The assembly of mammalian SWI/SNF chromatin remodeling complexes is regulated by lysine-methylation dependent proteolysis

The assembly of mammalian SWI/SNF chromatin remodeling complexes is developmentally programed, and loss/mutations of SWI/SNF subunits alter the levels of other components through proteolysis, causing cancers. Here, we show that mouse Lsd1/Kdm1a deletion causes dramatic dissolution of SWI/SNF complexes and that LSD1 demethylates the methylated lysine residues in SMARCC1 and SMARCC2 to preserve the structural integrity of SWI/SNF complexes. The methylated SMARCC1/SMARCC2 are targeted for proteolysis by L3MBTL3 and the CRL4DCAF5 ubiquitin ligase complex. We identify SMARCC1 as the critical target of LSD1 and L3MBTL3 to maintain the pluripotency and self-renewal of embryonic stem cells. L3MBTL3 also regulates SMARCC1/SMARCC2 proteolysis induced by the loss of SWI/SNF subunits. Consistently, mouse L3mbtl3 deletion causes striking accumulation of SWI/SNF components, associated with embryonic lethality. Our studies reveal that the assembly/disassembly of SWI/SNF complexes is dynamically controlled by a lysine-methylation dependent proteolytic mechanism to maintain the integrity of the SWI/SNF complexes.

The mammalian SWI/SNF (mSWI/SNF, also called BRG1/BRM-Associated Factor, BAF) complexes are large ATP-dependent chromatin remodeling complexes assembled from up to 15 subunits with multiple paralogues during development to regulate stem cells, differentiation, and cell fate determination 1,2 . The developmentally programed mSWI/ SNF complexes assemble into cell/tissue-specific remodeling complexes by changing subunits or incorporating various subunit paralogues for the development of different tissues 3 . In pluripotent mouse embryonic stem cells (mESCs) and mouse F9 embryonic carcinoma cells, a unique mSWI/SNF complex, the esBAF complex, is assembled with limited key subunits, including BRG1 (SMARCA4 or BAF190A), the ATP-dependent chromatin remodeling catalytic subunit, SMARCC1 (BAF155 or SRG3), ARID1A (SMARCF1 or BAF250A), PBRM1 (Polybromo-1 or BAF180), and SMARCB1 (BAF47, INI1, or SNF5), to functionally associate with OCT4 and NANOG to regulate the central transcription circuitry of mESCs [4][5][6][7] . Loss of either BRG1 or SMARCC1 impairs the pluripotency and self-renewal of mESCs 4,5,7,8 , whereas ectopic expression of BRG1 or SMARCC1 in mouse embryonic fibroblasts (MEFs) significantly enhances SOX2, OCT4, and Myc-mediated reprogramming into induced pluripotent stem cells (iPSCs) 8 . When mESCs undergo differentiation, additional mSWI/SNF proteins, such as BRM (SMARCA2 or BAF190B), a BRG1 paralogue, and SMARCC2 (BAF170), a SMARCC1 paralogue, are co-expressed to form BRG1-and BRM-based SWI/SNF complexes 6 . In the mouse, homozygous deletions of Brg1 or Smarcc1 are early peri-implantation lethal during embryogenesis 9,10 , whereas Brm is dispensable and Smarcc2 null mutant animals die shortly after birth 11 . Recent human cancer exome and whole-genome sequencing studies revealed that loss or mutations of mSWI/SNF genes encoding various subunits are associated with more than 20% of human cancers 2,3 . The mutations or loss of a single mSWI/SNF subunit gene often uniquely cause the proteolysis of other subunits to assemble aberrant mSWI/SNF complexes with changes in stoichiometry and composition [12][13][14][15] . Notably, cancer cells defective in a particular mSWI/SNF subunit are often vulnerable to synthetic lethality of losing its remaining paralogue in the residual SWI/SNF complexes [13][14][15][16][17] . For example, cancer cells defective in BRG1 are usually sensitive to the removal of BRM, whereas cancer cells with ARID1A mutations are vulnerable to the loss of ARID1B [13][14][15][16][17] . However, little is known about how the assembly or disassembly of mSWI/SNF complexes and the proteolysis of subunits are regulated during development and tumorigenesis.
Lysine methylation is an important post-translational protein modification and lysine residues in proteins can be mono-, di-, and trimethylated. Extensive research has established the key roles of various methylated lysine residues at the amino-terminal regions of histones in modulating chromatin structure and gene expression 18,19 . While trimethylations at lysines 9 and 27 in histone H3 (H3K9me3 and H3K27me3) are usually associated with repressive chromatin structure that suppresses gene expression, the trimethylation at lysine 4 in histone H3 (H3K4me3) typically associates with active transcription 18 . LSD1 (also called KDM1a) was originally identified as a histone demethylase that specifically removes methyl groups from the mono-and dimethylated lysine 4 in histone H3 (H3K4me1/me2), but not trimethylated H3K4, to repress transcription 18,20 . Null deletion of the mouse Lsd1 gene causes early embryonic lethality around embryonic day 5.5 (E5.5) 21,22 . Loss of LSD1 also profoundly impairs the pluripotency and self-renewal of ESCs and the differentiation of other stem/progenitor cells, such as mouse hematopoietic stem cells [23][24][25][26] . The tamoxifen-induced conditional null deletion of the floxed Lsd1 mice (4-9 weeks old) by actin-Cre-ER (CAGGCre-ER) in the hippocampal and cerebral cortex neurons leads to paralysis with widespread hippocampus and cortex neurodegeneration, as well as learning and memory defects 27 . However, the molecular targets of LSD1 deficiency that cause the devastating animal phenotypes remain unclear.

Results
To understand the molecular mechanism by which Lsd1 null deletion causes embryonic lethality and other devastating effects in animals, we bred the floxed Lsd1 conditional deletion mouse strain 24 with the nestin-Cre transgenic mice 42 to specifically delete Lsd1 in the neuronal and glial cell precursors of the central nervous system. We found that the nestin-Cre-mediated homozygous loss of Lsd1 conditional alleles caused postnatal lethality immediately after birth (P0, Fig. 1a). During the characterization of Lsd1 mutants, we repeatedly observed that the protein levels of several subunits of mSWI/SNF chromatin remodeling complexes, such as BRG1, SMARCC1, SMARCC2, ARID1A, and PBRM1, are dramatically reduced in the brain extracts of Lsd1 null mutants, but not in the extracts from the body parts that do not express nestin-Cre, as compared with that of wildtype littermates ( Fig. 1b and Supplementary Fig. 1b). The dramatic reduction of these mSWI/SNF subunit protein levels occurs post-transcriptionally, as there is no significant correlative differences at the mRNA levels of these mSWI/SNF subunits between the Lsd1 null mutants and the wildtype littermates in the brain tissues ( Supplementary Fig. 1a). Consistently, our immunostaining of brain sections from the wildtype and Lsd1 null mutant animals revealed that the protein level of SMARCC1, a core mSWI/SNF component 14,[43][44][45] , is markedly reduced in the Lsd1 mutant mice (Fig. 1c). To rule out potential degradation of mSWI/SNF proteins caused by animal lethality, we bred Lsd1 conditional deletion mouse strain 24 with a transgenic mouse line expressing a tamoxifen inducible Cre-ER recombinase under the Actin promoter control (CAGGCre-ER) 46 to establish mouse embryonic fibroblasts (MEFs) from the homozygous floxed LSD1 conditional (fl/fl) deletion mouse embryos with the actin-Cre-ER (CAGGCre-ER/Lsd1fl/fl). While these MEFs normally express significant levels of mSWI/SNF proteins, induced deletion of Lsd1 in the CAGGCre-ER/Lsd1fl/fl MEFs by addition of 4-hydroxytamoxifen (4-OH-Tam) led to the rapid disappearance of these mSWI/SNF proteins (Fig. 1d). The 4-OH-Tam induced downregulation of mSWI/SNF proteins was reversed by the treatment of these Lsd1 deficient MEFs with 26 S proteasome inhibitor, MG132 (Fig. 1d). To further confirm that the protein stability of mSWI/SNF subunits is dependent on Lsd1, we treated the CAGGCre-ER/Lsd1fl/fl MEFs with a specific LSD1 inhibitor, CBB3001 that we previously developed 41,47 . The protein levels of mSWI/SNF subunits decreased substantially within 10 h of CBB3001 treatment ( Supplementary Fig. 1c). Our studies indicate that LSD1 is required to maintain the protein stability and integrity of mSWI/SNF complexes.

SMARCC1 and SMARCC2 are targets of LSD1
Mutations or loss of a single critical mSWI/SNF subunit, such as SMARCB1 in malignant rhabdoid tumors and epithelioid sarcoma, ARID1A in a wide variety of cancers including ovarian clear cell carcinoma, endometrioid carcinoma, neuroblastoma, and bladder cancer, BRG1 in medulloblastoma, breast and lung cancers, SMARCC1 in small cell lung cancer, and SMARCC2 in pancreatic cancer, uniquely causes the proteolysis of other subunits 2,3,12-15 . Recent studies indicate that the structural integrity of the mSWI/SNF complexes requires the formation of homo-or heterodimer of SMARCC1 and SMARCC2 43 , and loss of SMARCC1 and/or SMARCC2 causes the disassembly of mSWI/SNF complexes and proteolytic degradation of many other mSWI/SNF subunits 15,44,45 . As LSD1 serves as a demethylase for DNMT1, E2F1, and SOX2 25,26,41 , we wondered whether mSWI/SNF proteins are regulated by lysine methylation. While lysine methylations of mSWI/SNF proteins have not been reported, our protein sequence examination revealed that SMARCC1 and SMARCC2 both contain several H3K4-like lysine residues that may be monomethylated by the SET7 methyltransferase (see below). To test whether SMARCC1 or SMARCC2 is regulated by SET7, we used siRNA-mediated silencing of SET7 with two representative siRNAs 41 and found that loss of SET7 indeed increased the levels of both SMARCC1 and SMARCC2 proteins in human cervical carcinoma HeLa cells ( Fig. 1e and Supplementary Fig. 1d). In addition, silencing of LSD1 with two representative siRNAs 41 reduced the protein levels of SMARCC1 and SMARCC2, and co-silencing of SET7 and LSD1 effectively re-stabilized SMARCC1 and SMARCC2 proteins in LSD1 deficient cells ( Fig. 1e and Supplementary Fig. 1d). To examine whether loss of LSD1 or SET7 affects the protein stability of SMARCC1 and SMARCC2, siRNA-transfected cells were treated with protein synthesis inhibitor cycloheximide to block translational initiation to measure protein decay rates 48 . We found that the half-lives of SMARCC1 and SMARCC2 proteins were reduced after LSD1 silencing. However, cosilencing of LSD1 and SET7 restored the half-lives of both SMARCC1 and SMARCC2 proteins analyzed by measuring the steady-state protein stability ( Supplementary Fig. 2). We also stably expressed Flagtagged SMARCC1 or SMARCC2 under the retroviral LTR promoter control 41 in human lung carcinoma H1299 cells and found that the ectopically expressed Flag-SMARCC1 and SMARCC2 are also sensitive to LSD1 silencing (see below). The downregulation of SMARCC1 protein induced by LSD1 silencing is also reversed by 26 S proteasome inhibitor, MG132 (Fig. 1f) 25,26 . To exclude the potential off-target effect of siRNAs, we ectopically expressed Flag-tagged LSD1 cDNA that does not contain the 3′ untranslated region (UTR), and then silenced the expression of endogenous LSD1 using the siRNA targeting the 3′UTR region of the endogenous LSD1 gene in H1299 cells. We found that the ectopically expressed Flag-LSD1 cDNA suppressed the 3'UTR silencing effects of endogenous LSD1 on SMARCC1 and SMARCC2 proteins (Fig. 1g). These observations indicate that loss of LSD1 induces the proteolysis of SMARCC1 and SMARCC2 proteins through the ubiquitin-dependent pathway.

L3MBTL3 regulates the proteolysis of SMARCC1 and SMARCC2
Since LSD1 is a demethylase, it is possible that SMARCC1 and SMARCC2 might be lysine methylated to trigger their proteolysis. We have recently shown that the monomethylated lysine residues in DNMT1, E2F1, and SOX2 are recognized by L3MBTL3, a Malignant Brain Tumor (MBT) domain protein that specifically binds to methyl-lysine residues 49 , to target substrate proteolysis by the CRL4 DCAF5 ubiquitin   Fig. 2c and Supplementary Fig. 3c). Moreover, our immuno-co-precipitation analysis revealed that endogenous L3MBTL3 and SMARCC1 proteins reciprocally interact with each other (Fig. 2f), and this interaction is significantly enhanced in the presence of wild-type SET7, but not when a methyltransferasedeficient mutant of SET7, in which the critical histidine 297 is converted to alanine (H297A) 37 , is expressed (Fig. 2g). Conversely, downregulation of SET7 greatly reduced the interaction between L3MBTL3 and SMARCC1 ( Supplementary Fig. 4a). In addition, while LSD1 silencing induced the degradation of ectopically expressed Flag-SMARCC1 protein, such degradation was prevented by cosilencing of L3MBTL3 or DCAF5 and LSD1 (Figs. 2d, e and Supplementary Fig. 3a, b). Either the expression of SET7 or the treatment of LSD1 silencing can also stimulate SMARCC1 polyubiquitination, which is further promoted by the co-expression of SET7, L3MBTL3, and DCAF5 ( Fig. 2h and Supplementary Fig. 4b).
Our protein sequence examination revealed that SMARCC1 contains three lysine residues, Lys201 (K201), Lys482 (K482), and Lys615 (K615), whereas SMARCC2 also has three lysine residues, Lys328 (K328), Lys457 (K457), and Lys592 (K592), in the putative H3K4-like methylation consensus motifs (Fig. 3a) 25,26,28,41 . Among these lysine residues, the K482 or K615 motif in SMARCC1 is conserved (K482) or partially conserved (K615) in the K457 or K592 motif of SMARCC2, respectively. To test whether these putative methylated lysine residues in SMARCC1 serve as substrates for L3MBTL3 and LSD1, we synthesized monomethylated K482 and K615 peptides and their unmethylated cognate peptides derived from SMARCC1 (Fig. 3b). We found that in vitro translated and 35 S-labeled L3MBTL3 protein selectively and directly binds to the methylated K482 and K615 peptides immobilized onto Sepharose beads 41 , but not to their cognate unmethylated peptide beads (Fig. 3b). We also examined whether LSD1 can remove the methyl groups from the synthetic monomethylated K482 and K615 peptides by mixing the methylated peptides with the purified recombinant GST-LSD1 fusion protein or GST control 25,41 . We found that GST-LSD1 can effectively demethylate the methylated K482 or K615 peptides (Figs. 3c, d, and Supplementary Fig. 4c). We next investigated the requirement of SET7 for SMARCC1 methylation in vivo 26,41 . While the affinity-purified anti-methylated K615 antibody selectively recognized a 155 kilodalton (kDa) protein corresponding to the size of SMARCC1 protein in the total HeLa cell lysate, silencing of SET7 greatly reduced the intensity of this protein, suggesting that it is the SMARCC1 protein methylated at K615 by SET7 (Fig. 3e). However, our affinity-purified anti-methylated K482 antibody produced a high background in cell lysates so we could not directly evaluate the methylated K482 levels in various cells.
To define the critical lysine residues in SMARCC1, we mutated K201, K482, or K615 of SMARCC1 to arginine to produce K201R, K482R, and K615R mutants either singly or in combination so that they cannot be methylated by SET7 (Fig. 3a). Expression of K201R/K482R or K201R/ K615R double mutants indicate that the presence of either K615 or K482 is sufficient for SMARCC1 to interact with L3MBTL3 ( Fig. 4a, b). However, the K201R/K482R/K615R triple mutant (TM) of SMARCC1 failed to interact with L3MBTL3 (Fig. 4c). To further test if SET7 promotes the methylation of SMARCC1 in vivo, we co-transfected either the wild-type GFP-SMARCC1 or the GFP-SMARCC1 triple-mutant and SET7, and we found that only the wild-type SMARCC1, but not the SMARCC1 triple-mutant, was methylated by SET7 ( Supplementary  Fig. 4d). Consistently, while the K201R/K482R and K201R/K615R double mutants of SMARCC1 are still sensitive to LSD1 silencing, the K482R/K615R double mutant and in particular the K201R/K482R/ K615R triple mutant are much more resistant to LSD1 silencing (Fig. 4d, e and Supplementary Fig. 5a, b). Furthermore, the K201R/K482R/ K615R triple mutant is also resistant to be polyubiquitinated by SET7, L3MBTL3, and DCAF5 (Fig. 2h). Our results indicate that the methylated SMARCC1 protein is directly recognized by L3MBTL3 to promote its ubiquitin-dependent degradation by the CRL4 DCAF5 ubiquitin E3 ligase and that LSD1 acts as a demethylase to remove the methyl groups from methyl-lysine 482 and methyl-lysine 615 of SMARCC1 to prevent the proteolysis of SMARCC1 by L3MBTL3 and the CRL4 DCAF5 ubiquitin E3 ligase complex.
Since SMARCC2 also contains lysine residues in the putative H3K4-like methylation motifs (Fig. 3a), our studies revealed that the binding of SMARCC2 to L3MBTL3 is stimulated by SET7 (Fig. 4f). We also constructed the lysine-to-arginine mutations of either K457 (K457R) alone, which is identical to K482 in SMARCC1, or the triple mutations of K328R/K458R/K592R of SMARCC2 (Fig. 3a). We found that the K457R single and the K328R/K457R/K592R triple mutants (TM) of the Flag-tagged SMARCC2 are much more resistant to the loss of LSD1, as analyzed by direct Western blotting or by measuring the steady-state protein stability of SMARCC2 and its mutants after LSD1 silencing (Fig. 4g, h and Supplementary Figs. 5c, d, 6b, c). Our studies indicate that LSD1 and L3MBTL3 act through the H3K4-like lysine residues, such as K457 of SMARCC2, to regulate SMARCC2 protein stability.

Regulation of SMARCC1 and mSWI/SNF complexes in embryonic stem cells
Mouse embryonic stem cells (mESCs) and mouse F9 embryonic carcinoma cells uniquely express a specific mSWI/SNF complex, the esBAF . The genomic DNA from the dissected brains from LSD1 flox/+ , LSD1 flox/flox , and LSD1 flox/flox /Nestin-Cre mutant mice immediately after birth were extracted and used for genome-typing. b Brain extracts from LSD1 flox/+ , LSD1 flox/flox , and LSD1 flox/flox /Nestin-Cre mice were Western blotted for indicated SWI/SNF proteins. Protein molecular weight markers are in kilodaltons (kDa). c Representative brain cryo-sections of postnatal LSD1 flox/flox and LSD1 flox/flox /Nestin-Cre mice (P0,10-µm thick coronal) were stained with anti-SMARCC1 antibody, counterstained with 4′,6-diamidino-2-phenylindole (DAPI). Scale bar: 500 μm. d Embryonic fibroblasts from CAGGCre-ER TM /LSD1 fl/fl mouse embryos (E13.5) were treated with 4-hydroxytamoxifen (20 μg/ml) for 12 h to delete Lsd1 by inducible Actin-Cre-ER. Dimethyl sulfoxide (DMSO) or proteasome inhibitor MG132 (5 μg/ml) was added for the last 6 h before lysing the cells. e HeLa cells were transfected with 50 nM siRNAs of luciferase (Luc), LSD1, LSD1 + SET7 (L + SET7), or SET7 for 48 h and representative Western blots from three independent experiments are shown with Actin as a loading control. Band intensities were quantified by ImageJ software and normalized to the Luciferase (Luc) siRNA control signal. Significance was indicated as a two-tailed, unpaired, t-test. Values are expressed as the mean ± SEM. ***p < 0.001. f H1299 cells stably expressing Flag-tagged SMARCC1 under the retroviral LTR promoter control were transfected with 50 nM of luciferase and LSD1 siRNAs for 48 h and then treated with either dimethyl sulfoxide (DMSO, 1%) or 5 μg/ml MG132 for last 6 h before lysing the cells for Western blotting. g H1299 cells with or without stably expressing Flag-tagged LSD1 cDNA under the retroviral LTR promoter control were transfected with 50 nM of luciferase and LSD1-3'UTR siRNAs for 48 h and a representative blot set is shown for Western blotting. Source data are provided as a Source Data file. Significance was indicated as a two-tailed, unpaired, t-test. Values are expressed as the mean ± SEM. ***p < 0.001. The experiments for a-d and f were repeated three times with the same results.
complex, that only contains SMARCC1 but not SMARCC2 4-7 (Supplementary Fig. 7e). We found that siRNA-mediated silencing of LSD1 caused the proteolysis of SMARCC1 and greatly impaired the selfrenewal and pluripotency of mESCs, indicated by the loss of alkaline phosphatase (AP) activity (Fig. 5a, b). The co-silencing of L3MBTL3 and LSD1 restored the SMARCC1 protein level and rescued the defects on self-renewal and pluripotency in LSD1 deficient mESCs (Fig. 5a), indicating that L3MBTL3 mediates the proteolysis of SMARCC1 in mESCs 26 . LSD1 silencing also blocked the proliferation of F9 cells, induced SMARCC1 proteolysis, and promoted the reduction of other mSWI/ , and other indicated mSWI/SNF proteins (c) were analyzed by Western blotting. Band intensities were quantified and normalized to the Luciferase (Luc) siRNA control signal, as described in Fig. 1e SNF proteins (Fig. 6a, b and Supplementary Fig. 7c, d). The effects of LSD1 silencing are unique for mESCs and F9 cells, as silencing of LSD1 in human embryonic kidney carcinoma 293 T and lung carcinoma H1299 cells that express both SMARCC1 and SMARCC2 did not exhibit any proliferation defects ( Supplementary Fig. 7c-e). While our studies and other reports indicate that LSD1 plays an essential role in both human and mouse ESCs 22,23,25,26 , the mechanism by which LSD1 regulates ESCs remains unclear. We found that SMARCC1 silencing using two representative siRNAs impairs the selfrenewal of both mESCs and F9 cells but not the proliferation of 293 T or H1299 cells 4,5 (Supplementary Figs. 7c-e, 8e), similar to the selective effects of LSD1 silencing in these cells. To analyze the role of SMARCC1 in mESCs and F9 cells, we ectopically and stably expressed Flag-SMARCC1 and its lysine-to-arginine mutants using the pMSCV-Puro retroviral LTR vector in both mESCs and F9 cells. Strikingly, we found that stable expression of the K201R/K482R/K615R triple mutant of Flag-SMARCC1, but not the wildtype counterpart, is sufficient to confer the resistance to LSD1 silencing to maintain the pluripotency and selfrenewal in mESCs (Fig. 5b, c). Similarly, the expression of Flag-SMARCC1 triple mutant, as well as the K482R/K615R double mutant, in F9 cells are sufficient to render the resistance to LSD1 silencing to prevent the degradation of the mSWI/SNF proteins and growth arrest The consensus lysine residues (K*) methylated by SET7 in H3K4-like methylation motifs with the R/K-S/T/V-K* consensus sequences. b L3MBTL3 preferentially binds to the monomethylated K482 and K615 peptides. Sulfolink-agarose immobilized SMARCC1 peptides containing the monomethylated K482 and K615, or cognate unmethylated peptides were incubated with 20 μl of in vitro translated and 35 S-methionine labeled L3MBTL3 for 1 h at room temperature. L3MBTL3 binding to the peptide resins was analyzed by autoradiography after protein electrophoresis separation. c, d LSD1 demethylates methylated K482 and K615 in SMARCC1. Purified 1 μg of GST (spot 1) or GST-LSD1 proteins (spot 2) were incubated with 100 ng of monomethylated K482 (c) or monomethylated K615 (d) peptides for 4 h at 37 o C and the resulting peptides and control input methylated peptides (spots 3 and 4) were spotted onto a nitrocellulose membrane. The methylated peptides were detected by affinity-purified anti-monomethylated K482 (c) or K615 (d) peptide antibodies (see Supplementary Fig 4c) as indicated. e HeLa cells were transfected with 50 nM siRNAs of luciferase or SET7 for 48 h and the K615-methylated SMARCC1, SMARCC1, and other proteins were analyzed by Western blotting analysis using indicated antibodies. Source data are provided as a Source Data file. (Fig. 6a, b and Supplementary Fig. 7a, b). Thus, the effect of LSD1 silencing appears to resemble the loss of SMARCC1 in mESCs and F9 cells, since the loss of either LSD1 or SMARCC1 blocked cell proliferation, induced the disassembly of the mSWI/SNF complex, and the degradation of mSWI/SNF subunits, as analyzed by the immuno-coprecipitation of the mSWI/SNF complex by anti-SMARCB1 antibodies (Figs. 5b, c, 6a-d). We further investigated whether ectopic expression of SMARCC2 affects the sensitivity of LSD1 deficiency in these pluripotent cells. The Flag-SMARCC2 in the pMSCV-Puro retroviral vector was stably expressed in F9 cells and the response to LSD1 inactivation was analyzed. We found that the ectopic and stable expression of the Flag-SMARCC2 and its triple mutant cannot confer resistance to LSD1 deficiency-induced growth inhibition and SMARCC1 degradation (Fig. 6e). These studies suggest that SMARCC1 is a key target of LSD1 in both mESCs and F9 cells and SMARCC2 cannot substitute SMARCC1 for this critical function. Notably, our analysis revealed that the silencing of either LSD1 or SMARCC1 also induced the expression of endogenous SMARCC2 in F9 cells or mESCs ( Supplementary Fig. 7d), suggesting that loss of SMARCC1/LSD1 and pluripotency might induce the expression of endogenous SMARCC2 in the LSD1-or SMARCC1deficient pluripotent cells. In contrast, silencing of LSD1 or SMARCC1 caused the downregulation of SMARCC2 in both 293 T and H1299 cells ( Supplementary Fig. 7d), indicating that the regulation of mSWI/SNF complex in mESCs and F9 cells is very distinct from that of 293 T and H1299 cells and that the expression of SMARCC2 alone may not be the only factor that affects the resistance of 293 T and H1299 cells to LSD1 silencing ( Supplementary Fig. 7c-e).
The mSWI/SNF complex is essential for mESCs and F9 cells through their regulation of the central OCT4-SOX2-NANOG transcription circuitry [4][5][6][7] . Indeed, silencing of SMARCC1 caused the   downregulation of OCT4 and SOX2, and the upregulation of NANOG proteins in mESCs (Fig. 5d). We have previously shown that loss of LSD1 causes the proteolysis of SOX2 protein through the L3MBTL3 and CRL4 DCAF5 complex 25,26 . Notably, while LSD1 silencing reduced SOX2 levels in mESCs or F9 cells expressing the wildtype Flag-SMARCC1, expression of the K482R/K615R double or the triple mutant of Flag-SMARCC1 significantly reversed the downregulation of SOX2 after LSD1 silencing in mESCs and F9 cells ( Fig. 5c and Supplementary  Figs. 6a, 7a, b). We further found that SMARCC1 and SOX2 interact with each other in mESCs (Fig. 5e). We also ectopically and stably coexpressed Flag-SOX2 in the pMSCV-zeocin vector and Flag-SMARCC1 or its triple mutant in the pMSCV-puromycin vector in F9 cells (Fig. 6f). We found that in F9 cells expressing the Flag-SMARCC1 triple mutant, but not the wildtype Flag-SMARCC1, the protein levels of Flag-SOX2 or the endogenous SOX2 became highly increased, although both forms of SOX2 proteins remain sensitive towards the loss of LSD1 (Fig. 6f). Intriguingly, the effect of SMARCC1 triple mutant on SOX2 is selective, as DNMT1, another LSD1 demethylase substrate, remained to be downregulated in response to LSD1 silencing in cells expressing either Flag-SMARCC1 or Flag-SMARCC1 triple mutant (Fig. 6f). These results demonstrate that the SMARCC1 triple mutant may selectively stabilize SOX2 protein, suggesting that SMARCC1 is a critical target of LSD1 in regulating the assembly of mSWI/SNF complex and the SOX2 protein levels to maintain the self-renewal and pluripotency of mESCs and F9 cells.

Cell cycle regulation of mSWI/SNF proteins by L3MBTL3
We repeatedly found that aphidicolin, a DNA polymerase inhibitor that arrests replicating cells at the G1/S border and in the S phase, reduced the steady-state levels of SMARCC1 and BRG1 in mESCs, as compared to asynchronously growing cells (Fig. 5f, g). The aphidicolin-arrested cells are associated with reduced levels of SOX2 and increased levels of cyclin E (Fig. 5f, g), a late G1 and early S phase cell cycle regulator, in mESCs 41 . Notably, in aphidicolin-arrested S phase mESCs, L3MBTL3 silencing is sufficient to restore the levels of SMARCC1, BRG1, and SOX2 proteins (Fig. 5g). To characterize potential cell cycle regulation of mSWI/SNF complexes, we synchronized HeLa cells at the G1/S border by the sequential thymidine/ aphidicolin treatment 41 , and measured the levels of mSWI/SNF proteins after releasing the synchronized cells into S phase in fresh media without aphidicolin. Our studies revealed that the protein levels of SMARCC1, SMARCC2, BRG1, PBRM1, and ARID1A are relatively low in the synchronized G1/S and early S phase cells, and their levels become gradually increased as cells progress towards late S phase and/or G2 phase ( Supplementary Fig. 9a, b). Since aphidicolin inhibits DNA polymerases, we also synchronized HeLa cells in mitosis by nocodazole, an inhibitor for the polymerization of microtubules. The mitotic cells were collected, washed, and released into fresh culture media without nocodazole to allow these mitotic synchronized cells to progress into G1, S, G2, and M phases 50 . We subsequently measured the levels of mSWI/SNF proteins at various times after mitotic release. Our studies showed that the protein levels of SMARCC1, SMARCC2, BRG1, PBRM1, and ARID1A are dynamically regulated in the cell cycle and these proteins were reduced to low levels in the S phase cells (Supplementary Fig. 10a, b). These observations are consistent with our previous findings that L3MBTL3-regulated DNMT1 degradation increases in the S phase and that SOX2 is a substrate of L3MBTL3 25,26,41 . Our findings are also consistent with previous observations that SMARCC2 protein is undetectable in the S and G2 phases of proliferating embryonic cortical progenitor cells during mouse cortex development between E10.5-E14.5 15 . Together, these observations suggest that the levels of mSWI/SNF proteins are regulated in the cell cycle and that L3MBTL3 plays a role in the regulation of mSWI/SNF proteins in the S phase of the cell cycle.

Control of subunit deficiency-induced SMARCC1/SMARCC2 proteolysis
It is well established that the subunit stoichiometry and composition of mSWI/SNF complexes are uniquely regulated that mutation/loss of a particular subunit, such as SMARCC1, SMARCC2, SMARCB1, BRG1, or ARID1A, destabilizes the mSWI/SNF complexes to induce proteolysis of other subunits via an unknown mechanism (Fig. 7a) 14,44,45,51 . Indeed, our studies showed that silencing of either SMARCC2 or SMARCB1 triggered the proteolysis of SMARCC1 and other mSWI/SNF proteins such as PBRM1 (Fig. 7b, c, f). We wondered whether L3MBTL3 and CRL4 are involved in SMARCC1 degradation during the mSWI/SNF disassembly induced by the loss of SMARCC2 or SMARCB1. Our tests revealed that SMARCC1 degradation, induced by SMARCC2 silencing, can be effectively reversed by MLN4924, a protein neddylation inhibitor that blocks the assembly of CRL ubiquitin ligase complexes including CRL4 52 (Fig. 7b). Importantly, we repeatedly found that while silencing of SMARCC2 or SMARCB1 triggers the degradation of the wildtype Flag-SMARCC1 protein, the Flag-SMARCC1 triple mutant is resistant to the loss of either SMARCC2 or SMARCB1 (Fig. 7c, f and Supplementary  Fig. 8a, c, d), and that the SMARCC1 triple mutant expression also stabilized other mSWI/SNF proteins such as PBRM1 (Fig. 7f). If SMARCC1 proteolysis during the silencing of SMARCC2 is mediated by L3MBTL3, we would expect that the interaction between SMARCC1 and L3MBTL3 to increase. Indeed, we found that SMARCC2 silencing enhanced the binding of SMARCC1 to L3MBTL3 (Fig. 7d) and silencing of SMARCC2 also enhanced the interaction between SMARCC1 and SET7 (Fig. 7e), suggesting that the released SMARCC1 after SMARCC2 silencing and subsequent disassembly of mSWI/SNF complexes may facilitate SMARCC1 methylation by SET7 and its subsequent binding to L3MBTL3. Conversely, while SMARCC1 silencing induces proteolysis of the wildtype Flag-SMARCC2 14,44,45 , the K457R and the triple mutants of Flag-SMARCC2 are more resistant to SMARCC1 silencing ( Fig. 7g and Supplementary Fig. 8b). Thus, our results reveal that the proteolysis of SMARCC1 or SMARCC2 triggered by the loss of individual subunit of mSWI/SNF complexes is regulated by the L3MBTL3-CRL4 DCAF5 dependent pathways.

Regulation of SMARCC1/SMARCC2 by L3MBTL3 during development
Human L3MBTL3 is mutated in medulloblastoma and is further implicated in other pathological disorders such as multiple sclerosis, insulin resistance, prostate cancer, and breast cancer 49,[53][54][55][56] . Mouse homozygous L3mbtl3 deletion is late embryonic lethal around E17.5 41,57 , although the molecular targets for the lethality remain unclear (Fig. 8a). We found that the protein levels of mSWI/SNF subunits, including SMARCC1, BRG1, SMARCC2, PBRM1, ARID1A, and SMARCB1 are markedly elevated in the head extracts of mouse L3mbtl3 (−/−) deletion mutant embryos at E14.5, as compared to that of the wildtype littermates (Fig. 8b). Further characterization showed that the protein levels of SMARCC1 were high and comparable in both wildtype and L3mbtl3 deletion mutant embryos at E9.5. In the wild-type embryos, SMARCC1 protein levels became gradually declined between E14.5-E17.5. However, the protein levels of SMARCC1 accumulated in the L3mbtl3 deletion embryos during the E14.5-E17.5 developmental period and these L3mbtl3 deficient embryos eventually died 41 (Fig. 8a, c).
These studies indicate that SMARCC1 protein is targeted for proteolysis by L3MBTL3 between E14.5-E17.5 during mouse embryonic development. In MEFs isolated from L3mbtl3 homozygous deletion mutant embryos (Fig. 8d), mSWI/SNF proteins also accumulated and L3mbtl3 deletion markedly increased the methylated K615 level in SMARCC1, as compared to the MEFs from the wildtype littermates (Fig. 8e). Immunostaining of brain tissues revealed that the protein levels of total and K615-methylated SMARCC1 and SMARCC2 proteins are significantly elevated in L3mbtl3 deleted embryos, as compared to that of wildtype littermates (Fig. 8f). These studies indicate that L3MBTL3 regulates the proteolysis of SMARCC1 and SMARCC2 during embryonic brain development and that the accumulation of mSWI/ SNF proteins may contribute to the embryonic lethality of mouse L3mbt3 deletion mutants. Our studies are consistent with a model by which SMARCC1 and SMARCC2 proteins are methylated by SET7 to recruit L3MBTL3 and its associated CRL4 DCAF5 ubiquitin ligase complex to target SMARCC1 and SMARCC2 proteins for degradation, thereby regulating the disassembly of the mSWI/SNF complex through the subsequent proteolysis of other mSWI/SNF subunits, and that LSD1 serves as a demethylase to remove the methyl groups from the  methylated SMARCC1 and SMARCC2 proteins to prevent their degradation to preserve the integrity of the mSWI/SNF complexes (Fig. 8g).

Discussion
In this report, we found that the assembly or disassembly of the mSWI/ SNF chromatin remodeling complexes is dynamically regulated by a lysine methylation-dependent proteolytic mechanism involving the activities of the LSD1 demethylase and the L3MBTL3-CRL4 DCAF5 ubiquitin ligase complexes. Our studies revealed that two key mSWI/SNF components, SMARCC1 and SMARCC2, contain critical lysine residues for SET7-mediated methylation and these methylated lysine residues serve as the substrates of LSD1 and L3MBTL3 that regulate the proteolysis of SMARCC1 and SMARCC2, promoting the disassembly of the mSWI/SNF complexes. Our studies are consistent with previous findings that SMARCC1 and SMARCC2 are the core mSWI/SNF components that are present in every mSWI/SNF complex for the nucleation of the mSWI/SNF complex assembly and that loss of SMARCC1 and SMARCC2 leads to the disassembly of entire mSWI/SNF complexes and the degradation of many mSWI/SNF subunits 15,[43][44][45] . Notably, the loss of LSD1 blocked the self-renewal of mESCs and F9 cells that express the  The genotypes were determined from genomic DNA isolated from the wildtype and the L3MBTL3 null embryos. Repeated three times with the same results. b Total lysates from the heads of mouse L3mbt3 (+/+) and deletion (−/−) mutant embryos (equal total proteins) were analyzed by Western blotting with antibodies for the indicated mSWI/SNF proteins. c Lysates from the heads of wildtype and L3mbtl3 deletion embryos were collected at various embryonic dates (E9.5, E14.5, and E17.5) and analyzed for SMARCC1 and actin protein levels. d Mouse embryonic fibroblasts from the wildtype and L3mbtl3 deletion mutant embryos (E13.5) were examined and representation of three independent experiments is shown for indicated SWI/SNF proteins by Western blotting. e The L3mbt3 (+/+) and deletion (−/−) mutant MEFs were analyzed for monomethylated K615 of SMARCC1, total SMARCC1, and L3MBTL3 proteins by Western blotting. f Embryonic brain cryosections of wildtype and L3mbtl3 deletion mutant (E14.5, 10-µm thick coronal) were immunostained with anti-SMARCC1, monomethylated K615 of SMARCC1, and SMARCC2 antibodies. Boxed regions are enlarged and arrows or arrowheads indicate the increased SMARCC1-K615me or SMARCC2 regions in the brains. Scale bar: 100 μm. Repeated three times with similar results. g Model: Specific lysine residues of SMARCC1 and SMARCC2, core components of the SWI/SNF complex, are methylated by SET7 methyltransferase and the levels of methyllysines are reversibly removed by LSD1 demethylase. L3MBTL3 preferentially binds to the methyl-lysine residues of SMARCC1 or SMARCC2 to recruit the CRL4 DCAF5 ubiquitin E3 ligase complex to target the methylated proteins for ubiquitin-dependent proteolysis, promoting the disassembly and proteolytic degradation of the mSWI/SNF complex. Source data are provided as a Source Data file.
ESC-specific esBAF complex that contains only SMARCC1 4,5 , and our work identifies SMARCC1 as the critical target of LSD1 in mESCs and F9 cells (Figs. 5, 6). However, loss of LSD1 in other cells, such as 293 T or H1299 cells that contain both SMARCC1 and SMARCC2, has no detectable growth defects ( Supplementary Fig. 7c-e), although the loss of LSD1 triggers the degradation of both SMARCC1 and SMARCC2 proteins in these cells. It is possible that the fractions of the methylated SMARCC1 and/or SMARCC2 proteins are not high enough to cover every mSWI/SNF complex in 293 T and H1299 cells for disassembly and degradation. Our studies also show that ectopic expression of SMARCC2 in F9 cells is not sufficient to confer the resistance to LSD1 inactivation (Fig. 6e), suggesting that additional mechanisms exist for 293 T and H1299 cells to resist LSD1 silencing. SMARCC1 or SMARCC2 are also proteolyzed when mSWI/SNF subunits are lost or mutated 14,44,45,51 . Our evidence suggests that the proteolysis of SMARCC1 and SMARCC2 during the loss of other mSWI/SNF subunits is L3MBTL3 dependent (Fig. 7). Although we found that SMARCC1 and SMARCC2 are regulated through lysine methylation, it remains to be further investigated whether additional mSWI/SNF proteins are similarly regulated by lysine methylation-dependent proteolysis through LSD1 and the L3MBTL3-CRL4 DCAF5 ubiquitin ligase complexes.

Cells and cell synchronization
Human lung carcinoma H1299, cervical carcinoma HeLa, embryonic kidney carcinoma 293 T, and mouse teratoma F9 cells were purchased from the American Type Culture Collection (ATCC) and cultured in RPMI-1640 or DMEM medium with 10% FBS and 1% antibiotics as described 25,26,41 . Mouse embryonic fibroblasts (MEFs) were generated from wildtype and L3mbtl3 deletion mutant embryos or CAGGCre-ER TM /LSD1 fl/fl embryos (E12.5-E13.5), as described previously 25,26,58 . Mouse embryonic stem cells (CMTI-2, strain C57/ BL6J, passage 16) were obtained from MilliporeSigma and were cultured on the mitomycin C treated mouse fibroblast feeder layer in knockout DMEM and knockout serum replacement, supplemented with leukemia inhibitory factor (LIF), GlutaMax, β-mercaptoethanol, MEM nonessential amino acid solution, and antibiotics 25,26 (all from Life Technologies). For stable expression, human LSD1, SMARCC1, and SMARCC2 were cloned into the retroviral pMSCV-Puromycin or Zeocin vector containing 3xFlag-3xHA epitope (clontech) and the recombinant retroviruses were packaged in 293 T cells 25,26 . Viral-infected H1299, F9, and mouse embryonic stem cells were selected by puromycin or zeocin resistance. For cell cycle synchronization, actively growing HeLa cells were treated with 2.5 mM thymidine for 18 h, released into the fresh cell culture medium for 9 h, and then treated again with 5 μg/ml aphidicolin (Sigma) for another 15 h to synchronously arrest the cells at the G1/S border 41 . For cell cycle synchronization in mitosis, actively growing HeLa cells were treated with 40 ng/ml nocodazole for 14 h. The rounded-up mitotic cells were collected by low-speed centrifugation, washed, and released into fresh complete growth medium without nocodazole, and collected at various time points 50 . A fraction of cells at various time points after releasing from mitosis was labeled with 10 μM 5-ethynyl-2′-deoxyuridine (Edu) for 30 min, trypsinized, and processed according to the protocol of Clik-iTTM Plus Edu Flow Cytometry Assay Kit and stained by FxCycleTM PI/RNase Staining Solution at room temperature for 30 min 59 . Cellular DNA was stained with 4′,6-diamidino-2-phenylindole (DAPI).

Peptide synthesis and preparation of methylated peptides
The monomethylated K482 (RALPEFFNGKNKS (Km1) TPEIYLAYRNF MIDTC) and K615 (GLRTDIYSKKTLAKS (Kme1) GASAGREWTEQETC) and cognate unmethylated peptides of SMARCC1 were synthesized at ABI Scientific. The monomethylated K482 and K615 peptides were used to raise rabbit polyclonal antibodies after covalently coupling these peptides to keyhole limpet hemocyanin (KLH) 41 . Affinity purification of methylated peptide antibodies were conducted as described in ref. 41, and the unmethylated and monomethylated K482 or K615 peptides were immobilized to Sulfolink-coupled-resins (Thermo Fisher) by covalently cross-linking with the cysteine residues at the end of the peptides to the resin 41 . The anti-monomethylated K482 or K615 peptide sera (5 ml each) were diluted in 1:1 in PBS and first passed through the unmethylated K482 or unmethylated K615 peptide columns (1 ml) three times to deplete anti-K482 or K615 peptide antibodies. The unbound flow-through antibody fractions were then loaded onto the monomethylated K482 or monomethylated K615 peptide column (0.5 ml), washed, and the bound antibody fractions were eluted by 5 ml of 100 mM glycine, pH2.5. The eluted antibodies (0.5 ml/fraction) were immediately neutralized by the addition of 100 μl of 2 M Tris, pH8.5, and tested for specificity towards the monomethylated K482 or K615 peptide but not to the unmethylated K482 or K615 peptide. For L3MBTL3 binding to methylated peptides, the monomethylated K482 and K615 peptides and cognate unmethylated peptides of SMARCC1 were covalently coupled to the Sulfolink-coupled resins, respectively, through the disulfide bond between the C-terminal cysteine of the peptides and the resins 25,26,41 . Human L3MBTL3 was cloned into pcDNA3 and in vitro translated and isotope-labeled using T7 polymerase TNT® Quick Coupled Transcription/Translation System (Promega) in the presence of 35 S-L-methionine (PerkinElmer).

RNA extraction and qRT-PCR analysis
RNA was extracted from the heads of mouse embryos or neonatal mice using Trizol regent according to the manufacturer's instructions. 0.5 μg of total RNA was reverse-transcribed using a first-strand cDNA synthesis kit (Roche). qRT-PCR assays were performed with SYBR Green Master mix (Bio-Rad) and specific primers for PCR amplification. qRT-PCR data were recorded and analyzed using iQ-PCR (Bio-Rad) equipment and software according to manufacturers' recommendations. For each primer pair, the primer efficiency was measured and the melting curve was analyzed.

Transfection and siRNAs
Oligofectamine was used for siRNA silencing in HeLa, H1299, or 293 T cells, and DharmaFECT1 was used for the siRNA silencing in mouse ESCs and F9 cells, whereas Lipofectamine 2000 was used for transient transfection as described previously 25,26,41 . Typically, 50 nM of each siRNA or their combinations were transfected into target cells for 48 h and cells were directly lysed in SDS or NP40 lysis buffers 41 . For verification of the effects, usually, two or three independent siRNAs were designed to examine the knockdown efficiency and the consequences of knockdown on target proteins 41

Demethylation analysis
The glutathione-S-transferase (GST) and the GST-LSD1 (human) fusion protein were expressed in E. coli BL21 strain and purified by the Glutathione Sepharose resin. Purified 1 μg of control GST or GST-LSD1 proteins were incubated with 100 ng of the monomethylated K482 or monomethylated K615 peptides for 4 h at room temperature and the resulting peptides were blotted onto nitrocellulose membrane 25,41 . The demethylated peptides were detected by immunoblotting with affinity-purified anti-monomethylated K482 or anti-monomethylated K615 antibodies.

Animals and histology
The LSD1fl/ + conditional mutant (B6.129-Kdm1a tm1.1Sho/J, stock No: 023969), transgenic actin-Cre-ER (CAGGCre-ER TM , B6.Cg-Tg(CAG cre/ Esr1*)5Amc/J, stock No: 004682), and transgenic Nestin-Cre (B6.Cg-Tg(Nes-cre)1Kln/J, stock No: 003771) mouse strains were obtained from Jackson Laboratory. The L3mbt3 deletion mutant (MBT-1-/+, B6;129-L3mbtl3tm1Tmiy) mouse strain was previously described 41 . All animal experiments, including breeding, housing, genotyping, and sample collection were conducted in accordance with the animal protocols approved by the Institutional Animal Use and Care Committee (IACUC) and complied with all relevant ethical regulations at the University of Nevada, Las Vegas. All procedures were conducted according to the National Institutes of Health (NIH) Guide for Care and Use of Laboratory Animals. The UNLV IACUC is an AAALAC-approved facility and meets the NIH Guide for the Care and Use of Animals. For embryonic analyses, usually, three pairs of the L3MBTL3 (−/+) male and female mice (10-12 weeks old) in three cages, each with one male and one female, were bred in the late afternoon and the breeding plugs were examined in the female mice in next morning. The positive plugs were counted as embryonic day 1 (E1) and the pregnant female mice between E14-E17.5 were euthanized by the primary method of CO 2 asphyxiation, followed by cervical dislocation (secondary method), as approved by the institutional IACUC committee. Usually, a single pregnant female mouse produced about 6-8 embryos, which segregated at the Mendelian ratio, usually with 1-2 L3MBTL3 (−/−), 1-2 wildtype, and 3-4 heterozygous L3MBTL3 (−/+) embryos. The L3MBTL3 null embryos between E17.5-19.5 usually died and became disintegrated, so they were excluded from protein analyses. For the analysis of mSWI/SNF proteins in LSD1 fl/fl /Nestin-Cre mice, usually, three to four pairs of the LSD1 flox/flox male and LSD1 flox/+ /Nestin-Cre mice female mice (10-12 weeks old) were bred. The animals were collected immediately after birth to avoid any delay in sample analysis. The brains or other body parts of the mice were dissected for protein or immunostaining analysis. For immunostaining, embryos or dissected brains were fixed in 4% paraformaldehyde (PFA) at 4°C overnight and embedded in an Optimal cutting temperature compound (OCT) according to standard procedures 27 . Sections (10-µm thick, coronal) were stained with specific antibodies and counter-stained with 4′,6-diamidino-2-phenylindole (DAPI). Images were acquired with the Nikon A1Rsi Confocal LSM. The sample size was chosen on the basis of our experience on L3mbtl3 or Lsd1 mutant mice and on cultured cells in order to detect the mSWI/SNF proteins for differences of at least 50% between the wildtype and mutant groups 41 . In the experimental analyses for the examination of proteins, the investigators were unaware of the genotypes of the experimental embryos. The investigators also randomly analyzed the wildtype, heterozygous and homozygous knockdown embryos.

Analysis of proteins and DNA from embryos
The experimental procedures for embryo isolation were approved by the UNLV Institutional Animal Use and Care Committee (IACUC). The L3mbtl3 embryos from the euthanized pregnant female mice or dissected brains or other body parts from the Lsd1 mice were washed with PBS and lysed in the NP40 lysis buffer 41 . The nuclear and cytosolic fractions were separated by centrifugation. Genomic DNA was isolated from nuclear pellets by Zymo genomic DNA-tissue prep kit and quantified. Proteins in the cytosolic supernatant of the lysates were quantified by protein assay dye (Bio-Rad), equalized, and boiled for 15 min after the addition of 1% SDS and 5% beta-mercaptoethanol to the lysates. Proteins were resolved in protein gel and analyzed by Western blotting.

Statistical information
Experiments were usually performed with at least two to three independent repeats to ensure the results. Statistical plot analyses were performed using Microsoft Excel. Protein bands were quantified using ImageJ. To quantify protein loading in each Western blot analysis of a set of protein samples, the same protein samples were analyzed with three repeated loading experiments (technical replicates). For cell-based assays, triplicated repeats in the same set of cells (technical replicates) were measured and the experiments were usually repeated in three independent experiments with independently cultured cells (biological replicates). Quantitative data are expressed by a bar graph, with mean and standard deviation (SD) for error bars from independent replicates. For siRNA-mediated knockdown experiments, statistically significant differences or variations between means of double and single knockdowns were normalized to the luciferase siRNA control or actin control and compared using a two-tailed paired Student's t-test. For animal experiments, triplicated breeding was used to obtain a statistically significant number of embryos; and statistically significant differences between means of protein levels in the control wildtype and knockout mutants were compared using the two-tailed equal-variance independent Student's t-test. All other data were determined using a two-tailed equal-variance independent Student's t-test. The data in all figures met the assumption of normal distribution for tests. Different data sets were considered to be statistically significant when the P value was <0.05 (*) or 0.01 (**) 41,60 .

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support this study are available from the corresponding author upon reasonable request. Source data are provided with this paper.