
To assess the transcriptional activities of WT OCT4 and the K222 variants, we generated constructs of luciferase reporters whose transcription is driven by three tandem repeats of one of the OCT4 binding motifs (designated as 3SORE, 3PORE, 3MORE and 3MONO, respectively). These OCT4-binding motif-luciferase reporter www.nature.com/scientificreports/ constructs were co-transfected with OCT4 variant genes into HeLa cells, and the mRNA levels of the transcribed luciferase as well as the OCT4 variants were determined by qRT-PCR. In the first set of experiment where different OCT4-binding motif-luciferase reporter constructs were co-transfected with the WT OCT4 construct only, the level of luciferase transcription driven by the 3MORE was the highest, followed by that of 3PORE, 3MONO, and 3SORE (Fig. 4D), suggesting that in the cellular context of somatic cells, OCT4 MORE-and POREhomodimers are probably the most favored configurations. Next, OCT4-K222A and OCT4-K222F proteins were compared with the WT OCT4 proteins for their capability of driving the transcription of the same OCT4binding motifs. There was no difference between WT OCT4 and the two K222 mutants in driving 3MONO-and 3SORE-mediated luciferase transcription, but both K222 mutants moderately reduced the 3MORE-luciferase transcription to the same extent (Fig. 4E). Most interestingly, compared with the WT OCT4, OCT4-K222F significantly reduced the 3PORE-mediated luciferase transcription by greater than 50% while the inhibitory effect of OCT4-K222A was less prominent (Fig. 4E). This observation indicated that constitutively-methylated K222 residue may stabilize the configuration of OCT4 PORE homodimers and lead to their prolonged binding to the PORE genes in a 'locked-in' mode and thereby suppressing the transcription of those genes.

OCT4-K222 methylation represses the transcription of PORE-motif-containing target genes in vivo and the differentiation capacity of PSCs.
To further explore the physiological functions of OCT4-K222 methylation in the context of PSCs, we transfected H9 human ESC cells with the shOCT4-FLAG-OCT4 constructs expressing WT OCT4, OCT4-K222A or OCT4-K222F, respectively, while knocking down endogenous OCT4 simultaneously. Compared with H9 cells transfected with the control vector and WT OCT4 which maintained compact colonies, K222A-, and particularly K222F-transfected H9 cells became flattened and the colonies were also loosened (Fig. 5A). Given K222 methylation promoted the formation and engagement of the OCT4 PORE-homodimers, we examined its impact on the transcription of PORE genes in PSCs. By nucleotide BLAST searching, we came up with a list of PORE genes that contain one or more PORE motif sequences (ATT TGA AAT GCA AAT) 9 (Table S2). The main functions of the PORE genes center on cell fate determination, cell junction/cytoskeleton organization and regulation of signaling pathways such as the Hippo, MAP, and calcium signaling pathways (Fig. 5B, Table S4). Remarkably, OCT4-K222F dramatically reduced the transcription of all the PORE genes in ESCs and OCT4-K222A also exhibited some suppressive effects ( Fig. 5C and S9), consistent with the results obtained in HeLa cells co-transfected with OCT4-K222 mutants and different OCT4binding motif-luciferase reporter constructs (Fig. 4E). As a control, we examined the transcription of multiple genes harboring the PORE-like-motif (ATT TGA AAGGCA AAT ) that only differs the PORE motif (ATT TGA AAT GCA AAT) by one single nucleotide (Table S2 and S6). Strikingly, the transcription of most PORE-likemotif-containing genes was only moderately inhibited by two OCT4-K222 mutants to the same extent ( Figure  S10), suggesting that the recognition of the PORE motif by the OCT4 PORE-homodimers is highly specific, and the suppression of the PORE genes by the OCT4-K222F mutant is likely to be specific and physiologically relevant. Next, we examined the transcription levels of some key stemness genes and three germ layer marker genes in the above H9 cells (female, XX) and H1 cells (male, XY) subjected to similar treatments. The expression of SOX2, NANOG, GATA4, GATA6, NKX3.1 and HES4 was reduced by OCT4-K222F and, to a lesser extent, by OCT4-K222A in H9 cells (Fig. 5D,E), consistent with the partially-differentiated morphology of the cells and colonies (Fig. 5A). In contrast, under the same conditions, only HES4 transcription was repressed in H1 cells (Fig. 5E). After searching for potential PORE-motif-like sequences in all the above stemness and germ layer marker genes, we could find no single sequence in any genes that completely matches that of the PORE motif (Table S3), indicating that the reduced expression of some genes seen in Fig. 5D,E is most likely a secondary event caused indirectly by OCT4-K222F.
Retinoic acid (RA) is a vitamin A metabolite that is essential for early embryonic development and promotes stem cell neural lineage specification at relatively high concentrations 22 . It is widely used for inducing    www.nature.com/scientificreports/ LSD1 inhibitors and RA for 1 h followed by adding in CHX and a further incubation for 12 h ( Figure S12A). Cells treated with the two LSD1 inhibitors showed similar patterns of OCT4 protein degradation and H3K9 methylation ( Figure S12B). CHX treatment per se appeared to elevate the expression of mesendoderm marker genes ( Figure S12C). Interestingly, the overall transcription patterns of the lineage-specific markers in NCCIT cells treated with the more potent LSD1 inhibitor OG-L002 were similar to those in H9 cells transfected with the OCT4-K222F mutant, supporting the assumption that OCT4-K222 methylation is likely a common target for these two experimental settings.

Discussion
Accumulating evidence indicates that the protein stability, transcriptional activity, subcellular localization, protein-protein interaction of OCT4 are comprehensively and delicately regulated by a variety of PTMs including phosphorylation 13,15,21,23 , O-GlcNAcylation 24 , sumoylation 25 , ubiquitination 19,26,27 , and acetylation 28 . Although methylation of non-histone proteins at lysine residues emerges to influence their stability and functionality, regulate their interaction with DNA and partner proteins, and prevent other PTMs at the same lysine residues 29 , systematic identification and in-depth functional characterization of methylated residues in OCT4 proteins is still lacking. In this study, we identified multiple methylated lysine and arginine residues in OCT4 proteins in the cellular contexts of PSCs and somatic cancer cells, and demonstrated that LSD1 can maintain the unmethylated status of the crucial OCT4-K222 residue to prevent the 'locked-in' mode engagement of the OCT4 PORE-homodimers and thereby allowing for the transcription of the PORE genes. LSD1 was the first identified histone lysine-specific demethylase that demethylates H3K4 or H3K9 in a reaction utilizing flavin adenosine dinucleotide (FAD) as a cofactor 30 . It represses a variety of genes via demethylation of H3K4me1/me2 30 . In human ESCs, high levels of LSD1 are required to suppress H3K4 methylation at the regulatory regions of differentiation genes and thereby maintaining their undifferentiated state, and downregulation of LSD1 during differentiation can favor the resolution of the bivalent domains towards H3K4 methylation and gene activation 17 . Interestingly, when forming a complex with androgen receptor, LSD1 changes its substrates to H3K9me2 and can regulate gene activation through demethylation of H3K9me1/me2 31 .
Besides histone family proteins, LSD1 is able to demethylate lysine residues at a growing number of nonhistone substrates 32 , such as p53 33 , DNMT1 34 , STAT3 35 , E2F1 36 , and HIF1α 37 . Given the finding that there is a significant overlap of genome-wide distribution of OCT4 binding sites with the LSD1 binding sites 17 , and OCT4 is associated with different subunits of the CoREST and NuRD complexes [38][39][40] , OCT4 may recruit LSD1 to target genes via direct or indirect interactions. In this study, we demonstrated direct interaction between LSD1 and OCT4 in PSCs, and the methylated OCT4-K222 site can be demethylated by LSD1 both in vitro and in vivo. Recently, it was shown that murine Oct4 can interact with Lsd1 to inhibit its catalytic activity in F9 ECCs, leading to retention of H3K4me1 and a "primed" chromatin state at pluripotency gene enhancers 41 . It remains to be seen if such inhibitory mechanism is also present in ESCs, and if this may represent a feedback regulation between OCT4 and LSD1.
Among multiple methyl-lysine sites identified in OCT4, K222 is of particular interest for the following reasons: firstly, it is situated at the highly flexible linker region (N213-A229) connecting the POU S domain with POU H domain that functions as a protein-protein interaction interface and plays a critical role during reprogramming by recruiting epigenetic modifiers to OCT4 target genes 12,42 . Secondly, it is highly conserved across many species despite its neighboring residues are not conserved. Thirdly, it is the only positively charged residue in the whole linker region. Based on the solved and modeled crystal structures 9,12 , the positively charged K222 (numbered as K85 in the cited reference) may be important for forming the intramolecular salt-bridge with the negatively-charged D166 (D29 in the cited reference) to facilitate forming the OCT4 PORE homodimers. Fourthly, although in this study methylation appears to be the only PTM for K222, the information in the PhosphoSitePlus database provided by Cell Signaling Technology, Inc. indicated that murine Oct4-K215 (corresponding to human OCT4-K222) can be ubiquitinated. Therefore, different PTMs may crosstalk at the OCT4-K222 site. For instance, methylation of K222 would be expected to block its ubiquitination, which may well explain our observation that K222 methylation induced by LSD1 ablation promotes the proteasome-independent degradation of OCT4, supporting the notion that methylation increases the stability of proteins by competing with ubiquitination 43 . As methylation of a lysine residue does not modify the side chain's positive charge 43 , the above-mentioned intramolecular salt-bridge between K222 and D166 is unlikely to be disrupted by K222 methylation. However, since the phenylalanine to mimic methylated lysine does not carry any charge, K222F mutant is expected to lose the salt-bridge interaction with D166. Nevertheless, our biochemical data indicated that K222F stabilized the OCT4 PORE-homodimers and strengthened their 'locked-in' mode engagement onto the PORE genes and thereby continuously suppressed their transcription. Further structural-based studies are warranted to better understand the detailed molecular mechanisms for such a 'locked-in' mode engagement. Paradoxically, K222A mutant that has reduced protein stability and diminished binding with the PORE genes also suppressed their transcription to some extent. These seemingly contradicting results may be reconciled by hypothesizing that a dynamic 'on-and-off ' mode binding of the OCT4 PORE-homodimers to the PORE genes is required for their normal transcription.
Since its first characterization over two decades ago 11 , the canonical PORE motif sequence (ATT TGA AAT GCA AAT) has been analyzed mainly for its interaction with the OCT4 homodimers structurally and biochemically 12,13,44,45 . Surprisingly, few studies have been reported to systematically investigate the physiological functions of the PORE genes and their regulation by OCT4 in PSCs. We show here that the PORE genes are mainly involved in cell fate determination, cytoskeleton/cell junction organization, and a variety of related signaling pathways that are crucial for PSC self-renewal. Importantly, we reveal that the 'locked-in' mode engagement of the OCT4 PORE-homodimers inhibits rather than activates the transcription of most PORE genes in PSCs, Scientific Reports | (2021) 11:10285 | https://doi.org/10.1038/s41598-021-89734-y www.nature.com/scientificreports/ and LSD1 can prevent such engagement by maintaining the un-methylated status of the crucial OCT4-K222 residue ( Figure S13). Taken together, we provide evidence in this study that LSD1-mediated demethylation of OCT4-K222 plays a crucial role in safeguarding the pluripotency of PSCs by restricting the 'locked-in' mode engagement of the OCT4 PORE-homodimers to maintain the transcription of the PORE genes. These results reveal a novel interplay between an epigenetic regulator and a pluripotency factor, and exemplify how a sitespecific PTM of OCT4 critically regulates the pluripotency of PSCs.

Methods
Cell culture, transfection and treatment. The sources and culture conditions for 293 T, HeLa, HepG2, U87, HT29, LO2, HUVEC, MCF7, H1, H9 and NCCIT cells were described in detail previously 16,46,47 . All cells were maintained in a 37 °C incubator with 5% CO2, and free of mycoplasma contamination based on routine tests every 3 months. H1 and H9 cells were cultured in mTeSR1 medium from STEMCELL (catalog no. 85850). 293 T, HeLa and U87 cells were transfected with GenEscort II Transfection Reagent reagents (catalog no. WIS 2100) as described from WISGEN. H1 and H9 cells were transfected with Lipofectamine 3000 Transfection Reagent (catalog no. L3000015) as described in the instructions provide by Thermo-Fisher. In some experiments, cells were treated with 20 μg/ml CHX  Plasmids and viral infection. The prokaryotic human His-OCT4 and His-SOX2 were generated by PCR and subcloned into the pET28a vector via EcoRI and SacI sites, and His-LSD1 (852 AA) was generated via NheI and XhoI sites. Lentiviral vector was constructed targeting the 3'-UTR of human POU5F1 mRNA to knock down the endogenous OCT4 while expressing the FLAG-tagged exogenous OCT4 WT (referred to as "shOCT4 + FLAG-POU5F1-WT") was described previously 15 , which was applied to infect and transfect the cells for 48-96 h as described in detail previously 15 . The OCT4 variants (K222A, K222R, K222F and K222D) were generated from OCT4 WT (His-OCT4 or shOCT4 + POU5F1) plasmids by site-directed mutagenesis and applied to the experiments. The FLAG-OCT4-6His portion was subcloned into the EcoRI and XbaI sites of the pLKO.1-TRC-shOCT4-FLAG-OCT4 plasmid 15 .
Immunoprecipitation and immunoblotting. Immunoprecipitation and immunoblotting were conducted as described previously 15,16 . 1 mg whole cell lysates (293 T and NCCIT) were incubated with 50 μl anti-FLAG M2 magnetic beads or 50 μl Pierce protein A/G agarose over 8 h at 4℃ respectively. 1 mg of whole cell lysates (U87, NCCIT and H1) was incubated with 100 μl of Ni-NTA agarose resin over 2 h at 4℃ respectively. Immunoblotting and Coomassie Brilliant Blue staining were performed after SDS-PAGE. In most cases, cropped immunoblot images were presented in the main figures, with their uncropped or partly-cropped images being shown in the Supplementary Information.
Immunofluorescence staining and co-localization study. Immunofluorescence staining was conducted as described previously 15 . Briefly, anti-OCT4 and anti-LSD1 antibodies were applied for immunofluorescence staining and their co-localization in NCCIT cells was determined by confocal microscopy (Carl Zeiss LSM-510).
Electrophoretic mobility shift assay, EMSA. EMSAs were carried out using a LightShift chemilumi- OCT4 SORE probe, Sense: 5'-biotin-CCG TCT TCT TGG CAG ACA GCA GAG AGA TGC ATG ACA AAG GTG CCG TGA TGG TTC  TGT CC-3' .  Anti-sense:5'-GGA CAG AAC CAT CAC GGC ACC TTT GTC ATG CAT CTC TCT GCT GTC TGC CAA GAA  GACGG-3' . OCT4 Quantitative real-time PCR, qRT-PCR. Quantitative real time PCR analysis and primers used for lineage marker quantitation were as described previously 15 . Firstly, total RNA was extracted by RNAiso Plus (TaKaRa, catalog no. 9109). Secondly, cDNA was synthesized using PrimeScript RT reagent kit with gDNA eraser (TaKaRa, catalog no. RR047A) according to the manufacturer's instructions. Thirdly, qRT-PCR was performed using the iTaq Universal SYBR Green Supermix (Bio-Rad, catalog no. 1721-5124) in an ABI 7500 Real-Time PCR instrument. All the PCR amplifications were performed in triplicates and repeated in three independent experiments. The relative quantities of mRNAs were normalized by the mRNA levels of the housekeeping gene PBGD (HMBS). The sequences of all RT-PCR primers were listed in Table S5.
Luciferase reporter-based OCT4 transactivity assay. The pGL6-TA vector (Beyotime, catalog no. D2105) was inserted with three tandem repeats of one of the OCT4 binding motifs (3SORE, 3PORE, 3MORE and 3MONO) via XhoI and HindIII sites. The resulting constructs together with WT OCT4 or OCT4 variant plasmids were co-transfected into HeLa cells, and the luciferase mRNA expression levels driven by the OCT4 binding motifs were determined by qRT-PCR and normalized by the GAPDH mRNA levels. The sequences of the three tandem repeats of OCT4 binding motifs were as follows: and SOX2 constructs were transformed into Rosseta E. coli strains and the purification procedures were as described previously 15 . After sonication, the 15 ml E. coli lysate was centrifuged at 12,000 g for 10 min at 4℃ and the supernatants were mixed with 1 ml of 50% (w/v) slurry of Ni-NTA Agarose resin/beads incubated at 4℃ on a rotary shaker for 2 h. The mixture was then centrifuged at 800 g for 2 min and the supernatant was discarded. The saving beads were washed with lysis buffer containing 50 mM imidazole and the proteins binding on Ni-NTA beads were used for subsequent in vitro PTMs assay (prokaryotic proteins eluted with the same buffer containing 200 mM imidazole for EMSA experiments).
Mass spectrometric identification of PTMs in OCT4 proteins. OCT4 samples were subjected to overnight digestion with trypsin or chymotrypsin as described by Liu et al. 48 . The peptides were extracted with acetonitrile containing 0.1% formic acid and vacuum dried. Proteolytic peptides were reconstituted with mobile phase A (2% acetonitrile containing 0.1% formic acid) and then separated on an on-line C18 column (75 μM inner diameter, 360 μM outer diameter, 10 cm, 3 μM C18). Mass spectrometry analysis was carried out on an LTQ-Orbitrap Velos mass spectrometer (Thermo Fisher Scientific, Waltham, MA, USA) operated in data dependent scan mode. Survey scan (m/z 375-1300) was performed at a resolution of 60,000 followed by MS2 scans to fragment the 50 most abundant precursors with collision induced dissociation. The activation time was set at 30 ms, the isolation width was 1.5 amu, the normalized activation energy was 35%, and the activation q was Scientific Reports | (2021) 11:10285 | https://doi.org/10.1038/s41598-021-89734-y www.nature.com/scientificreports/ 0.25. Mass spectrometry raw file was searched by Proteome Discovery version 1.3 using MASCOT search engine with percolator against the human ref-sequence protein database (updated on 07-04-2013). Phosphorylation of Ser/Thr and Tyr, acetylation of Lys, mono-methylation and di-methylation of Lys and Arg, and di-glycine modification of Lys (ubiquitination) were used as variable modifications. A filter of 90% peptide confidence was applied according to the Peptide-Prophet and Protein-Prophet parsimony algorithms. Fragment assignment of each modified peptide was subject to manual inspection and validation using the original tandem mass spectra acquired in profile mode using Xcalibur software 16 .
Statistical analyses. The band intensities of all the immunoblots were quantified by the ImageJ software and presented as means ± SD of triplicate measurements in one experiment representative of three similar ones. All RT-PCR quantitative data were presented as means ± SD of three independent experiments. The statistical significance of normally distributed data was evaluated using the two-tailed unpaired Student's test, and differences were considered significant at *P < 0.05 and **P < 0.01.