Recruitment of TRIM33 to cell‐context specific PML nuclear bodies regulates nodal signaling in mESCs

Abstract

TRIM33 is a chromatin reader required for mammalian mesendoderm differentiation after activation of Nodal signaling, while its role in mESCs is still elusive. Here, we report that TRIM33 co‐localizes with promyelocytic leukemia nuclear bodies (PML‐NBs) specifically in mESCs, to mediate Nodal signaling‐directed transcription of Lefty1/2. We show that TRIM33 puncta formation in mESCs depends on PML and on specific assembly of PML‐NBs. Moreover, TRIM33 and PML co‐regulate Lefty1/2 expression in mESCs, with both PML protein and formation of mESCs‐specific PML‐NBs being required for TRIM33 recruitment to these loci, and PML‐NBs directly associating with the Lefty1/2 loci. Finally, a TurboID proximity‐labeling experiment confirmed that TRIM33 is highly enriched only in mESCs‐specific PML‐NBs. Thus, our study supports a model in which TRIM33 condensates regulate Nodal signaling‐directed transcription in mESCs and shows that PML‐NBs can recruit distinct sets of client proteins in a cell‐context‐dependent manner.

mESC‐specific PML‐NBs including TRIM33 bind to Lefty1/2 loci and regulate gene cluster expression.

Introduction

Transforming growth factor beta (TGF‐β) superfamily proteins play indispensable roles in early embryonic development, immunity, and tissue homeostasis, among other processes (Massague, 2008; Arnold & Robertson, 2009; Bai et al, 2013; David & Massague, 2018). TRIM33, also called TIF1γ/Ectodermin (Dupont et al, 2005; He et al, 2006; Bai et al, 2010; Rossmann et al, 2021), is specifically associated with SMAD2/3 by Nodal signaling and can recognize H3K9me3 and H3K18ac histone marks through its PHD‐Bromo cassette to regulate the expression of mesendodermal master regulator genes (Xi et al, 2011). Consistent with cell‐context‐dependent functions of the TGF‐β signaling pathway, TRIM33 has been reported to regulate different groups of genes in mESCs and differentiated cells following activation of Nodal signaling (Xi et al, 2011; Quail et al, 2013). But the mechanisms by which TRIM33 participates in this cell‐context‐dependent gene regulation are still unknown.

Biomacromolecules such as proteins and nucleic acids can condense into liquid‐like membrane‐less organelles via liquid–liquid phase separation (LLPS), which enables their increased local concentration and segregation from other cellular components in a spatiotemporally defined manner, as required for diverse biological processes (Alberti, 2017). The occurrence of LLPS has been reported to provide a mechanistic framework that helps to explain many previously unresolved biological phenomena (Brangwynne et al, 2009). For example, several transcription factors, RNA‐binding proteins, and chromatin readers have been found to undergo LLPS as part of their transcriptional regulatory activity (Hnisz et al, 2017; Strom et al, 2017; Cho et al, 2018; Lu et al, 2018; Sabari et al, 2018; Wang et al, 2019; Daneshvar et al, 2020; Wei et al, 2020; Zhang et al, 2022). However, whether LLPS might contribute to cell‐context‐dependent transcriptional regulation upon activation of TGF‐β signaling need to be further investigated.

The membrane‐less organelle PML nuclear bodies (NBs) consist of a shell layer comprised of PML proteins surrounding an inner core that contains dozens of client proteins (Corpet et al, 2020). These client proteins have been proposed to function in specific cellular context mediated by PML NBs (Lallemand‐Breitenbach & de The, 2018), including transcriptional regulation, chromatin organization, RNA splicing, stem cell self‐renewal, DNA damage response, apoptosis, and defense against viral infection (Wu et al, 2001; Conlan et al, 2004; Daniels et al, 2004; de Stanchina et al, 2004; Jiang et al, 2009; Chuang et al, 2011; Chang et al, 2013; Lusic et al, 2013; Wu et al, 2014; Dutrieux et al, 2015; Hadjimichael et al, 2017; Xu & Roizman, 2017; Cabral et al, 2018; Wang et al, 2018; Li et al, 2019a; Wang et al, 2020). Indeed, increasing evidence supports that PML NBs serve as a structural platform to recruit other TFs to the promoter region of their regulatory targets (Corpet et al, 2020; Li et al, 2020).

In this study, we report that TRIM33 condenses with PML NBs in mESCs but not in the differentiated cells. Moreover, genetic evidence showed that PML and TRIM33 co‐regulate transcription of pluripotency‐related genes, Lefty1/2, under activated Nodal signaling. ChIP‐seq results showed that TRIM33 association with chromatin at specific loci depends on both PML and properly assembled PML NBs. Remarkably, DNA‐FISH showed that PML NBs specifically associate with the Lefty1/2 loci in mESCs. Lastly, TurboID proximal labeling experiment further confirmed the enrichment of TRIM33 in mESCs‐specific PML NBs and revealed that PML NBs recruit distinctly different sets of client proteins depending on the cell contexts. Taken together, PML NBs form a regulatory hub for TRIM33 control of Lefty1/2 expression in mouse embryonic stem cells.

Results

TRIM33 forms puncta in the nucleus of mESCs but not in the differentiated cells

The histone H3K4‐K9me3/H3K18ac reader protein TRIM33 is known to transcriptionally regulate TGFβ (Nodal, hereafter) signaling during early embryonic development (Xi et al, 2011; Massague & Xi, 2012). We observed TRIM33 puncta in the nuclei of the TRIM33‐expressing (GFP‐TRIM33) cells but not in the GFP‐alone control mESCs (Figs 1A and EV1A). Nor were TRIM33 puncta detectable in differentiated cells (i.e., mESCs cultured without LIF [leukemia inhibitory factor] at day 4; Fig 1A). It should be noted that the puncta of TRM33 in mESCs are not overlapped with the DAPI (Fig 1B), although TRIM33 is known as chromatin reader. Given the DAPI puncta is in general considered as the heterochromatin, it suggests the TRIM33 is associated with transcriptionally active chromatin in mESCs. Our previous study showed that mESCs are known to differentiate into mesendoderm lineages around Day 3 and 4 upon withdrawal of LIF (Xi et al, 2011; Luo et al, 2019), so “differentiated cells” hereafter refers to differentiated mesendoderm cells.

Figure 1. TRIM33 forms puncta in the nucleus of mESCs but not in the differentiated cells and interacts stronger with PML in mESCs than in differentiated cells.

1. Confocal microscopy of GFP‐TRIM33‐overexpressing mESCs and differentiated cells. Scale bar = 5 μm.
2. Line scans of GFP‐TRIM33 and DAPI fluorescence imaging in each cell context at the position depicted by the white line (from left to right). These line scans were processed by plot profile using image J software.
3. FRAP assays showing recovery time of normalized fluorescence signals for GFP‐TRIM33 puncta in mESCs. Photobleaching occurs at t = 5 s. Data are the mean ± SD of three biological experiments.
4. Sphericity analysis and 3D imaging by NIS‐Elements 3D tools of TRIM33 puncta in live mESCs. Panels show the xz, xy, and yz planes, from left to right. Scale bar = 1 μm.
5. Table of top ranking GFP‐TRIM33 interacting proteins. GFP‐trap‐based affinity purification of extracts prepared from both mESCs and differentiated cells were carried out to identify factor(s) that associate with GFP‐TRIM33. Mass spectrometry (MS) analysis was performed following the affinity purification.
6. Western blot analysis of TRIM33 interactions with PML in lysates of HEK293T cells co‐transfected with plasmids encoding Flag‐PML and GFP‐TRIM33. GFP‐TRIM33 was immunoprecipitated with anti‐GFP affinity beads and immune complexes were detected using antibodies targeting Flag and GFP. Protein inputs were detected using antibodies against Flag and GFP in the same amount of cell lysates (n = 3).
7. Lysates from mESCs and differentiated cells were immunoprecipitated with anti‐PML antibody, and protein complexes and inputs were analyzed by immunoblotting with antibodies against TRIM33 and PML (n = 3).
8. Lysates from HEK293T cells co‐transfected with plasmids encoding either full‐length Flag‐TRIM33 or the Flag‐TRIM33‐RBCC domain alone and full‐length HA‐PML or the HA‐PML‐RBCC domain alone immunoprecipitated with anti‐Flag affinity beads. Protein complexes were detected using antibodies against Flag and HA. The protein inputs were detected using antibodies against Flag and HA in the same amount of cell lysates (n = 3).
9. Schematic diagram of TRIM33 and PML interaction domains.

Data information: All data are from three biological replicates.

Source data are available online for this figure.

Figure EV1. TRIM33 forms puncta in the nucleus of mESCs but not in the differentiated cells and interacts stronger with PML in mESCs than in differentiated cells. Related to Fig 1 .

1. GFP‐TRIM33 puncta images in mESCs. Scale bar = 10 μm (n = 3).
2. Confocal microscopy for KO + GFP‐TRIM33 puncta of mESCs and differentiated cells, KO + GFP‐EV as a control. Scale bar = 5 μm (n = 3).
3. Western blot analysis to detect protein levels of endogenous TRIM33 and overexpressed GFP‐TRIM33 in WT, Trim33 null, KO + GFP‐EV and KO + GFP‐TRIM33 mESCs (n = 3).
4. AP staining of WT, Trim33 null, KO + GFP‐EV and KO + GFP‐TRIM33 mESCs. Scale bar = 50 μm (n = 3).
5. Western blot analysis to protein levels of endogenous TRIM33 during embryoid differentiation (n = 3).
6. Table of top ranking GFP‐TRIM33 interacting proteins in nucleoplasm of mESCs compared to differentiated cells. GFP‐trap‐based affinity purification of extracts prepared from both mESCs and differentiated cells were carried out to identify factor(s) that associate with GFP‐TRIM33. Mass spectrometry (MS) analysis was performed following the affinity purification.
7. Identification of domains required for TRIM33‐PML interaction. Lysates from HEK293T cells co‐transfected with plasmids encoding either full‐length Flag‐TRIM33, the Flag‐TRIM33‐RBCC or Flag‐TRIM33‐PB domain alone and full‐length HA‐PML immunoprecipitated with anti‐Flag affinity beads. Protein complexes were detected using antibodies against Flag and HA. The protein inputs were detected using antibodies against Flag and HA in the same amount of cell lysates (n = 3).

Data information: All data have three biological replicates.

To exclude the possibility that the observed TRIM33 puncta in the undifferentiated mESCs represent some artifact of aberrantly high TRIM33 expression, we also detected TRIM33 puncta in nuclei of TRIM33‐KO mESCs complemented with the GFP‐TRIM33 reporter at a level similar to the endogenous level detected in unmodified (wild‐type, WT) cells (Fig EV1B and C). Additionally, AP (alkaline phosphatase) staining assays showed no changes in AP activity among these cell lines (Fig EV1D), suggesting that our experimental modulation of TRIM33 expression in mESCs does not affect mESC pluripotency.

A common hallmark of proteins that undergo LLPS is their capacity to rapidly shift in and out of the solution (Li et al, 2012; Hyman et al, 2014; Patel et al, 2015). FRAP (Fluorescence Recovery After Photobleaching) assays in mESCs showed that more than 50% of the original GFP‐TRIM33 fluorescence was restored within 25 s of bleaching, suggesting a liquid‐like behavior of TRIM33 puncta (Fig 1C). Additionally, sphericity analysis showed that these puncta were highly spherical (sphericity value: 0.905) (Fig 1D and Movies [Link], [Link]). Note that the TRIM33 protein level is similar between undifferentiated and differentiated mESCs (Fig EV1E). Thus, our results revealed that TRIM33 puncta form specifically in undifferentiated mESCs and are not overlapped with DAPI.

TRIM33‐PML interaction is stronger in mESCs than in differentiated cells

Given that TRIM33 puncta are present in mESCs but not differentiated cells, we performed GFP‐trap‐based affinity purification of extracts prepared from both mESCs and differentiated cells to identify factor(s) that may contribute to the observed difference. Mass spectrometry (MS) analysis identifies candidate interacting partner proteins of TRIM33 including several known binding partners of TRIM33 (e.g., TRIM24, TRIM28, HDAC1, and SMAD2) in mESCs and in differentiated cells (Fig 1E; Dataset EV1; He et al, 2006; Bai et al, 2010). Additionally—and consistent with current understanding of TRIM33 interactions with activated Nodal signaling in the regulation of mesendoderm differentiation (Xi et al, 2011; Luo et al, 2019)—MS analysis showed that the interaction between SMAD2 and TRIM33 is significantly stronger in differentiated cells than in mESCs (Fig 1E). Conversely, we detected that the promyelocytic leukemia (PML) protein (also called TRIM19) had 2.58‐fold higher affinity for TRIM33 in mESCs than in differentiated cells and among the top of all the proteins that show stronger binding affinity with TRIM33 in the ES than that of the differentiated cells (Figs 1E and EV1F).

Our co‐immunoprecipitation (Co‐IP) assays confirmed that TRIM33 interacts with PML in HEK293T cells ectopically co‐expressing GFP‐TRIM33 and FLAG‐PML (Fig 1F). And endogenous Co‐IP experiments showed that while a TRIM33‐PML complex forms in both mESCs and differentiated cells (Fig 1G), the interaction between TRIM33 and PML was obviously stronger in mESCs than in differentiated cells (Fig 1G). Both PML and TRIM33 are TRIM (tripartite motif) family proteins that contain a characteristic conserved RBCC domain (Hatakeyama, 2011). Experiments examining serial truncation variants of both proteins showed that the TRIM33‐PML interaction is mediated by binding of their respective RBCC domains (Figs 1H and I and EV1G).

TRIM33 puncta formation is dependent on proper assembly of PML NBs

Given that PML NBs are considered to be formed via LLPS (Banani et al, 2016), we next investigate whether TRIM33 puncta formation depends on PML NBs. We established Pml null mESCs using CRISPR/Cas9 (Figs EV2A and B). Immunofluorescence (IF) staining of endogenous PML NBs confirmed the presence of phase‐separated puncta in WT mESCs, but not in Pml null mESCs (Fig 2A). Next, we confirmed that the expression of either mCherry‐ or GFP‐tagged PML did not interfere with normal self‐renewal or differentiation in mESCs (Fig EV2C), and PML NBs could be clearly observed in both fluorescence‐tagged mESC lines (Fig 2B). FRAP assays showed that PML NBs could recover fluorescence after photobleaching, but with slower recovery time (203 s) than that of TRIM33 puncta (25 s) in mESCs (Fig 2C).

Figure EV2. TRIM33 puncta formation is dependent on proper assemble of PML NBs and TRIM33 puncta colocalize with DAPI puncta in the absence of PML. Related to Fig 2 .

1. Schematic diagram for establishment of Pml KO mESCs using CRISPR/Cas9 system.
2. Western blot analysis using anti‐PML antibody to validate the knockout efficiency. *present PML major band.
3. AP staining of mCherry‐tagged or GFP‐tagged PML overexpression mESCs. Scale bar = 50 μm (n = 3).
4. Representative images of co‐localization assays with mCherry‐PML and GFP‐TRIM33 in mESCs, differentiated cells, NaAsO₂‐treated mESCs and Pml null mESCs. Scale bar = 5 μm.
5. Western blot analysis of endogenous TRIM33, PML and GFP‐PML with the indicated concentrations of NaAsO₂ treatment in mESCs. GAPDH as a control (n = 3).
6. Scatter plot of numbers and volume of PML NBs in mESCs, NaAsO₂ treated mESCs and differentiated cells. Twenty individual puncta with three biological experiments were analyzed. Significant differences were determined by t‐tests (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).
7. Western blot analysis of TRIM33 level in three GFP‐TRIM33 expressing mESCs cell lines under Pml KO and GFP‐TRIM33 expressing under Trim33 KO background (n = 3).

Data information: All data have at least three biological replicates.

Figure 2. TRIM33 puncta formation is dependent on proper assembly of PML NBs and TRIM33 puncta colocalize with DAPI in the absence of PML.

• A
  Representative images of immunofluorescence staining of PML in WT and Pml null mESCs. Scale bar = 5 μm.
• B
  Representative confocal microscopy images of mESCs overexpressing mCherry‐PML or GFP‐PML (top). Representative confocal microscopy images of mESCs overexpressing mCherry‐EV or GFP‐EV (bottom). Scale bar = 5 μm.
• C
  FRAP assays quantifying normalized fluorescence recovery of GFP‐PML signal in mESCs. Photobleaching occurs at t = 5 s. Data are the mean ± SD of four biological experiments.
• D–L
  (D, G, I, K) Representative images of co‐localization assays with mCherry‐PML and GFP‐TRIM33 in mESCs (D), differentiated cells (G), NaAsO₂‐treated mESCs (200 μM for 2 h) (I) and Pml null (K). Scale bar = 5 μm. (E, H, J, L) Line scans of GFP‐TRIM33 and mCherry‐PML fluorescence imaging in each cell context at the position depicted by the white line (from left to right). These line scans were processed by plot profile using image J software. (F) Column graph of the numbers of GFP‐TRIM33 puncta and mCherry‐PML NBs in mESCs.
• M
  Column graph of the numbers of GFP‐TRIM33 puncta and DAPI in mESCs.
• N
  Scatter plot of the numbers of GFP‐TRIM33 puncta (per cell) in WT and Pml KO mESCs (n = 20).
• O
  Scatter plot of the volume of GFP‐TRIM33 puncta (per cell) in WT and Pml KO mESCs (n = 20).
• P
  Representative images of immunofluorescence staining of PML in WT and Trim33 null mESCs. Scale bar = 10 μm.
• Q
  Scatter plot of the numbers of PML NBs (per cell) in WT and Trim33 KO mESCs (n = 20).
• R
  Scatter plot of the proportion of PML NBs (per cell) in WT and Trim33 KO mESCs (n = 20).

Data information: In panels (N, O, Q, and R), 20 individual puncta with three biological experiments were analyzed. Significant differences were determined by t‐tests (*P < 0.05, ****P < 0.0001).

Source data are available online for this figure.

We then examined the subcellular localization patterns of TRIM33 and PML in both mESCs and differentiated cells using mESCs co‐expressing mCherry‐PML and GFP‐TRIM33 (which was previously shown to not interfere with mESCs functionality; Luo et al, 2019; Xi et al, 2011). The imaging results showed that almost 95% of GFP‐TRIM33 puncta co‐localized with PML NBs in the nucleus of mESCs, whereas 34% of mCherry‐PML NBs do not co‐localize with GFP‐TRIM33 puncta (Figs 2D–F and EV2D), suggesting that TRIM33 puncta formation relies on PML NBs. In addition, no TRIM33 puncta were detectable in differentiated cells (Fig 2G and H). The number of PML NBs in differentiated cells was significantly less than that in mESCs but had greater volume (Figs 2G and EV2D and F). In light of our above finding that TRIM33 association with PML was weaker in differentiated cells (Fig 1G), we hypothesized that this decreased association could result in the dissolution of TRIM33 puncta in differentiated cells. Conversely, TRIM33 puncta formation in mESCs would thus depend on stronger interaction between PML and TRIM33.

PML NBs are highly sensitive to oxidative stress (Lallemand‐Breitenbach et al, 2008; Lallemand‐Breitenbach et al, 2012; Gartner & Muller, 2014; Sahin et al, 2014). Western blotting analysis showed that treatment with 200 μM NaAsO₂ for 2 h led to the degradation of 70% of the PML in mESCs (Fig EV2E), while PML NBs in NaAsO₂‐treated ESCs were both fewer in number and exhibited larger volumes than those in untreated mESCs (Figs 2I and EV2D and F). These results indicated that the client proteins of PML NBs differed among these three conditions. No TRIM33 puncta were detected in NaAsO₂‐treated mESCs (Figs 2I and J and EV2D). This result further confirmed that TRIM33 puncta formation in mESCs would depend on proper assembly of PML NBs.

TRIM33 puncta colocalize with DAPI in the absence of PML

We then examined the TRIM33 puncta formation in Pml KO cells. The expression of GFP‐TRIM33 under Pml KO background is about the same level of the GFP‐TRIM33 under Trim33 KO background (Fig EV2G). Interestingly, the numbers of the TRIM33 puncta were significantly decreased, however, the size of the remained TRIM33 puncta is dramatically enlarged in Pml KO cells compared to WT mESCs (Figs 2K–O and EV2D). Remarkably, the TRIM33 puncta are completely overlapped with DAPI under PML KO (Fig 2K–M). Recalling that TRIM33 puncta are not overlapped with DAPI in WT mESCs, these results suggest that PML might prevent TRIM33 puncta colocalization with DAPI staining by recruiting TRIM33 into the PML NBs.

We also examined whether deletion of TRIM33 has impact on PML NBs. The numbers of PML NBs are not altered significantly by depletion of TRIM33, nor the volume of PML NBs in the Trim33 KO cells (Fig 2P–R). Thus, TRIM33 does not impact on PML NBs numbers and volumes.

TRIM33 and PML regulate transcriptional activation of Lefty1/2 in mESCs

Given that TRIM33 puncta formation depends on proper assembly of PML NBs in mESCs, in conjunction with PML is essential for pluripotency in mESCs (Chuang et al, 2011), we next sought to characterize the full suite of PML target genes in mESCs. Analysis of the sequencing data from Pml KO and WT mESCs revealed the expression of some pluripotency‐related genes was significantly decreased in Pml KO mESCs (Fig 3A and B) (¦log₂FC¦ > 0.67 and −log₁₀(p) > 1.5). The AP activity was significantly lower in Pml KO cells than that in WT mESCs by AP staining assay and the morphology of the cells is dramatically changed (Fig 3C), validating the essential role of PML for maintaining pluripotency in mESCs (Chuang et al, 2011). In addition, deletion of PML also led to earlier induction differentiation‐associated gene expression (Fig 3D). GSEA (Gene Set Enrichment Analysis) further showed enrichment for upregulated genes involved in EMT (epithelial–mesenchymal transition), apoptosis and hypoxia (Figs 3E and EV3B). The upregulation of EMT‐related genes in Pml KO mESCs is consistent with our observation that knock‐out PML led to early onset of mesendoderm differentiation.

Figure 3. TRIM33 and PML regulate transcriptional activation of Lefty1/2 in mESCs.

1. Volcano plot of RNA‐seq data showing WT versus Pml KOs. Genes with ¦log₂FC¦ > 0.67 and −log₁₀(p) > 1.5 are highlighted in blue (downregulated) or red (upregulated).
2. Scatter plot of pairwise comparison of pluripotency‐related gene expression in WT versus Pml KO mESCs.
3. Alkaline phosphatase (AP) staining of Pml WT and null mESCs. Scale bar = 20 μm.
4. Pml‐WT and KO mESCs were induced to embryoid bodies formation for the indicated times. Total RNA was analyzed by qPCR using primers for MixL1, Gsc and T.
5. GSEA of RNA‐seq data for WT versus Pml KO mESCs to show significant enrichment for upregulated genes associated with EMT and apoptosis in Pml KO mESCs. NES, normalized enrichment score (Data are from two biological replicates).
6. Heatmap of RNA‐seq comparison of WT versus Trim33 KO mESCs treated with activin A (50 ng/ml) or SB431542 (10 μM) for 2 h. The results represent differential expression values between activin A/SB431542 treatments (Data are from two biological replicates).
7. GSEA of RNA‐seq data for WT versus Trim33 KOs to show significant enrichment for upregulated genes associated with EMT in activin A‐treated Trim33 KO mESCs (Data are from two biological replicates).
8. Venn diagram of overlapping gene list between DEGs of Trim33 null and WT mESCs treated with activin A or SB431542 and DEGs of Pml null and WT mESCs.
9. qPCR analysis of the normalized Lefty1/2 response to Nodal signaling activation in WT and Trim33 null mESCs treated with activin A (50 ng/ml) or SB431542 (10 μM) for 2 h.
10. qPCR analysis for normalized Lefty1/2 response to Nodal signaling response of the in Pml WT and null mESCs treated with activin A (50 ng/ml) or SB431542 (10 μM) for 2 h.
11. qPCR analysis of Lefty1/2 expression levels in WT, Pml KO, and GFP‐PML complemented Pml KO mESCs.
12. Transcript counts of Lefty1/2 during mesendoderm differentiation at the indicated times (Data are from two biological replicates) (GSE70486).
13. Volcano plot of RNA‐seq data for mESCs treated or not with NaAsO₂. Genes with ¦log₂FC¦ > 1.5 and ‐log₁₀(p) > 1.5 are highlighted in blue (downregulated) or red (upregulated).

Data information: In panels (D, I, J, and K), data are the mean ± SD of triplicate measurements and are representative of three biological replicates. Significant differences were determined by t‐tests (**P < 0.01, ***P < 0.001, ****P < 0.0001). In panels (E, F, G, and L) data have two biological replicates.

Figure EV3. TRIM33 and PML regulate transcriptional activation of Lefty1/2 in mESCs. Related to Fig 3 .

1. GSEA of WT and Pml KO RNA‐seq to show significant enrichment for upregulated genes associated with hypoxia and UV response in Pml KO mESCs (Data are from two biological replicates).
2. AP staining of mESCs with/without NaAsO₂ (200 μM) treatment for 2 h. Scale bar = 20 μm (Data are from two biological replicates).
3. qPCR analysis for mRNA levels of Lefty1/2, Oct4, Nanog and Sox2 in mESCs with NaAsO₂ treatments for 2 h at indicated concentration.
4. qPCR analysis for mRNA levels of Lefty1/2, Oct4, Nanog and Sox2 at the indicated conditions.

Data information: In panels (C, D), data are the mean ± SD of triplicate measurements and are representative of three independent experiments with similar results. Significant differences were determined by t‐tests (**P < 0.01, ***P < 0.001, ****P < 0.0001).

Nodal signaling is responsible for predominant regulatory functions in the control of EMT (Lamouille et al, 2014). We next sought to characterize the full suite of TRIM33 target genes upon activation by Nodal signaling in mESCs. To this end, we performed RNA‐seq in Trim33 null and WT mESCs treated with activin A or SB431542 (Fig 3F). Here, activin A was used as a substitute for Nodal in our study because it is easier to obtain than Nodal and because these two protein ligands act through the same receptors (Yeo & Whitman, 2001). Consistent with previous studies (Xi et al, 2011; Luo et al, 2019), Lefty1/2 are included among the targets of TRIM33 identified in cells with activation of Nodal signaling (Fig 3F). LEFTY1 and LEFTY2 belong to the TGFβ family and are transcriptional targets of Nodal signaling (Hamada et al, 2002; Schier, 2003; Arnold & Robertson, 2009; Zabala et al, 2020). Consistently, GSEA showed significant enrichment for upregulated genes associated with EMT in activin A‐treated Trim33 KO mESCs compared to activin A‐treated WT mESCs (Fig 3G).

Given that KO of either Pml or Trim33 leads to upregulation of EMT‐related genes, it was reasonable to speculate that PML might also participate in regulating Nodal signaling. Identification of genes regulated by TRIM33 via Nodal signaling and genes regulated by PML in mESCs in the above RNA‐seq data, revealed Lefty1/2 were among the most significantly impacted genes on this list of overlapping regulatory targets (Fig 3H). Moreover, we found that the induction of Lefty1/2 was impaired under activation of Nodal signaling in both Trim33 KO and Pml KO mESCs (Fig 3I and J). In addition, re‐expression of Pml led to the rescue of decreased Lefty1/2 expression in Pml null mESCs (Fig 3K). These results indicated that both TRIM33 and PML participated in regulating the induction of Lefty1/2 following activation of Nodal signaling. Moreover, Lefty1/2 expression both rapidly decreased during mESC differentiation (Fig 3L).

AP staining showed the significantly decreased AP activity in the NaAsO₂‐treated cells (Fig EV3B), which is similar as Pml KO mESCs. Consistently, RNA‐seq analysis of mESCs treated with NaAsO₂ showed significantly decreased expression of pluripotency genes, including Lefty1/2 (Fig 3M). And qPCR analysis showed that the decreased expression of these genes is in a dosage‐ and time‐dependent manner by NaAsO₂ treatments (Fig EV3C and D). These findings indicated that both PML and specific assembly of PML NBs are essential for Lefty1/2 expression.

PML is required for TRIM33 association at Lefty1/2 loci in mESCs

To determine the specific role of PML in transcriptional regulation of Lefty1/2 in mESCs, we performed GFP‐TRIM33 ChIP‐seq in GFP‐TRIM33‐expressing WT and Pml KO mESCs. We treated cells with activin A or SB431542 to respectively activate or suppress Nodal signaling. As expected, PML deletion not only impaired TRIM33 puncta formation, it also led to significant genome‐wide disruption of TRIM33 association with chromatin. Specifically, the 27,369 total peaks of TRIM33 (including 8,950 peaks in promoter regions) detected in activin A‐treated WT cells decreased to 9,414 peaks (including 2,391 peaks in promoter regions) in activin A‐treated Pml KO cells (Fig EV4A), suggesting that PML promotes TRIM33 association with chromatin at specific loci. Furthermore, TRIM33 association with Lefty1/2 loci appeared weaker under PML deletion (Fig 4A). Consistent with this effect, PML deletion also led to reduced SMAD2/3 binding and decreased H3K27ac enrichment at the Lefty1/2 promoters (Figs 4A and EV4B). These findings confirmed that PML facilitates TRIM33‐mediated transcription regulation of Lefty1/2 in mESCs.

Figure EV4. PML is required for TRIM33 association at Lefty1/2 loci in mESCs and transcriptionally regulates a gene cluster in mESCs. Related to Fig 4 .

• A
  Pie chart of genome wide distribution in Pml WT and KO mESCs.
• B
  IGV tracks displaying enrichment for SMAD2/3 at the Lefty1 locus in Pml KO and WT mESCs (n = 2).
• C
  Venn diagram of ChIP‐seq targeting TRIM33 and CUT&TAG targeting PML signals at mouse chromosome 1.
• D, E
  IGV tracks displaying enrichment of ChIP‐seq data for TRIM33, PML (CUT&TAG), H3K27ac and H3K18ac at Vat1, Etv4, Sox21, Abcc4, Frat1, Frat2 and Pi4k2a gene loci in mESCs (n = 2).
• F
  IGV tracks displaying enrichment of ChIP‐seq data for TRIM33, PML (CUT&TAG), H3K27ac, and H3K9me3 at Y‐linked genes cluster: Eif2s3y, Uty, and Ddx3y (n = 2).
• G
  Heatmaps of relative expression for each gene clusters in Pml KO or WT and Trim33 KO or WT mESCs (n = 2).

Data information: All data have two biological replicates.

Figure 4. PML is required for TRIM33 association at Lefty1/2 loci in mESCs and transcriptionally regulates a gene cluster in mESCs.

1. Integrative Genomics Viewer (IGV) tracks displaying enrichment for TRIM33 at the Lefty1 and Lefty2 loci in Pml KO and WT mESCs treated with activin A (50 ng/ml) or SB431542 (10 μM) for 2 h. IVG tracks displaying enrichment for SMAD2/3 and H3K27ac at the Lefty1 and Lefty2 loci in Pml KO or WT mESCs. The Rrs1 locus was shown as a negative control (n = 2).
2. Heatmaps of ChIP‐seq signals for TRIM33, PML (CUT&TAG), H3K18ac, H3K27ac, H3K4me3, H3K9me3, H3K27me3, SMAD2/3 (CUT&TAG), and SMAD4 aligned at ±3 kb of TSS, sorted by overlapping genes between TRIM33 and PML (n = 2).
3. Hi‐C heatmap data aligned with TRIM33 ChIP‐seq and PML CUT&TAG signals.
4. IGV tracks displaying enrichment of ChIP‐seq data for TRIM33, PML (CUT&TAG), SMAD2/3 (CUT&TAG), SMAD4, OCT4, NANOG, H3K27ac and H3K18ac at Lefty1/2 gene cluster loci of Parp1, Sde2, Lefty2, Pycr2, and Lefty1.
5. Heatmaps of relative expression for each gene in Lefty1/2 gene cluster in Pml KO or WT and Trim33 KO or WT mESCs (n = 2).

Data information: All data are from two biological replicates.

PML transcriptionally regulates a gene cluster in mESCs

We then explored the genome‐wide chromatin binding sites by TRIM33 and PML through ChIP‐seq targeting TRIM33 and CUT&TAG assays targeting PML and SMAD2/3 in mESCs (Fig 4B). These analyses, combined with published ChIP‐seq datasets for SMAD4, H3K18ac, H3K27ac, H3K4me3, H3K9me3 and H3K27me3 (GSE125116, GSE169450, GSE99009) revealed a set of 9,571 sites (including 5,678 peaks at promoter regions) co‐occupied by TRIM33 and PML in mESCs that were also sites associated by SMAD2/3 and SMAD4 (Fig 4B). Moreover, these sites were enriched with active H3K18ac, H3K27ac, and H3K4me3 histone marks, but not H3K9me3 or H3K27me3 marks (Fig 4B). These results suggested that TRIM33 and PML could co‐occupy these transcriptionally active sites.

Given that Lefty1/2 are shared targets of PML and TRIM33 and are known to be located on mouse chromosome 1, we next aligned Hi‐C (High‐through chromosome conformation capture) data of chromosome 1 (Kloetgen et al, 2020) (GSE74055) with ChIP‐seq data of TRIM33 and CUT&TAG data of PML (Fig 4C) to generate a 3D genomic map of TRIM33 and PML localization sites. Intriguingly, the overlapping peaks (i.e., co‐localization sites) between TRIM33 and PML were located at the boundaries of TADs (Topologically Associated Domains) on chromosome 1 (Fig 4C), suggesting that both PML and TRIM33 function in these chromatin accessible regions (Achinger‐Kawecka et al, 2016; Madani Tonekaboni et al, 2021; Wang et al, 2021). Moreover, visualization by venn diagram showed that a total of 483 peaks overlapped between PML and TRIM33 on chromosome 1, which included > 95% of the PML peaks for this chromosome (Fig EV4C).

Next, a magnified expanded view of Lefty1/2 loci region that these two genes were part of a gene cluster that included Parp1, Sde2, Lefty2, Pycr2, and Lefty1, in that chromosomal order. Interestingly, TRIM33, PML, SMAD2/3, and SMAD4, were all enriched at the distal and proximal promoters of Lefty1 and at distal promoter of Lefty2, as were H3K27ac and H3K18ac histone marks (Fig 4D). Consistent with previous studies that R‐SMADs are recruited to target gene loci by cell‐specific master regulators (Young, 2011), ChIP‐seq data indicated the presence of both OCT4 and NANOG at the Lefty1/2 loci in mESCs (Fig 4D). In addition, TRIM33, PML, and active histone marks were also enriched at the proximal promoters of Parp1, Sde2, and Pycr2 (Fig 4D). Hence, it is likely PML NBs facilitate their client proteins (i.e., TRIM33, OCT4, and NANOG) to be recruited to the chromatin.

Intriguingly, RNA‐seq data from Pml null and WT mESCs showed that the expression of all genes in this putative cluster were significantly decreased in the absence of PML (Fig 4E), whereas TRIM33 only affected the expression of Lefty1/2 (Fig 4E). Moreover, RNA‐seq combined with CUT&TAG revealed that PML participates in regulating the expression of gene clusters at several loci genome wide, including a previously reported Y‐linked gene cluster (Fig EV4C–G) (Kurihara et al, 2020). Cumulatively, these analyses support the possibility that PML can serve as a hub to regulate the transcriptional activation of various gene clusters, and the Lefty1/2 loci in particular, in mESCs.

PML NBs specifically localize at Lefty1/2 loci in mESCs

This finding prompted us to examine whether PML NBs localized at the Lefty1/2 loci in mESCs, NaAsO₂‐treated mESCs, or differentiated cells. For this purpose, we first performed PML CUT&TAG. The results showed that PML exhibited weaker association with chromatin at the Lefty1/2 loci in both NaAsO₂‐treated and differentiated cells compared to that in WT mESCs (Fig 5A), which was consistent with our above observations of decreased Lefty1/2 expression in these cells. Then, DNA‐FISH (fluorescence in situ hybridization) using a genome fragment containing the Lefty1/2 loci as a probe in GFP‐PML‐expressing mESCs and differentiated cells showed that PML NBs indeed colocalized with the Lefty1/2 loci in mESCs, but not in NaAsO₂‐treated mESCs or in differentiated cells (Fig 5B–D).

Figure 5. PML NBs specifically condense at Lefty1/2 loci in mESCs.

1. IGV tracks of CUT&TAG targeting PML and RNA‐seq at the Lefty1 and Lefty2 loci under the indicated conditions (Data are from two biological replicates).
2. DNA‐FISH (fluorescence in situ hybridization) of Lefty1/2 in GFP‐PML mESCs with or without NaAsO₂ treatment (2 h) and differentiated cells. Scale bar = 5 μm (Data are from three biological replicates).
3. DNA‐FISH of Lefty1/2 in GFP‐EV mESCs (as a negative control). Scale bar = 5 μm (Data are from three biological replicates).
4. Statistical analysis of distance between PML NBs and Lefty1/2 signals in the indicated cell conditions (17 individual puncta with three biological experiments were analyzed. Significant differences were determined by t‐tests, **P < 0.01).
5. IGV tracks displaying enrichment for TRIM33 at the Lefty1 and Lefty2 loci in GFP‐TRIM33 ChIP‐seq with indicated conditions. The Rrs1 locus was shown as a negative control (Data are from two biological replicates).

We then performed TRIM33 ChIP‐seq to investigate whether it also localized at the Lefty1/2 loci in NaAsO₂‐treated mESCs, or differentiated cells. This ChIP‐seq analysis confirmed that TRIM33 showed significantly less association with the Lefty1/2 loci in both mESCs treated with NaAsO₂ and differentiated cells compared to in mESCs (Fig 5E). Thus, it is possible that not only PML, but also proper assembly of PML NBs specific to WT mESCs was required to facilitate TRIM33 recruitment to the Lefty1/2 loci. Thus, it is likely PML functions as a hub in the form of PML NBs to regulate the transcriptional activation various gene clusters, and the Lefty1/2 loci in particular, in mESCs.

Proximity labeling with TurboID shows that client proteins of PML NBs differ in three cell contexts

Given that PML recruits different client proteins under different cell context (Corpet et al, 2020), it hints that the cell‐context‐specific client proteins might determine whether PML NBs colocalize with the Lefty1/2 loci. Then, we used TurboID proximity labeling (PL) method followed by MS to determine the composition of PML NBs in mESCs, differentiated cells, and NaAsO₂‐treated mESCs. We constructed a TurboID‐PML mES cell line with stable expression of GFP‐PML‐TurboID‐FLAG fusion protein. Proximity‐dependent biotinylation was activated by incubating cells in medium with exogenous 200 μM biotin, while equal volume of DMSO was added in parallel as negative control. Proteins within PML NBs would be properly biotinylated and enriched through streptavidin purification (Fig 6A).

Figure 6. Proximity labeling with TurboID shows that client proteins of PML NBs differ in three cell contexts.

1. Schematic diagram of PML NBs purification and client protein identification through proximity labeling mediated by TurboID and subsequent MS analysis.
2. Venn diagram of overlapping and unique PML NB proteins among mESCs treated or untreated with NaAsO₂ and differentiated cells identified by TurboID followed with MS analysis. Threshold was determined by Biotin/DMSO > 2.
3. Heatmap of identified PML NB client proteins in mESCs and NaAsO₂‐treated cells; the threshold was Biotin/DMSO > 2 (Data are from two biological replicates).
4. Heatmap of pluripotent master regulator proteins in mESCs, NaAsO₂‐treated mESCs, and differentiated cells; the threshold was Biotin/DMSO > 2 (Data are from two biological replicates).
5. STRING network analysis of mESC‐specific proteins filtered by Biotin/DMSO > 2 as a cut‐off.
6. STRING network analysis of differentiated cell‐specific proteins filtered by Biotin/DMSO > 2 as a cut‐off.
7. Working model for LLPS‐mediated PML NB recruitment of TRIM33 on Lefty1/2 transcriptional regulation in mESCs.

The pull‐down efficiency was validated for each sample by silver staining and resolved by SDS–PAGE (Fig EV5A). The PML NB client proteins were then identified by MS (Liu et al, 2020; Barroso‐Gomila et al, 2021). We used Biotin/DMSO ratio > 2 as a cut‐off to filter the positive candidate of PML NB client proteins in each group (Fig 6B and Dataset EV2). Consistent with previous reports, all three PML proximity labeling groups trapped primary core client proteins of PML NBs (Barroso‐Gomila et al, 2021), GO analysis confirmed that these core components of PML NBs are proteins involved in transcription, chromatin modification, RNA processing etc. (Fig EV5B). Interestingly, these relative stable components had a dynamic capacity in different environments and some of them showed higher enrichment during stem cell stage: NaAsO₂ treatment indeed altered the well‐known client proteins of PML NBs such as PIAS1/3/4, HDAC1, KDM1A, DAXX, HIRA, RANBP2, and SUMO1/3 (Fig 6C; Hofmann & Will, 2003; Corpet et al, 2020); to be noted, previously identified client protein TRIM33 (Barroso‐Gomila et al, 2021) is also highly enriched in mESCs compared to the differentiated cells and the NaAsO₂‐treated cells (Fig 6C), which explained the regulatory role of PML NBs in TRIM33‐mediated Lefty1/2 induction under Nodal signaling in mESCs.

Figure EV5. Proximity labeling with TurboID shows that client proteins of PML NBs differ in three cell contexts. Related to Fig 6 .

1. Silver staining of PML NBs with TurboID pull‐down. Input as an internal control and DMSO as a negative control. Western blot analysis using anti‐GFP antibody further confirmed PML‐GFP fusion protein was successfully pull‐down (Data are from two biological replicates).
2. Gene Ontology (GO) analysis of the overlapping proteins (527) in the ES‐No treat, Diff and NaAsO₂‐treated cells.
3. STRING network analysis of NaAsO₂ treated cell‐specific proteins filtered by Biotin/DMSO > 2 as a cut‐off.

Closer scrutiny of the candidates revealed that several pluripotency master regulators (i.e., POU5F1, SOX2, NANOG, SALL1, and SALL4 etc.) scored much lower in both differentiated and NaAsO₂‐treated cells than in mESCs (Fig 6D). Given that previous studies have shown PML is essential for self‐renewal and pluripotency in mESCs (Hadjimichael et al, 2017), this result suggests that PML NBs could contribute to maintaining pluripotency by recruiting these master regulators. Thus, it is plausible that PML NBs perform essential roles in the maintenance of pluripotency in mESCs by recruiting clusters of these master regulators to coordinate their transcriptional regulatory activity including promoting Lefty1/2 expression.

Particularly, each group had clustered a batch of specific proteins. In mESCs group, some ES group‐specific proteins contained RNA recognition motif which have been reported in regulation of RNA processing and transporting (Fig 6E). NaAsO₂‐treated group displayed a high activity on DNA damage response (Fig EV5C). Moreover, PML NBs in differentiated cells recruited a vast number of embryonic development‐related proteins, and protein involved in regulation of chromatin organization (Fig 6F). Thus, the different client proteins could be the determining factors for functions of PML NBs under specific cell context.

Discussion

Here, our findings reveal that TRIM33 co‐condenses with PML NBs specifically in mESCs to transcriptionally regulate Lefty1/2 expression upon activation of Nodal signaling. First, we demonstrated that TRIM33 puncta formation dependent on the proper assembly of mESCs‐specific PML NBs and TRIM33 puncta colocalization with DAPI staining in the absence of PML. Moreover, TRIM33 association with chromatin at specific loci depends on PML, and PML appears to regulate a gene cluster including the Lefty1/2 loci in mESCs. Accordingly, PML NBs specifically associate with the Lefty1/2 loci in mESCs. PML NBs recruit distinctly different sets of client proteins in mESCs, NaAsO₂‐treated mESCs, and differentiated cells. Taken together, PML NBs facilitate client protein TRIM33 localization to target gene loci in mESCs and specifically serve as a hub for transcriptional regulation of Lefty1/2 directed by Nodal signaling (Fig 6G).

It is widely accepted that PML NBs undergo LLPS. Previous studies have shown that a mixture of polySUMO‐polySIM polymers can form droplets and recruit SUMO/SIM clients in vitro (Banani et al, 2016). Moreover, LLPS has been demonstrated to facilitate the recruitment of transcription regulators in concentrated spatial proximity to the chromosomal target site (Boija et al, 2018; Cho et al, 2018; Sabari et al, 2018). Given that our study and others show that PML is essential for mESC pluripotency, by extension, stage‐specific component dynamics of PML NB is also required to regulate cellular homeostasis. Our results show that PML NBs help to recruit TRIM33 and potentially other pluripotency factors (i.e., POU5F1 and NANOG) at the Lefty1/2 promoters to activate their expression in mESCs. Thus, PML NBs can serve as a transcriptional regulatory hub for pluripotency factors in mESCs. This relationship could at least partially explain the dissociation of PML NBs from the Lefty1/2 loci in differentiated cells, since these factors no longer actively maintain pluripotency. In light of our results that show TRIM33 co‐condenses with PML NBs in mESCs and co‐regulate Lefty1/2 expression in an mESC‐specific manner, further investigation is need to explore whether and how PML NBs and TRIM33 regulate gene expression in other cellular contexts (such as cancers).

Our TurboID analysis confirmed that PML NBs differentially incorporate client proteins in a manner depending on cellular context (i.e., NaAsO₂‐treated or‐untreated mESCs and differentiated cells). Moreover, FRAP assays indicated that recovery time for PML NBs was substantially longer than that of the TRIM33 puncta, suggesting that PML NBs could provide structural stability to promote TRIM33‐mediated regulatory events. The shorter recovery time of client proteins such as TRIM33 indicates that they could be quickly mobilize in or out of the PML NBs as needed, potentially mediated by LLPS (Lallemand‐Breitenbach & de The, 2010; Lallemand‐Breitenbach & de The, 2018). It also warrants mention that we could observe puncta formation by TRIM33 and PML NBs in mESCs using live cell imaging, and we could detect PML NB puncta formation by IF staining, but could not detect TRIM33 puncta using IF. This anomaly could also be explained by lower stability of TRIM33 puncta compared to that of PML NBs.

Limitations of the study

Our study showed that TRIM33 interacts with PML more weakly in the differentiated cells, although the underlying mechanism remains unknown. Sumoylation of PML leads to development of a spherical, mature PML NBs and is essential for its function in recruitment of client proteins (Lallemand‐Breitenbach et al, 2012; Sahin et al, 2014; Lallemand‐Breitenbach & de The, 2018; Wang et al, 2018). Given that the client proteins are recruited to the PML NBs depend on the cell contexts, it is speculated that the sumoylation of PML might be different in variant cell contexts. Thus, examination the status of the PML sumoylation and identification of the enzyme responsible for modulating PML sumoylation could help to define this mechanism.

Materials and Methods

Reagents and Tools table

Reagent/resource	Reference/source	Identifier
Experimental models
HEK‐293 cells (H. sapiens)	ATCC	Cat# CRL‐1821; RRID: CVCL_Y481
E14Tg2a.IV (M. musculus)	ATCC	Cat# CRL‐11268; RRID: CVCL_1926
Recombinant DNA
FUGW‐GFP‐PML	This paper	NA
FUGW‐GFP‐PML‐TurboID	This paper	NA
FUGW‐TuboID‐GFP‐PML	This paper	NA
FUGW‐GFP‐TRIM33	This paper	NA
FUGW‐GFP‐EV	This paper	NA
FUGW‐mCherry‐PML	This paper	NA
FUGW‐mCherry‐EV	This paper	NA
pCI‐FLAG‐PML	This paper	NA
pCI‐FLAG‐TRIM33	This paper	NA
pCI‐FLAG‐TRIM33‐RBCC	This paper	NA
pCI‐FLAG‐TRIM33‐PB	This paper	NA
pCI‐HA‐PML	This paper	NA
pCI‐HA‐PML‐RBCC	This paper	NA
pCI‐FLAG‐EV	This paper	NA
pCI‐HA‐EV	This paper	NA
Antibodies
Rabbit polyclonal anti TRIM33 Antibody	Bethyl	Cat# A301‐060A‐1; RRID: AB_2920524
Mouse monoclonal anti‐PML	Merck	Cat# 36.1‐104; RRID: AB_309932
Mouse monoclonal anti FLAG® M2 antibody	Sigma‐Aldrich	Cat# F3165; RRID: AB_259529
Rabbit monoclonal anti HA	Millipore	Cat# 04‐902; RRID: AB_1977526
Mouse monoclonal anti‐GAPDH	ZSGB‐BIO	Cat# ta‐08; RRID: AB_2747414
Rabbit monoclonal anti‐H3K27ac	Cell Signaling Technology	Cat# 8173s; RRID: AB_10949503
Rabbit monoclonal anti SMAD2/3	Cell Signaling Technology	Cat# D7G7; RRID: AB_2799861
Chemicals, peptides, and recombinant proteins
Puromycin	Sigma‐Aldrich	Cat# P8833
Polybrene	Sigma‐Aldrich	Cat# TR‐1003
BSA	Amresco	Cat# 332
2‐Mecraptoethanol	Sigma‐Aldrich	Cat# M3148
Activin A	R&D Systems	Cat# 338‐AC‐010
SB431542	selleck	Cat# S1067
DMEM	Thermofisher	Cat# C11995500BT
L‐Glutamine Solution	Biological Industries	Cat# 02‐020‐1B
NEAA	Thermofisher	Cat# 11140050
sodium pyruvate	Sigma‐Aldrich	Cat# S8636
Pen‐Strep‐Ampho.B Solution	Biological Industries	Cat# 03‐033‐1B
FBS	ExCell Bio	Cat# FND500
Lipofectamine 2000	Thermofisher	Cat# 11668019
Opti‐MEM	Thermofisher	Cat# 31985070
Gelatin from porcine skin	Sigma‐Aldrich	Cat# V900863
Lenti‐concentin	Genestar	Cat# C103‐05
Protease Inhibitor Cocktail	Biotool	Cat# B14002
GFP‐Trap Agarose	Chromotek	Cat# gta‐20
Anti‐Flag Affinity Gel	Biotool	Cat# B23102
Protein G Agarose Resin	Smart‐Lifesciences	Cat# 36405ES08
Trypsin–EDTA (0.25%), phenol red	Thermofisher	Cat# 25200072
Critical commercial assays
FISH Tag DNA Red Kit with Alexa Fluor 594 dye	Thermofisher	Cat# F32949
2xRealStar Green Mixture qPCR	Genestar	Cat# A311‐TZ10
Total RNA Purification Kit	DAKEWE	Cat# 8034111
RevertAid First Strand cDNA Synthesis Kit	Thermofisher	Cat# K1622
ClonExpress MultiS One Step Cloning Kit	Vazyme	Cat# C113‐02
Deposited data
RNA‐seq data and ChIP‐seq data	This paper	GEO: GSE199738
Hi‐C data in mESCs	Yan et al (2018)	GEO: GSE74055
TurboID‐PML mass spectrometry	This paper	PRIDE: PXD037432
Lefty1/2 RNA‐seq in D0, D2, D3 and D4	Wang et al (2017)	GEO: GSE70486
H3K4me3 and H3K27me3 ChIP‐seq in mESCs	Cossec et al (2018)	GEO: GSE99009
SMAD4 ChIP‐seq in mESCs	Wang et al (2019)	GEO: GSE125116
OCT4 and NANOG ChIP‐seq in mESCs	Narita et al (2021)	GEO: GSE146328
H3K18ac and H3K9me3 ChIP‐seq in mESCs	Zuo et al (2022)	GEO: GSE169450
Primers (Oligonucleotides)
Primers for gRNA and shRNA	This paper Table EV1	NA
Primers for qPCR	This paper Table EV2	NA
Software and Algorithms
Image J	N/A	https://imagej.nih.gov/ij/; RRID: SCR_003070
GraphPad Prism	GraphPad software (V 8.4.2 (679))	https://www.graphpad.com/scientific-%20software/prism/; RRID: SCR_002798
NIS‐Elements	Nikon	https://www.nikoninNikonstruments.com/Products/Software
Integrated Genomics Viewer	Thorvaldsdottir et al (2013)	http://software.broadinstitute.org/software/igv/
R	R Core Team (2018)	https://www.r-project.org/
TrimGalore	N/A	https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/
Bowtie2 v2.1.0	Langmead & Salzberg (2012)	http://bowtie-bio.sourceforge.net/bowtie2/
PICARD tools	Li et al (2019b)	https://broadinstitute.github.io/picard/
HOMER v3.12		http://homer.ucsd.edu/homer/motif/
DeepTools	Ramirez et al (2016)	https://deeptools.readthedocs.io/en/develop/
MACS2 v2.1.1	Zhang et al (2008)	https://github.com/taoliu/MACS/wiki
DiffBind	N/A	https://bioconductor.org/packages/release/bioc/html/DiffBind.html
Samtools v0.1.19	Li et al (2019b)	https://github.com/samtools/samtools
FASTQC v0.11.2	N/A	http://www.bioinformatics.babraham.ac.uk/projects/fastqc/
BEDtools v2.17.0	Quinlan and Hall (2010)	http://bedtools.readthedocs.io/en/latest/
DESeq2	Love et al (2014)	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
Integrative Genomic Viewer	Thorvaldsdottir et al (2013)	http://software.broadinstitute.org/software/igv/
pheatmap		https://cran.r-project.org/web/packages/pheatmap/index.html
Rank ordering of super‐enhancers (ROSE)		https://bitbucket.org/young_computation/rose
Hisat2 v2.1.0	Pertea et al (2016)	https://daehwankimlab.github.io/hisat2/
cuffdiff v2.2.1	Trapnell et al (2012)	http://cufflinks.cbcb.umd.edu/
Seurat v3.2.0	Butler et al (2018)	https://satijalab.org/seurat/

Methods and Protocols

Cell culture

All the mouse ES cell lines were cultured in DMEM (GIBCO) supplemented with 15% FBS (ExCell Bio), 1% L‐Glutamine (Biological Industries), 1% MEM Non‐Essential Amino Acids (HyClone), 1% sodium pyruvate (Sigma), 1% penicillin/streptomycin (Biological Industries), 103 U/ml mLIF and 0.1 mM β‐mercaptoethanol. Cells were cultured at 37°C under 5% CO₂ in air.

HEK293T human embryonic kidney cells were cultured in DMEM (GIBCO) supplemented with 10% FBS (ExCell Bio) and 1% penicillin/streptomycin (Biological Industries).

Preparation of cell lines

E14Tg2a.IV (E14), Trim33 null, and GFP‐TRIM33 mESCs were from Xi et al (2011).

For Pml null cell line, E14 was transfected with PX458 and sgRNAs targeting Pml allele by using Lipofectamine 2000. Single colonies were selected by flow cytometry. After 2 weeks, the PML knockout clones were screened by genotyping PCR and Western blot analysis. Primers for gRNA and genotyping primers are listed in the STAR Method (Dataset EV1).

For PML‐expressing mESCs, the GFP‐PML or mCherry‐PML was cloned into the FUGW lentiviral construct and transfected with packaging vectors into 293T cells. Then the viruses were collected three times. And E14 cells were infected with viruses. The GFP‐PML or mCherry‐PML cells were selected by flow cytometry and then validated by fluorescence and Western blot.

For mESCs with GFP‐TRIM33 and mCherry‐PML, mCherry‐PML was cloned into the FUGW lentiviral construct and transfected with packaging vectors into 293T cells. Then the viruses were collected three times. And GFP‐TRIM33 mESCs were infected with viruses. The mESCs with GFP‐TRIM33 and mCherry‐PML were selected by flow cytometry and then validated by fluorescence.

For Pml null with GFP‐PML complemented mESCs, Pml KOs were infected with FUGW‐GFP‐PML lentivirus and selected by flow cytometry. The positive cells were validated by fluorescence.

Embryoid body (EB) differentiation

Embryoid body (EB) differentiation was carried out according to Xi et al (2011). Briefly, mESCs were digested to single cells by 0.25% trypsin. Then seeded 1.5 × 10⁶ cells for one 10 cm low adhesive petri dish (Nest, 752001) with 10 ml EB medium containing DMEM with 15% FBS, L‐Glutamine, MEM Non‐Essential Amino Acids, sodium pyruvate, penicillin/streptomycin and 0.1 mM β‐mercaptoethanol.

RNA extraction and RT–qPCR

Total RNA was isolated using Total RNA Purification Kit (DAKEWE, 8034111). Reverse transcription was carried out by RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, K1622). Real‐time PCR was performed by SYBR Green (GenStar, A311‐TZ10) on a Viia7 Real‐Time PCR system (Thermo Fisher Scientific). The relative RNA level was normalized by β‐actin. Primers for RT–qPCR are listed in the STAR Method (Table EV2).

Chromatin immunoprecipitation (ChIP) and CUT&TAG

ChIP was performed as previously described (Wang et al, 2017; Xi et al, 2011). CUT&TAG was performed with Hieff NGS Fast Tagment DNA Library Prep Kit for Illumina kit as manufacture protocol described. Illumina sequencing was performed after ChIP or CUT&TAG experiment.

Immunoprecipitation (IP) and western blot (WB)

Cells were collected with 94 g, 5 min and were washed three times with pre‐cooled PBS buffer. Cell pellets were lysed with whole cell lysis buffer (10 mM Tris–HCl pH8.0, 150 mM NaCl, 1% Triton X‐100, 5 mM EDTA) and were placed on ice for 30 min with vortex every 5 min. The cell lysates were collected by spin at 13,523 g 10 min. The supernatants were pre‐cleared with Protein A/G beads for 1.5 h at 4°C. Next, the supernatants were taken by spin at 845 g 3 min. GFP (ChromoTek) or Flag (Biotool) beads were added to the supernatants at 4°C overnight. The beads were washed by whole cell lysis 5 times and boiled with SDS loading buffer for 10 min.

The denatured protein samples were run on a 10% SDS–PAGE at 120 V and the SDS–PAGE was transferred to the PVDF membrane for 150 min at 300 mA. Then the membrane was blocked with 5% BSA in TBST (0.1% Tween‐20 in TBS) at room temperature for 1 h. The membrane was incubated with primary antibody at 4°C overnight and was washed four times with TBST. Membranes were incubated with the appropriate secondary antibodies conjugated with horseradish peroxidase for 1 h at room temperature. Enhanced chemiluminescence system (GE healthcare, RPN2108) was used for Western blot detection.

Alkaline phosphatase (AP) staining

Alkaline phosphatase staining was performed with BCIP/NBT Alkaline Phosphatase Color Development Kit (Beyotime, C3206). Briefly, cells were fixed with 1% PFA at room temperature for 30 min. Then, the alkaline phosphatase color developing working solution was added to the cells at room temperature without light for 10–30 min. The reaction can be terminated by removing the working solution and washed with ddH₂O.

Immunofluorescence (IF)

The cells were cultured on a glass bottom dish with 60% confluence and were fixed with 4% PFA for 30 min at room temperature. Then the cells were permeabilized with 1% Triton X100 for 10 min. Permeabilized cells were incubated with 1% BSA, in PBST (PBS and 0.1% Tween 20) for 60 min to block non‐specific binding of the antibodies, and then were incubated with diluted primary antibody in 1% BSA in PBST overnight at 4°C. in the next day, glass bottom dish was first washed with PBST for 3 times (5 min each), and then incubated with FITC labeled secondary antibodies (Invitrogen, F‐2765) for 60 min at RT. After PBST washing for 3 times (5 min each), glass dish was taken to confocal microscopy (Nikon A1) for imaging.

DNA FISH (fluorescence in situ hybridization) probe preparation and DNA FISH

DNA FISH probe was prepared by FISH Tag DNA Red Kit with Alexa Fluor 594 dye (Thermo Fisher, F32949) as manufacture protocol described. Briefly, nick translation was used to enzymatically incorporate an amine‐modified nucleotide into the probe template. Then, dye labeling of the purified amine‐modified DNA was achieved by incubation with amine‐reactive dyes. The purified probe was then ready for hybridization to the specimen.

For DNA FISH, the experiment was performed as previously described (Kurihara et al, 2020). Briefly, cells were cultured on the glass bottom dish at 60% confluent and were fixed with 4% PFA for 10 min at 37°C. Then the cells were treated with PBS containing 0.5 μg/ml RNase A and 0.5% Triton X‐100 for 10 min at room temperature. Next, the cells were treated with 2 N HCl for 2 min and were immediately washed out with PBS. Then the cells were re‐fixed with 1% PFA for 5 min. The cells were denatured by incubating with the hybrid buffer (2× Saline Sodium Citrate buffer (SSC), 50% formamide, 10% dextran sulfate, 1 mg/ml BSA, 1 mg/ml polyvinyl pyrrolidone (PVP), 0.01% Triton X‐100) with 10 ng/μl of DNA‐FISH probe at 80°C for 5 min and were transferred to 37°C heatblock overnight. After hybridization, the cells were washed with 2× SSC containing 0.01% Triton X‐100 at room temperature twice and with 0.2× SSC containing 0.01% Triton X‐100 at 50°C twice. The specimen was imaged by Nikon A1 confocal microscope.

3D reconstruction of PML NBs and GFP‐TRIM33 puncta

Cells for live‐cell imaging were cultured in confocal cell culture dish and stained with Hoechst (Beyotime, C1022). The fluorescent images were collected with Nikon A1RHD25 laser scanning confocal microscope (100× oil objective lens) and processed by NIS‐Elements AR 5.4 software. 405 and 488 nm channels were chosen to record Hoechst and GFP signal. Hoechst‐positive cells were selected for imaging. Z‐stack images are obtained with 0.3 μm per optical section. PML NBs and nuclei were reconstructed and quantitative analyzed by NIS‐Elements 3D tools. Images were processed with “BrightSpots detection” to set 3D binary of nuclei and PML NBs. For quantitative analysis, PML NBs located in nuclei were selected to calculate granule volume, diameter, and sphericity according to GFP signal and exported into excel data for calculation. Images from approximately 20 cells were taken for each group.

Fluorescence recovery after photobleaching (FRAP) measurements

In vivo FRAP experiments were carried out with Nikon A1 HD25 confocal microscope. Target granules were bleached with a 488 nm laser pulse. Recovery from photobleaching was recorded for 3–8 min. NIS‐Elements AR 5.4 software was used to analyze the fluorescence intensity of the bleached part. To calibrate possible dynamic position error in living cells, 3D tools were also included in statistical analyses. Fluorescence data were processed using Excel, GraphPad Prism. XY table and graph was used to show the recovery curve.

TurboID biotinylation and protein pull‐down

For biotinylation of PML‐TurboID proximity proteins, cells were incubated in medium with 200 μM biotin for 30 min, then rinsed with ice‐cold PBS for five times to quench the reaction. Cells were collected and lysed in chilled Buffer A (10 mM HEPES pH6.5, 1.5 mM MgCl₂, 10 mM KCl, 0.2% TritonX‐100, 0.5 mM DTT, protease inhibitor) for 10–30 min to release nuclei. Pelleted the nuclei by centrifugation at 1,700 g at 4°C for 5 min. Nuclei were resuspended in the lysis buffer (50 mM Tris pH8.0, 1 mM EDTA pH 8.0, 150 mM NaCl, 0.5% TritonX‐100, 0.5 mM DTT, protease inhibitor) and sonicated (60 W, 6 s on, 6 s off, 1 min) on ice. Nuclear lysates were centrifugated at 3,000 g for 10 min at 4°C to remove insoluble part, and nucleoplasm supernatant was transferred to a new tube for total protein quantification by BCA assay (Thermo‐fisher scientific, 23227, Pierce BCA protein assay kit).

Biotinylated proteins were captured by Streptavidin magnetic beads (Thermo‐fisher scientific, 88816). Pre‐equilibrated beads were incubated with nucleoplasm for 3 h in cold room to thoroughly bind target proteins. Enriched beads were washed twice with lysis buffer, once with 1 M KCl, once with 0.1 M Na₂CO₃, once with 2 M urea in 10 mM Tris–HCl (pH 8.0), and twice with lysis buffer. Enriched material was eluted from beads by boiling each sample in 30 μl of 2× protein loading buffer supplemented with 2 mM biotin and 20 mM DTT at 95°C for 10 min. Run input and eluate samples on a 4–20% SDS–PAGE gel (Genscript, M00655) for subsequent LC–MS/MS sequencing and data analysis.

Mass spectrometry data analysis

Protein–protein interaction network analysis was performed on STRING web server (https://cn.string‐db.org/, version 11.5), then exported into Cytoscape software (version 3.9.1). Proteins located in nucleus were selected by compartment filters. Node size was continuously mapped to Log₂ (TurboID/DMSO). Width of the edges were continuously mapped to the String score. Gene ontology analysis was performed using DAVID 6.8 web server (https://david.ncifcrf.gov/tools.jsp).

Gene expression analysis

The resulting FASTQ files were trimmed using TrimGalore (v0.5.0) and were mapped to the mouse reference genome (mm10) using HISAT2 (v 2.1.0). The levels of gene expression were calculated by featureCounts (v2.0.1) based on mm10 annotations. Differential testing and log2 fold change calculations were performed using DESeq2 (v 3.10) with the two biological replicates. We generated Volcano plots using R package ggplot2 showing the log transformed P values in the y‐axis and the log transformed fold‐change in the x‐axis. The gene set enrichment analysis (GSEA) of mESCs was performed as previously described (Subramanian et al, 2005; www.broadinstitute.org/software/gsea). We calculated the Normalized Enrichment Score (NES) and the false discovery rate (FDR) q‐value by permuting the gene‐set types. We assigned a cutoff of FDR ≤ 0.25 to identify significantly enriched gene sets.

ChIP‐seq and Hi‐C analysis

The resulting FASTQ files were trimmed using TrimGalore (v0.5.0) and then aligned to mm10 using Bowtie2 (v 2.3.3) (Langmead & Salzberg, 2012). PCR duplicates were removed using PicardTools. We chose MACS2 to call the peaks (FDR < 0.1). We subsequently generated bigwig files from the bam files using the bamCoverage function in deepTools (Ramirez et al, 2016). For visualization purposes, representative track diagrams were generated using the Integrated Genomics Viewer (IGV) software (Thorvaldsdottir et al, 2013).

Processed Hi‐C data in hic format was downloaded from 4DN data portal (Dekker et al, 2017) (https://data.4dnucleome.org, ID = 4DNFI3JYF9VS) and converted to cool format using hic2cool (https://github.com/4dn‐dcic/hic2cool), which was then inputted to hicFindTADs (Wolff et al, 2020) for identification of TADs. Track plot was drawn by pyGenomeTracks (Lopez‐Delisle et al, 2021; Lopez‐Delisle & Delisle, 2022) with input of Hi‐C cool file, TAD bed file, bw files of PML and TRIM33 ChIP‐seq, peak bed file called from PML ChIP‐seq data and GENCODE mm10 GTF file.

Author contributions

Hongyao Sun: Formal analysis; validation; methodology; writing—original draft. Yutong Chen: Methodology. Kun Yan: Formal analysis. Yanqiu Shao: Formal analysis. Qiangfeng C Zhang: Supervision. Yi Lin: Supervision; funding acquisition. Qiaoran Xi: Conceptualization; data curation; supervision; funding acquisition; writing—review and editing.

Disclosure and competing interests statement

The authors declare that they have no conflict of interest.

Supporting information

Expanded View Figures PDF

Table EV1

Table EV2

Movie EV1

Movie EV2

Movie EV3

Movie EV4

Movie EV5

Dataset EV1

Dataset EV2

PDF+

Source Data for Figure 1

Source Data for Figure 2

The EMBO Journal (2023) 42: e112058

Data availability

The RNA‐seq, ChIP‐seq, and CUT&TAG datasets have been deposited under Gene Expression Omnibus: GEO accession number GSE199738 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE199738). The TurboID‐PML mass spectrometry data have been deposited under Proteomics Identification database: PRIDE number PXD037432 (https://www.ebi.ac.uk/pride/archive/projects/PXD037432/private). We did not generate code in this study. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.