Essential Cytoplasmic Role of the Histone Lysine Methyltransferase Setdb1 in Post-Transcriptional Regulation in Embryonic Stem Cells

Embryonic stem cells (ESCs) fate is regulated both at transcriptional and post-transcriptional levels. Indeed, several studies showed that, in addition to gene transcription, mRNA stability and protein synthesis are finely tuned and strongly control the ESCs pluripotency and fate changes. An increasing number of RNA-binding proteins (RBPs) involved in post-transcriptional and translational regulation of gene expression has been identified as regulators of ESC identity. The major lysine methyltransferase Setdb1 is essential for the self-renewal and viability of ESCs. Setdb1 was primarily known to methylate the lysine 9 of histone 3 (H3K9) in the nucleus, where it regulates chromatin functions. However, Setdb1 is also massively localized in the cytoplasm, including in mouse ESCs, where its role remains unknown. Here we show that the cytoplasmic Setdb1 (cSetdb1) is essential for the survival of mESCs. Functional assays further demonstrate that cSetdb1 regulates gene expression post-transcriptionally, affecting the abundance of mRNAs and the rate of newly synthetized proteins. A yeast-two-hybrid assay shows that cSetdb1 interacts with several regulators of mRNA stability and protein translation machinery, such as the ESCs-specific E3 ubiquitin ligase and mRNA silencer Trim71/Lin41. Finally, proteomic analyses reveal that cSetdb1 is required for the integrity of Trim71 complexes involved in mRNA metabolism and translation. Altogether, our data uncover the essential cytoplasmic function of a firstly supposed nuclear “histone” lysine methyltransferase, Setdb1, and provide new insights into the cytoplasmic/post-transcriptional regulation of gene expression mediated by a key epigenetic regulator.


Embryonic stem cells (ESCs) fate is influenced by transcriptional and post-transcriptional regulation
of gene expression. The role of transcription factor regulatory networks in combination with chromatin modifying proteins in ESCs fate have been extensively studied. Increasing evidences show that also mRNA stability and mRNA translation are finely tuned and strongly control the pluripotency and the fate of ESCs (Gabut et al., 2020). Moreover, an increasing number of RNA-binding proteins (RBPs) involved in post-transcriptional and translational regulation of gene expression has been identified and described as regulatory of ESC identity (Kwon et al., 2013;Ye and Blelloch, 2014;You et al., 2015). Setdb1 (also called ESET/KMT1E) is a histone 3 lysine 9 (H3K9) lysine methyltransferase (KMT) which belongs to the SUV39 family, and a major epigenetic regulator of transcriptional gene repression.
Setdb1 knockout (KO) is lethal in mice at the peri-implantation stage (Dodge et al., 2004). Setdb1 is essential for the survival of mESCs and it influences ESCs identity, pluripotency and self-renewal (Bilodeau et al., 2009;Lohmann et al., 2010;Yeap et al., 2009). Setdb1 is also required for the cell survival, pluripotency maintenance and terminal differentiation of many progenitor cell types.
Setdb1 is necessary for the survival of spermatogonial progenitors in mice ) and a mesenchyme-specific Setdb1 knock-out resulted in an increase of articular chondrocytes terminal differentiation . Furthermore, Setdb1 is crucial for early neurogenesis in mice by promoting proliferation and cell survival (Tan et al., 2012). In contrast, Setdb1 is crucial for osteoblast differentiation during bone development  and is involved in the terminal differentiation of skeletal muscle adult stem cells (Beyer et al., 2016) and growth plate chondrocytes . Setdb1 shows unusual distinct subcellular and subnuclear distributions, including in ESCs (Beyer et al., 2016;Loyola et al., 2006;Schultz et al., 2002;Yeap et al., 2009), but the meaning of the different Setdb1 subcellular localisations remains mostly unknown. Although, the role of nuclear Setdb1 in transcription regulation, especially in mESCs, has been extensively studied, its cytoplasmic role remains elusive.
Since Setdb1 is essential for ESCs, we first asked whether cytoplasmic Setdb1 (cSetdb1) is required for the survival of ESCs. Our cellular studies showed that the loss of normal cytoplasmic localization of Setdb1 in ESCs lowers the cell amount and increases the number of apoptotic cells. The loss of cSetdb1 affects specific mRNA abundance and the overall rate of newly synthetized proteins.
Biochemical protein-protein interaction studies unraveled that cSetdb1 interacts with several regulators of mRNA processing and translation, among which the ESC-specific RNA binding protein (RBP) Trim71/Lin41. Proteomic studies further demonstrated that cSetdb1 is required for the integrity of Trim71 protein complexes. Taken together, our results suggest a new fundamental role of cSetdb1 in mESCs, which influences post-transcriptional regulation of gene expression and provide new insights into the cytoplasmic regulation of gene expression mediated by a key epigenetic regulator.

Results
Cytoplasmic Setdb1 is an active lysine methyltransferase required for the survival of embryonic stem cells Setdb1 is essential for mouse embryonic stem cells survival (mESCs, see introduction). While Setdb1 function, extensively studied in mESCs, has been always associated with its nuclear role, it has been shown that Setdb1 localizes also in the cytoplasm. However, the importance and the function of cytoplasmic Setdb1 (cSetdb1) in mESCs has never been investigated.
To address the cytoplasmic role of Setdb1 in mESCs, we used the TT2 mESCs which have been genetically modified to induce an acute Setdb1 knockout (KO) by tamoxifen (TA). Indeed, an acute Setdb1 protein depletion was obtained from 48h post-TA treatment (Fig 1A). We also used inducible TT2 Setdb1 KO cells where exogenous wild-type Setdb1 (KO+WT) or enzymatically-dead point mutant (KO+CA) were stably expressed in the nucleus and in the cytoplasm (Fig 1B). Using these cell systems, we first investigated whether cSetdb1 is enzymatically active. To this end, we performed an immunoprecipitation (IP) of Setdb1 from the cytoplasmic fractions of Ctr, KO, KO+WT or KO+CA mESC lines, using anti-Setdb1 antibody (Ab, Fig 1C). We used the IP products in an in vitro radioactive methylation assay using core histones as a substrate and radioactive S-Adenosyl Methionine (SAM) as methyl donor. We observed an enzymatic activity towards H3 in the Ctr and KO+WT conditions, and, as expected, not in KO and KO+CA (Fig 1C). Since the immunoprecipitates from KO and CA Setdb1 do not show any detectable methyltransferase activity (Fig 1C), we deduced that previous detected signals were specific for Setdb1. Thus, mESCs cSetdb1 is enzymatically active.
To investigate the role of cSetdb1 in mESCs, we generated an inducible Setdb1 KO cells where Setdb1 expression is stably rescued only in the nucleus (KO+NLS) by adding an artificial strong nuclear localization signal (NLS) to the N-terminal part of the protein (Fig 1B and 1D). Setdb1 depletion is known to cause growth defects in ESCs (Fisher et al., 2017), but the importance of Setdb1 cytoplasmic function has never been specifically examined. To this end, we first performed an apoptosis test using Annexin V-FITC/IP staining in KO and KO+NLS cells, compared to Ctr and KO+WT cells, after a long-term acute Setdb1 KO (96h of TA treatment). As expected, we found almost twice more apoptotic cells in early stage (Annexin V+) and late stage (Annexin V and Propidium Iodide-PI+) of apoptosis upon acute Setdb1 depletion (KO) compared to control (Ctr) (Fig 1E). Rescue of WT Setdb1 (KO+WT) restored the normal apoptotic index, while the expression of Setdb1 only in the nucleus (KO+NLS), did not (Fig 1E), showing an increase in the number of apoptotic cells. To go further, we quantified the number of attached cells, which are living (trypan blue-positive). In accordance with the previous result, we observed a significant decrease in the number of KO and KO+NLS attached cells, compared to control KO+WT cells (Fig 1F). In addition, KO+NLS cells show less viability than KO+WT cells (Fig 1F and S1A). To further check if cSetdb1 is essential in ESCs, we looked after 6-days of TA treatment. Our data showed that the expression of nuclear-only Setdb1 exacerbated even more the number of apoptotic cells and lowers the number of attached cells (KO+NLS, Fig 1G). Moreover, by immunofluorescence, an increase in cleaved caspase-3 was detected in KO cells (Fig S1B).
Together, these data highlight the importance of cytoplasmic Setdb1 for ESCs survival.

Cytoplasmic Setdb1 regulates the abundance of mRNAs and newly synthetized proteins
Setdb1 in the nucleus regulates chromatin functions affecting transcription regulation in ESCs (Bilodeau et al., 2009;Yeap et al., 2009). We wondered whether cSetdb1 could influence gene expression at post-transcriptional level considering its importance in the regulation of protein abundance and ESC identity (Gabut et al., 2020;Schwanhausser et al., 2011). First, we looked at the amount of mRNAs, via RNAseq assay after ribodepletion, in KO+NLS compared to KO+WT mESCs cells. First, the PCA analysis showed that the three KO+NLS replicates were well defined and appear to form a distinct group from the KO+WT replicates (Fig S2A). 291 mRNAs were found differentially abundant in KO+NLS compared to KO+WT cells (Fig 2A and Table 1); of those genes, 76 were less abundant and 215 more abundant. While the role of Setdb1 in regulating gene expression has been always associated with its nuclear roles, our RNA-seq data suggest that cSetdb1 is also involved in gene expression regulation, at the mRNA transcript abundance level.
Gene ontology (GO) analysis for biological processes showed that most of differentially abundant mRNAs are enriched in gene categories related to Setdb1 phenotypes in mice (http://www.informatics.jax.org/marker/MGI:1934229).
Notably, embryogenesis, including development (vascular, reproductive, placenta and somite development), tissue morphogenesis (including also branchial, epithelial and placenta morphogenesis), segmentation and regulation of cellular response to growth factor stimuli (Fig. 2B). In particular, we identified more abundant mRNAs related to reproductive and placenta development (such as Gcm1 and Tex19.1) and to vascular, vessel development and angiogenesis (such as Gpx1,Egr3,Adm and Enpp2). Moreover, in line with the phenotype that we observed in absence of cSetdb1, we found, less abundant mRNAs associated with cell proliferation and cell survival, including Dab2, kit, Fgf1 and Pdgfra (Fig. 2B).
Further, enrichment analysis for biological processes and biological pathways, showed that the differentially abundant mRNAs are enriched in gene categories related to extracellular matrix (EM), including EM organization, collagen formation and multimerization and cell adhesion (Fig. 2C). In particular we identified less abundant mRNAs related to EM structure/organization (such as collagen genes Col23a1, Col4a1 and Col4a2) and more abundant mRNAs related to EM degradation (such as the matrix metalloproteinases Mmp12, Mmp9 and Mmp25). All these categories and pathways are notably involved in ESCs self-renewal, proliferation and differentiation, thus associated with ESCs identity.
Interestingly, in accordance with the enrichment analysis, KO+NLS ESCs showed more spread distribution and less compact clones compared to KO+WT control ESCs, a phenotype characteristic of more committed cells (Fig S2B).
All these data show that in mESCs cSetdb1 affects the abundance of mRNAs mainly involved in development and cell adhesion, and suggest a loss of ESC state and a potential acquisition of a differentiated phenotype upon the depletion of Setdb1 cytoplasmic function.
Although Setdb1 was shown to be associated with ribosomes (Rivera et al., 2015), the possible role of Setdb1 in translation regulation has not been investigated. Thus, to pursue the study of the role of cSetdb1 in post-transcriptional gene expression control, we checked its potential participation in protein translation regulation. To this end, we performed a SUnSET assay, a non-radioactive method to quantify the amount of newly synthetized proteins (Schmidt et al., 2009), which mirrors ongoing protein synthesis through puromycin incorporation. We acutely depleted Setdb1 in ESCs after 48h of tamoxifen treatment, a timepoint which does not alter the viability (Fig S2C), neither proliferation (Fig S2D), size and granularity of the cells (Fig S2E), rRNAs level (Fig S2F) or RNA synthesis (Fig S2G).
The SUnSET assay specifically showed that an early acute Setdb1 depletion (KO) affects ongoing protein synthesis, revealing a statistically significant almost 40% decrease in the amount of newly synthetized proteins, compared to control cells (Ctr) (Fig 2D). This phenotype is rescued by WT Setdb1 in KO+WT ESCs. However, very interestingly, the expression of nuclear-specific Setdb1 in KO+NLS ESCs did not rescue the protein synthesis defect, showing a statistically significant 42% decrease in the amount of newly synthetized proteins in KO+NLS compared to KO+WT ESCs (Fig. 2D).
Together, the SUnSET assay and the rescue experiments suggest that the loss of Setdb1 in the cytoplasm affects the amount of newly synthetized proteins, therefore that cSetdb1 is required for protein synthesis in ESCs. Altogether, our results of RNA-seq, enrichment analysis and protein synthesis show that cSetdb1 plays key roles in post-transcriptional regulation of gene expression in ESCs, influencing mRNA abundance and the synthesis of proteins.

Cytoplasmic Setdb1 interacts with Trim71, a mESC-specific regulator of gene expression
We next sought to decipher the molecular mechanism(s) through which cSetdb1 posttranscriptionally affects gene expression. To this end, we conducted a yeast two-hybrid assay (Y2H, performed by Hybrigenics Services, France) to identify potential cSetdb1 protein partners. We used as bait the full-length WT mouse Setdb1 cDNA, screened against a mESC cDNA library. Interestingly, the Y2H assay showed that Sedb1 interacts with several regulators of mRNA stability and translation (not shown). Among other proteins, our data assigned a robust and high-fidelity score to the interaction between Setdb1 and the RNA-binding protein and mRNA silencer Trim71 (also called Lin41), specifically expressed in ESCs. We confirmed this interaction by 1-by-1 Y2H assay (Fig. 3A).
The mESC-specific RNA-binding protein Trim71 has a double function: it is a E3 ubiquitin ligase and it regulates gene expression post-transcriptionally. Notably, Trim71 localizes to Processing bodies (Pbodies), where mRNA repression occurs, and it is involved in the regulation of mRNA stability and translation (Chang et al., 2012;Loedige et al., 2013;Torres-Fernandez et al., 2019). The Y2H assay showed that the Setdb1/Trim71 interaction is overlapping the coiled-coil domain of Trim71 (Fig. 3A), which is required for Trim71 to repress mRNA function (Loedige et al., 2013). The interaction between cSetdb1 and Trim71 was confirmed by co-immunoprecipitation in mESCs of ectopically expressed proteins (Fig. 3B, left), and of exogenous Trim71 and endogenous Setdb1 (Fig. 3B, right).
The coiled-coil domain of Trim71 is required for its colocalization with DCP1a and directs Trim71 to Pbodies, where mRNA repression occurs (Torres-Fernandez et al., 2019). We found that this domain mediates also the interaction with Setdb1 (Fig. 3). Furthermore, using PLA, we show a spatial close proximity between Trim71 and DCP1a, a mRNA decapping enzyme involved in mRNA stability known to co-localize, indicating a direct interaction through physical contact or close functional coupling between Trim71 and DCP1a ( Fig. 3D and S3A, Video S2). These results suggest a functional link between Setdb1, Trim71 and DCP1a to regulate mRNA behavior in ES cells.
We then evaluated the functional significance of the cSetdb-Trim71 interaction by checking if Trim71-regulated mRNAs are also regulated by cSetdb1. Comparing our RNA-seq results (described in Fig. 2A) with the transcripts differentially regulated after knock-out of Trim71 in mESCs (Welte et al., 2019), we found 93 mRNAs commonly regulated by cSetdb1 (Setdb1 NLS) and Trim71 (Trim71 KO) ( Fig. 3E), 18 of them linked to multicellular organismal reproductive process (top biological process category from our ontology analysis (in bold in: Table1).
Next, we wondered whether cSetdb1 could influence the abundance of selected Trim71 mRNA targets, mainly involved in cell proliferation and embryogenesis, including Inhbb and Ccnd2 and potential Trim71 targets, including H60b and Ccnd1. By RNA-seq and RT-qPCR we found that Inhbb and H60b are dysregulated in absence of Setdb1 (KO) or upon the expression of the nuclear-specific Setdb1 (KO+NLS), compared to the controls (Ctr and KO+WT) ( Table 1 and Fig. 3F). By RT-qPCR we observed the same for Ccnd1 and Ccnd2 (Fig. 3F). Our data show that cSetdb1 and Trim71 coregulate the abundance of common mRNAs.
Finally, as for Setdb1 (Fig. 2D), we wondered whether Trim71 could regulate the amount of newly synthetized proteins. Interestingly, the amount of newly synthetized proteins decreases in absence of Trim71 (Trim71 KD), compared to control (Ctr) (Fig. 3G) Our results show the involvement of Trim71 in ongoing protein synthesis in ESCs as for cSetdb1.
All together, these new data confirm the previously demonstrated role of Trim71 in the regulation of mRNAs stability and translation, and highlights a functional role of cSetdb1, associated to Trim71, in post-transcriptional regulation of gene expression.

Integrity of Trim71 complex(es), involved in mRNA stability and translation, is affected in the absence of Setdb1
Our data showed that Setdb1 is a cytoplasmic partner of the ESC-specific RNA-binding protein and mRNA silencer Trim71/Lin41. To better understand the functional relation between Setdb1 and Trim71, we wondered whether Setdb1 is essential for the integrity of Trim71 complex, therefore for its functions. To address this question, we performed a proteomic analysis on Trim71 complex(es) in absence of Setdb1. We immunoprecipitated the cytoplasmic GFP-tagged Trim71 (Trim71) and its specific partners in ESCs, where we acutely depleted Setdb1 after 48h of tamoxifen treatment (KO), a timepoint which does not alter the viability, neither proliferation, size and granularity of the cells, rRNAs level or RNA synthesis (shown in Fig. S2C-G), and in control ESCs (ES) and we compared them to their corresponding GFP controls (Ctr) ( Fig. 4A-B). Then, the purified Trim71 complexes were analyzed by mass spectrometry. After discarding the sample with low sum of intensity (Fig. S4A), we filtered for proteins that were well represented in the immunoprecipitates; we then applied a scale to interval normalization to better compare the samples (see Material and Methods). As expected, the PCA analysis ( Fig. 4C) separates well all Trim71 samples from all Ctr samples. We then performed an ANOVA multi-sample test, which conducted to the selection of 108 proteins that were found differentially present in at least 2 of the 4 conditions ( Fig. 4D). In order to select the partners of Trim71 that could differentially bind to Trim71 after an acute Setdb1 knockout (KO-Trim71 against ES-Trim71), we performed a Post-Hoc Tukey's HSD test. We found, as displayed on the heatmap ( Fig.   4E), that some Trim71 partners are more or less associated with Trim71 in absence of Setdb1 (KO-Trim71) compared to control cells (ES-Trim71). Notably, the RNA-binding protein hnRNP-C and Hmgcs1 are more enriched in KO-Trim71 compared to ES-Trim71, and, conversely, Eif5a2, Elp3, Gnl3, Ranbp10 and Smyd5 are less enriched in KO-Trim71 compared to ES-Trim71. Very interestingly, these proteins are mainly involved and associated with mRNA stability, ribosome biogenesis and translation regulation. These data illustrate that the profile of Trim71 partners, whereby Trim71 executes its cellular functions, is dependent on Setdb1 and provide important mechanistic hypothesis to explain the role of cSetdb1 on post-transcriptional regulation.
Altogether, these observations shed new light on Trim71 complex function in mRNA stability and translation, revealing the importance of Setdb1 for its architecture and confirming the intimate relation between Setdb1 and Trim71.

An RBP partner of Trim-71, hnRNP-C, regulates the abundance level of key mRNAs involved in the control of cell fate
We have shown for the first time that the RNA-binding protein hnRNP-C, known to be involved in mRNA turnover and translation, is a member of Trim71 complex. Furthermore, we found the RBP hnRNP-C associates more with Trim71 in absence of cSetdb1 ( Fig. 4E), highlighting a Setdb1dependent interaction between hnRNP-C and Trim71. Therefore, we wondered whether there is a functional correlation between cSetdb1 and hnRNP-C; we checked if some of the mRNAs regulated by cSetdb1 are targets of hnRNP-C. 21 of the mRNAs found associated to hnRNP-C by eCLIP assay reprogramming. In addition to eCLIP assay, the scanning of all mouse 5' and 3'UTRs for hnRNP-C hits, through an established online prediction tool RBPmap, has identified that the mouse Myc, as the human, is a hnRNP-C target through its 5'UTRs ( Fig. S4B). It has been shown that hnRNP-C binds human c-Myc mRNA on 5'UTR, regulating c-Myc translation level (Kim et al., 2003). Our data suggest that cSetdb1 would influence the c-Myc levels through hnRNP-C. In conclusion, we found interesting mRNAs regulated by cSetdb1 that are also hnRNP-C targets, strengthening our hypothesis concerning the functional correlation between hnRNP-C and cSetdb1 on post-translation regulation of gene expression.
Finally, we have performed a predictive analysis to check for the binding profile of hnRNP-C on the mRNAs regulated by cSetdb1 (mRNAs described in Fig. 2 and Table 1). Indeed, having a well characterized hnRNP-C RNA binding motif, we performed a stringent computational prediction using RBPmap (see Materials and Methods), to scan the 5' and 3'UTRs of the 291 differentially abundant mRNAs found in absence of cytoplasmic Setdb1 by our RNA-seq for hnRNP-C binding motifs targets.
Out of the 157 3'UTRs that we managed to retrieve from ENSEMBL (Materials and Methods), most (128) were identified as potential targets of hnRNP-C (Fig 4G and Table 2). In the quest to more finely dissect the function and role of hnRNP-C in the absence of cytoplasmic Setdb1 in mESCs, we clustered the gene expression correlations of all the 291 differentially abundant mRNAs. Clustering of the correlation coefficients of gene expression profiles of all 291 DEGs identified 5 distinct clusters with 4 of them demonstrating significant correlations and forming 4 distinct co-expressed modules ( Fig. 4H and Table 3). (Fig. 1 and S1). The diagram in Fig 4H indicates that the 20 differentially abundant mRNAs in cluster 2 (the first that appears on the Fig. 4H diagram) are highly correlated between them and highly anti-correlated with the rest of the inferred clusters and therefore they bare the most informative signal, which might provide possible mechanistic links to the observed cellular phenotypes ( Fig. 1 and S1). Remarkably, all mRNAs in cluster 2 (except one) were predicted as hnRNP-C targets (Fig. 4G) and are less abundant in the absence of cSetdb1 function (Table 1).
Further, in line with the observed mESCs survival phenotype, downstream analysis of GO enrichment identified proliferation-related categories, notably Wnt signaling pathway and MAPK activity, including key genes (such as Dab2 and Dixdc1) (Fig. 4I).
All these data analyses and prediction show that hnRNP-C, one of the Trim71 partners that we found more associated with Trim71 in absence of Setdb1, targets almost half of the mRNAs that we found differentially abundant upon cSetdb1 function depletion, and provide a potential molecular mechanism through which cSetdb1 regulates gene expression post-transcriptionally.
Overall, the results of the binding , the enrichments and the human/mouse UTR conservation suggest that Setdb1 affects the integrity and the function of Trim71 complex, responsible for the regulation of major pathways, essential for mESC survival and identity.

Discussion
The major H3K9 lysine methyltransferase Setdb1 (mouse Eset) is known to be required for the maintenance and the normal growth of mESCs (Fisher et al., 2017;Lohmann et al., 2010). Eset KO embryos lack normal Inner cell mass (ICM), do not form the epiblast, while the extra-embryonic trophectoderm is not affected. Indeed, the inactivation of the mouse gene encoding Setdb1 (Eset), induces de-repression of differentiation genes caused by the loss of H3K9 methylation, in particular the trophectoderm genes, inducing the loss of ESC identity. These nuclear roles of Setdb1 in mESCs are thus well studied. However, Setdb1 is localized both in the nucleus and cytoplasm where it is enzymatically active. Our study addressed the potential cytoplasmic role of Setdb1 in mESCs. To this end, we have set a cellular system in which the cytoplasmic fraction of Setdb1 was lost while the nuclear one was maintained. Overall, we found that the cytoplasmic function of Setdb1 is required for the survival and the normal growth of mESCs. This indicates that cytoplasmic Setdb1 (cSetdb1) regulates ESCs fate independently of its nuclear role..
Next, we asked whether cSetdb1 could regulate gene expression at the post-transcriptional level.
Indeed, not only transcription regulation but also post-transcriptional regulations are both fundamental in ESCs fate and identity control (Gabut et al., 2020;Sampath et al., 2008;Schwanhausser et al., 2011). Since Setdb1 has been shown to be associated with ribosomes (Rivera et al., 2015), we first checked whether cSetdb1 could influence protein synthesis. Our data showed that loss of cytoplasmic Setdb1 in mESCs results in a considerable decrease in global ongoing protein synthesis. This was further supported by our proteomic study showing a physical association of Setdb1 with several translational regulators in mESCs. Our data unravel an unsuspected role of Setdb1 in protein synthesis regulation in mESCs. Moreover, the considerable level of protein synthesis dysregulation, observed in a early cSetdb1 LOF timepoint, strongly evokes that as a potential caused of ES cell death.
Next, we asked whether cSetdb1 could regulate mRNA abundance. While the LOF of cSetdb1 function did not induce a significant change in the total mRNA levels, it induced changes in the abundance of specific mRNAs involved in embryogenesis, including development (vascular, reproductive, placenta and somite development), tissue morphogenesis (including also branchial, epithelial and placenta morphogenesis), segmentation and regulation of cellular response to growth factor stimuli. Thus, in addition to the well-described role of Setdb1 in regulating gene expression in the nucleus via H3K9 methylation, cSedtb1 seems to be crucial in mRNA abundance regulation in the cytoplasm independently of gene transcription.
Our proteomic study showed that Setdb1 strongly interacts with the RNA-binding protein Trim71, a cytoplasmic key post-transcription regulator in ESCs. Trim71, which is both a E3 ubiquitin ligase and a mRNA silencer, co-localizes with DCP1a in P-bodies (Aeschimann et al., 2017;Rybak et al., 2009), and is required for ESCs self-renewal (Chang et al., 2012;Worringer et al., 2014).
Interestingly, cSetdb1 and Trim71 regulate the abundance of common mRNAs and Trim71 LOF induces similar effect on global protein synthesis as Setdb1 LOF does. These data illustrate the complementary roles and the functional relations that the two partners cSetdb1 and Trim71 have in mESCs.
Setdb1 LOF induced significant changes in Trim71 complex composition, including the presence of many proteins that are involved in mRNA stability and translation: hnRNP-C (RNA-binding protein involved in mRNA turnover and translation, among its different functions, Eif5a2 (translation elongation factor, involved also in mRNA turnover in addition to translation), Gnl3 (associated with proteins responsible of ribosome biogenesis) , Smyd5 (associated with the RNA helicase Mov10, a key protein involved in mRNA silencing together with the RISC complex (Castello et al., 2012)). These data illustrate that the profile of Trim71 partners, whereby Trim71 executes its cellular functions, is dependent on Setdb1 and provide important mechanistic hypothesis to explain the role of cSetdb1 in post-transcriptional regulation of gene expression, through Trim71.
The top Trim71 partner, whose interaction depends on cSetdb1, is the heterogeneous nuclear ribonucleoprotein hnRNP-C, which is also localized in the cytoplasm where it regulates the translation efficiency of specific mRNAs (Christian et al., 2008;Kim et al., 2003). We observed that more hnRNP-C protein interacts with Trim71 in cSetdb1 LOF condition. Thus, cSetdb1 is essential for a balanced and harmonious interaction between hnRNP-C and Trim71. Moreover, cSetdb1 and hnRNP-C share common mRNA targets. Interestingly, previous studies showed that the human hnRNP-C binds c-myc mRNA (Kim et al., 2003;Van Nostrand et al., 2016)   penicillin/streptomycin (Sigma), 0.1 mM β-mercaptoethanol (Thermo), 1 mM non-essential amino acids (Sigma), and 1000 U/mL of Leukemia Inhibitory Factor (LIF) (Millipore). All mESC lines were cultured in standard feeder-free conditions with 0.2% gelatin, maintained at 37°C and 8% CO2. To induce deletion of the Setdb1 cKO allele, mESCs were cultured in 800 nM 4-Hydroxy Tamoxifen (Sigma).

Cloning and generation of stable cell lines
Setdb1 cDNA, in addition to 3 NLS sequences, were cloned into pCAG-3XFLAG-IRESbsd vector used by Pr Yoichi Shinkai lab (the same vector that has been used for to introduce 3xFLAG-WT Setdb1 and 3xFLAG-CA Setdb1). Clones were selected in medium containing Blasticidin (7 mg/mL). Full-length Trim71 cDNA was cloned into pCS2 vector given by Liu Yang, which contains a Myc epitope tag.

Cell count
Harvested cells were resuspended in media or PBS in 0.2% Trypan Blue. Live cells were counted using countess automated Cell Counter (Invitrogen) according to the manufacturer instructions.

RNA purification and quantitative reverse transcription-PCR (RT-qPCR)
Total RNAs were extracted using RNeasy mini-kit (Qiagen) following manufacturer's procedures.

DNase (Qiagen) treatment was performed to remove residual DNA. With High Capacity cDNA Reverse
Transcription Kit (Applied Biosystems), 1 μg of total RNAs was reverse-transcribed (RT). Real-time quantitative PCR was performed to analyze relative gene expression levels using SYBR Green Master mix (Promega) following manufacturer indications. Relative expression values were normalized to the housekeeping mRNA Gapdh. H20 and RT-were used as negative controls.

RNA-seq
RNA was isolated as described above (RNA purification section). After RNA extraction, the amount of RNAs was quantified by Nanodrop. The quality of the RNA was determined on the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA), the RNA integrity number was above 8 for all the samples. To construct libraries, 1 mg of high-quality total RNA sample was processed using Truseq® stranded total RNA kit (Illumina®). After the removal of ribosomal RNAs (using Ribo-zero® rRNA), confirmed by QC control on pico chipTM on the Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA, USA), total RNA molecules are fragmented and reverse-transcribed using random primers. Replacement of dTTP by dUTP during the second strand synthesis will permit to achieve the strand specificity. Addition of a single A base to the cDNA is followed by ligation of adapters. Libraries were quantified by qPCR using the KAPA Library Quantification Kit for Illumina Libraries (KapaBiosystems) and library profiles were assessed using the DNA High SensitivityTMHS kit on an Agilent Bioanalyzer 2100. Libraries were sequenced on an Illumina® Nextseq 500 instrument using 75 base-lengths read V2 chemistry in a paired-end mode.

Bioinformatic analyses of RNA-seq data
All sequenced reads were aligned on the GRCm38.p6 version of mouse genome, obtained from ENSEMBL (Cunningham et al., 2019), using the STAR RNA-seq aligner version 2.5.2b (Dobin et al., 2013 After fitting the linear model with DESq2, differentially expressed genes (DEGs) were predicted by applying a correction for multiple testing pvalue of 0.01. From the remaining list of genes two groups for downstream analysis were selected: one with log2 Fold Change (log2FC) more than one (log2FC >= 1 and log2FC <= -1) to investigate manually several key genes and a second more strict selection of log2FC more than 1.5 (log2FC >= 1.5 and log2FC <= -1.5) to use for the automated downstream enrichment and clustering analysis. All enrichment analyses including GO, Pathways and WikiGenes enrichment as well as Gene Set Enrichment Analysis (GSEA), and visualizations were performed with the R package ClusterProfiler (Yu et al., 2012). Clustering of DEG TPMs (Transcripts Per kilobase Millions) was performed by using R base functions for hierarchical and k-means clustering using an array of distance measures and cluster numbers such that robust clusters were obtained. Gene expression correlation matrices for DGEs were calculated by the R package NetCluster and clustering and matrix visualization with the routines of the same package. 3' and 5'UTRs were retrieved for all DEGs by the BioMart service of the ENSEMBL database (Yates et al., 2020) for the mouse C57BL/6NJ strain and all the annotated 3' and 5'UTRs were retrieved. For each gene the longest (or at least >8bps) UTR was selected for the RNA-Binding Protein analysis, which was conducted by using the human/mouse conserved motifs of previously characterized RNA-BPs, as indicated on the online tool RBPmap (Paz et al., 2014). Venn diagram analysis were performed by the R package eulerr.
All analyses were performed in R versions 3.5 (for the DEG identification) and 4.0 for the downstream analysis by using in-house developed scripts available at https://github.com/parisepigenetics/setdb1downstream-anal.

Annexin V assay
Cells

Immunofluorescence (IF) and imaging
Cells were grown on glass coverslips coated with laminin 10 mg/ml in PBS for 1 h at 37 °C. mESCs cells were fixed with 4% formaldehyde (Electron Microscopy Sciences) for 20 min, quench with 50 mM NH4Cl for 10 min and permeabilized with 0.5% Triton-X-100 for 5 min. Primary and secondary antibodies were diluted in PBS containing 2% SVF serum and 0.1% Tween and incubated with coverslips overnight at 4°C or 1h at room temperature, respectively. DNA was stained with 1 μg ml−1 DAPI (Life Technology). Coverslips were mounted with Vectashield mounting media (Vector labs).
Microscopy was performed using inverted microscope Leica DMI-6000. Images were taken with the HQ2 Coolsnap motorized by Metamorph software. All images were processed with ImageJ software.

DuoLink
Cells were prepared and fixed as described above (IF section). Primary and secondary antibodies were incubated according manufacturer procedure from Duolink™ In Situ kit (Sigma DUO92101). The addition of Minus and Plus probes, the ligation and the amplification steps were carried out following manufacturer indications. Coverslips were stained with DAPI and mounted, following manufacturer indications. Microscopy was performed using inverted microscope Leica DMI-6000. Images were taken with the HQ2 Coolsnap motorized by Metamorph software. All images were processed with ImageJ software.

Total protein extract
Cells were lysed in RIPA buffer to which protease and phosphatase inhibitors (Sigma) were added.

Cell fractionation, co-immunoprecipitation and mass spectrometry sample preparation
Cells were lysed in buffer A (20 mM HEPES pH 7, 0.15 mM EDTA, 0.15 mM EGTA, 10 mM KCl), 10% NP40 and SR buffer (50 mM HEPES pH 7, 0.25 mM EDTA, 10 mM KCl, 70% (m/v) sucrose) with the addition of protease and phosphatase inhibitors (Sigma). Cell lysates were centrifuged at 2000 g for 5 min and the supernatant was recovered (cytoplasmic fraction) (as in (Fritsch et al., 2010)). Then high and low salt buffer were added (20 mM Tris-HCl pH 7.65; 0.2 mM EDTA; 25% glycerol; 900 mM NaCl; 1.5 mM MgCl2) to a final NaCl concentration of 300 mM to extract nuclear proteins. The nuclear extracts were next treated with Micrococcal nuclease (0.0125 U/ml) at 37°C during 10 min in addition to 1 mM CaCl2. EDTA was then added to 4 mM final concentration in order to stop the nuclease reaction. Finally, sonication was performed for 10 min (30 sec ON, 30 sec OFF) at medium frequency (Bioruptor Diagenode).
The nuclear and cytoplasmic fractions were ultra-centrifugated at 40000 rpm for 30 min and precleared with protein G-agarose beads (Sigma) during 2h at 4°C. Immunoprecipitations were carried out overnight at 4°C using 5 μg of each antibody. Ultralink A/G beads (Perbio) were blocked overnight at 4°C with 0,3 % BSA and 0.5 μg/μl ssDNA and then incubated with the immunocomplexes the next day for 2 h at 4°C.
For immunoprecipitation using Myc-Trap technology, the Myc-Trap beads were equilibrated with wash buffer (10 mM Tris pH 7.5, 150 mM NaCl, 0.15 mM EDTA). Immunoprecipitations were carried out for 1 h or overnight at 4°C using 25 μL of Myc-Trap bead slurry.
The immunocomplexes were washed four times in wash buffer and the proteins were eluted in NuPAGE 4X loading buffer (Life Technologies) and 10X reducing agent at 96°C during 5 min. Finally, the samples were examined either by western blot or by mass spectrometry at the Taplin facility (as in (Fritsch et al., 2010)).

Mass spectrometry data analyses
Analysis were performed on triplicates except for KO-Trim71 since one sample was distant from the other samples in PCA and had a very low sum of peptide intensities (Fig. S4A). This sample was removed for the analyses, consequently two samples were used for KO-Trim71. As the Trim 71 interactors are expected to bind only to the GFP-Trim column, only the peptides with 50% of valid values were filtered, and, for the consistency of the results, displaying at least 2 valid values in at least one triplicate, allowing the selection of 826 proteins. Missing values were replaced from normal distribution before scale to interval normalization. ANOVA multi-sample test was done with a FDR of 0.1, and the Post Hoc Tukey's HSD test with a FDR of 0.05.

Western blot
Total, nuclear or cytoplasmic protein extracts, or IP samples were prepared in NuPAGE 4X loading buffer (Life Technologies) and 10X reducing agent and hit at 96°C for 5 min. Separations of proteins were performed using pre-cast polyacrylamide gel cassettes (NuPAGE® Novex® 4-12% Bis-Tris) (Life technologies) and 1X NuPAGE MOPS SDS Running Buffer and transferred into nitrocellulose membranes in 20 mM phosphate transfer buffer (pH 6.7). Membranes were blocked in PBS supplemented with 0.1% Tween 20 and 5% nonfat dried milk and incubated with primary antibodies.
Primary antibodies were detected with IRDYE-conjugated secondary antibodies and scanned on the LI-COR imaging system. Quantification of bands was performed with LI-COR software.
- - Table 3: Gene composition of the clusters described in Figure 4G-I. Figure 1 Cytoplasmic Setdb1   Cycloheximide, a translation inhibitor (Ctr (Puromycin+CHX)) as negative control. FACS data were quanti ed (using the median values) and represented as scatter plot with bars. N = 4. Statistic test: ttest *p value < 0.05; **p value < 0.01; ***p value < 0.001; ns: not signi cant. Directed Yeast-two-hybrid test showing the interaction between Setdb1 and Trim71. Bottom panel: representation of the Trim71 major domains and the domain which mediates the interaction between Trim71 and Setdb1 (in red, aa 358-397). (B) Co-Immunoprecipitation assay performed from the cytoplasmic fraction of ES cells stably expressing agged WT-Setdb1 and transiently expressing GFP-Trim71 (on the left) or Myc-Trim71 (on the right). Trim71 was detected with anti-GFP (on the left) or anti-Myc and anti-Trim71 (on the right) antibodies and the co-immunoprecipitated Setdb1 was detected with anti-Flag (on the left) or anti-Setdb1 (on the right) antibodies. The plasmid expressing GFP (on the left) or the empty plasmid (on the right, W/O) were used as negative controls. MW: molecular weight marker. (C) In situ proximity ligation assay using anti-Myc and anti-GFP antibodies, performed in ES cells transiently expressing Myc-Setdb1 and GFP-Trim71. ES cells transiently expressing Myc-Setdb1 and GFP-Par4 were used as negative control. Green: PLA signal. Red: DAPI. PLA signals were quanti ed (number of spots) and represented as scatter plot with bars. ****p value < 0.0001. (D) In situ proximity ligation assay using anti-Setdb1 and anti-Dcp1a performed in ES cells. Anti-GM130 was used as negative control Green: PLA signal. Red: DAPI. PLA signals were quanti ed (number of spots) and represented as scatter plot with bars. ****p value < 0.0001. (E) Venn diagram showing the numbers and overlap of RNAs that are differentially abundant between KO+WT and KO+NLS (differentially expressed genes (DEGs) were selected for a logFC >= 1.5 and p-value< 0.01) and differentially regulated genes after Trim71-KO (PMID: 31371437). (F) RT-qPCR for endogenous Ccnd2, H60b, Ccnd1, Inhbb mRNAs performed in Ctr, KO, KO + WT, KO + NLS to validate RNAseq. N=3. Statistic test: t-test *p value < 0.05; **p value < 0.01; ***p value < 0.001; ****p value < 0.0001; ns=not signi cant. (G) SUnSET experiment performed in Ctr and Trim71 KD cells. Trim71 KD was obtained after 72 hours of siRNA treatment. N=4. Statistic test: t-test **p value < 0.01. KO-Ctr (diagonal cross), ES-Trim71 (square), KO-Trim71 ( lled square). (D) Differential analysis (multiple sample test ANOVA) after quanti cation of almost 800 proteins of which 108 are differentially enriched at FDR < 0.1 level. Heatmap after clustering of these signi cant proteins with the sum of their intensities. (E) Heatmap after clustering of the signi cant proteins with the score obtained by the Tukey's honestly signi cant difference (THSD, B). (F) List of gene names of the mRNAs that are differentially expressed in KO+WT and found associated to hnRNP-c by eCLIP, see text for details. Number refers to LogFC. (G) Venn diagram showing the clusters of the 157 mRNAs differentially abundant in absence of cSetdb1 function and the mRNAs whose 3'UTR are potential binding site of hnRNP-C. (H) Clustered correlation plot reveals 4 distinct co-expressed modules. Soft-thresholded correlation matrix of the 291 DEG TPMs clustered by hierarchical clustering and the Ward D2 linkage identi es the structure of 5 modules (the 5th module containing the rest 117 DEGs with relatively low correlation, was excluded from the gure). Each module represents a set of co-expressed genes which might also be co-regulated. (I) GO enrichment DotPlot of the second co-expressed module (Cluster 2) of the correlation plot of panel H. Enrichment identi es GO categories related to MAPK activation and WNT signaling. Again (like in Fig. 2