https://orcid.org/0000-0001-5487-1117
Abstract
Histone H3 Lysine 9 (H3K9) methylation, a characteristic mark of heterochromatin, is progressively implemented during development to contribute to cell fate restriction as differentiation proceeds. Accordingly, in undifferentiated and pluripotent mouse Embryonic Stem (ES) cells the global levels of H3K9 methylation are rather low and increase only upon differentiation. How global H3K9 methylation levels are coupled with the loss of pluripotency remains largely unknown. Here, we identify SUV39H1, a major H3K9 di- and tri-methylase, as an indirect target of the pluripotency network of Transcription Factors (TFs). We find that pluripotency TFs, principally OCT4, activate the expression of Suv39h1as, an antisense long non-coding RNA to Suv39h1. In turn, Suv39h1as downregulates Suv39h1 transcription in cis via a mechanism involving the modulation of the chromatin status of the locus. The targeted deletion of the Suv39h1as promoter region triggers increased SUV39H1 expression and H3K9me2 and H3K9me3 levels, affecting all heterochromatic regions, particularly peri-centromeric major satellites and retrotransposons. This increase in heterochromatinization efficiency leads to accelerated and more efficient commitment into differentiation. We report, therefore, a simple genetic circuitry coupling the genetic control of pluripotency with the global efficiency of H3K9 methylation associated with a major cell fate restriction, the irreversible loss of pluripotency.
INTRODUCTION
During development, the establishment and maintenance of distinct gene expression patterns supporting the identity of each cell type are closely linked to the regulation of chromatin states (1). Two broad states have been clearly and unambiguously identified: euchromatin, associated with transcriptionally active regions, and heterochromatin, associated with gene repression (2–5). Two major states of heterochromatin have been traditionally considered. Facultative heterochromatin refers to a repressive chromatin environment displaying high variability across developmental stages, cell types and cell states. Indeed, silent developmental genes are usually embedded in facultative heterochromatin (3,4). In contrast, ubiquitously silent elements such as retrotransposons and pericentromeric regions are locked by constitutive heterochromatin (4,5). These two types of heterochromatin have been thought to be distinguishable by several molecular signatures, with facultative heterochromatin being characterized by trimethylation of histone H3 lysine 27 (H3K27me3) and constitutive heterochromatin by H3K9me3, among other chromatin features (2–5). Nevertheless, recent data has challenged these strict definitions (3). On the one hand, constitutive heterochromatin can under some circumstances be transcribed or decorated by marks previously associated with facultative heterochromatin (6–8). On the other, while H3K27me3 and H3K9me2 were considered as major repressive mark for developmental genes, an increasing body of evidence points to H3K9me3 as an additional mean to silence developmental regulators as their expression is definitely shut down in particular lineages (9). Hence, even though the role of H3K9 methylation in genome stability is unquestionable (10), its importance in gene regulatory mechanisms during development appears to be equally important. Indeed, mouse knock-out (KO) models of H3K9 histone methyltransferases (HMTs) display penetrant phenotypes, particularly during gastrulation when pluripotency is lost and major differentiation events take place (11,12). Conversely, before reaching pluripotency during early mouse embryogenesis, the levels of H3K9 methylation are strictly controlled; promoting their increase, for instance by overexpressing the HMT SUV39H1, leads to developmental defects at the compaction stage (13,14).
While extensive research has contributed to our understanding of how the establishment and maintenance of H3K27me3 regulates developmental transitions (1–10), how the levels of H3K9 methylation are developmentally regulated is less clear. Yet, a major distinction has been identified, particularly using pluripotent cells such as mouse Embryonic Stem (ES) cells. Indeed, H3K27me3 characterizes developmental genes even before differentiation, when they are embedded in the so-called bivalent chromatin, which is simultaneously enriched for H3K27me3 and for marks of activity (15). Upon differentiation, H3K27me3 is either consolidated or erased in a cell-type-dependent manner (16). On the contrary, H3K9 methylation is more largely controlled at the level of its abundance: during differentiation the global levels of H3K9me2 and H3K9me3 increase drastically (17,18). Conversely, during the induction of pluripotency in vitro through reprogramming processes, H3K9 methylation has been shown to act as a major epigenetic barrier that is in part overcome by globally reducing its levels (18,19). Therefore, while H3K27 methylation is mainly controlled by altering its genomic distribution, the global levels of H3K9 methylation display correlated changes to the differentiation status. Beyond the role of H3K9 methylation to stabilise somatic cell identities (20), how its global levels are seemingly coupled to the acquisition and loss of pluripotency, and what consequences this coupling has, remain open questions.
In this study, we aimed at understanding the molecular basis of the link between H3K9 methylation and pluripotency. We find Suv39h1 to be the only HMT tightly connected to the network of transcription factors (TFs) supporting pluripotency, particularly to its main player Oct4. The analysis of the mechanisms of Suv39h1 repression by OCT4 led us to identify Suv39h1as (21) as a Suv39h1-repressive antisense long non-coding RNA (lncRNA (22)) directly activated by OCT4. Using CRISPR-Cas9 mediated deletion of the antisense promoters, we further show that its activity controls the efficiency of H3K9 methylation in ES cells and the timing of commitment into differentiation. Thus, our work identifies a simple genetic network that provides a mechanistic perspective into how the global levels of H3K9 methylation are regulated at the onset of differentiation to irreversibly exit pluripotency.
MATERIALS AND METHODS
Cell lines and generation of A8 and D8 Suv39h1as mutant cells
WT cells in this study are E14Tg2a ES cells, from which all mutant cells were derived. Dox-inducible knock-out cells as well as OCT-AID cells have been previously described (Esrrb: EKOiE (23), Oct4: ZHBTC4 (24), Nanog: 44iN (25), OCT-AID (26)). To generate ES cells deleted for the Suv39h1as promoter, a CRISPR-Cas9 approach was followed using gRNAs available in Supplementary Table S5. Additional details are available in supplementary methods and in Supplementary Figure S4.
Regular cell culture and differentiation
Cells were cultured at 37°C, 7% CO2 on gelatine-coated plates in either FCS/LIF or in 2i/LIF, as indicated, and passaged every 2–3 days. Cells cultured in FCS/LIF were differentiated by withdrawing LIF for 3 days. N2B27 and EpiLC differentiation assays were performed with cells cultured in 2i/LIF for a minimum of 3 passages, after which LIF and 2i were withdrawn (for both N2B27 and EpiLC) and activin A, FGF and KSR supplemented (for EpiLC only). All details are available in supplementary methods.
Commitment assays
For commitment assays, 600 cells obtained every day of differentiation in N2B27 were plated in poly-l-ornithine/laminin-coated wells of a six-well plate, cultured for 7 days in 2i/LIF and stained for alkaline phosphatase activity (supplementary methods).
Immunostainings
To ensure direct comparisons, cell lines or conditions to be compared were individually labelled either with Rhodamine Red or with Deep Red dyes. The labelled cells were then collected, mixed at a 1:1 ratio and seeded onto poly-l-ornithine/laminin coated μ-slides. After 6 h of culture at 37°C and 7% CO2, they were prepared for immunostaining and imaged. Additional details are available in supplementary methods.
Single molecule FISH and DNA-FISH
Single-strand probes for Suv39h1 (47 oligos, 30 in exons, see sequences Supplementary Table S5) and Suv39h1as (35 exonic oligos, see sequences Supplementary Table S5) were used for smFISH and the position of each image recorded on the microscope. Subsequently, DNA-FISH was performed with a labelled fosmid (WIBR1-2188H11—from bacpac.chori.org) at the same positions (supplementary methods).
ChIP-seq and ATAC-seq
Both assays and the corresponding libraries were performed and generated as previously described (27), with the exception that fixed Drosophila chromatin was spiked in to be used as an internal normaliser, and sequenced for 75 cycles in a NextSeq 500 (SR for ChIP-seq and PE for ATAC-seq). For ChIP-seq, adapters with UMIs to enable distinguishing true identical reads from PCR duplicates (27) were used to improve mapping of repetitive DNA. After alignment (Bowtie2) to both mouse and Drosophila genomes, mouse peaks were identified with MACS2 for ATAC-seq and with a previously described approach (28) for H3K9me3. Read counts at these regions were normalised to the total number of reads aligning to the Drosophila genome, considering for ChIP-seq both the ChIP and corresponding input of each replicate for H3K9me3. For ChIP-seq, for reads mapping at multiple positions only one was randomly kept; for ATAC-seq, all multi-mappers were excluded. Global analysis of repetitive elements was performed with RepEnrich (29) followed by DESeq2 (30).
RESULTS
Suv39h1 expression is under the control of OCT4 in ES cells
Using immunofluorescence, we first confirmed that differentiation of ES cells by LIF withdrawal leads to an increase of both H3K9me2 and H3K9me3, an increase that can also be observed in spontaneously differentiating cells in regular ES cell cultures, which express low levels of the pluripotency TFs OCT4 or NANOG (Figure 1A and Supplementary Figure S1A, B). Therefore, we hypothesized that one or several histone methyl-transferases or lysine demethylases (HMTs and KDMs, respectively) (2,5) could be directly controlled by pluripotency TFs and differentially expressed upon differentiation, linking the loss of pluripotency to increased H3K9 methylation. To assess this, we monitored mRNA levels of HMTs and KDMs using published RNA-seq datasets of undifferentiated and differentiating ES cells (31) (Figure 1B, Supplementary Figure S1C, Supplementary Table S1). We found three HMTs to be upregulated upon differentiation: Suv39h1, Suv39h2 and Glp. In contrast, all tested KDM displayed minor changes below 2-fold (Supplementary Figure S1C). We reasoned that the increase of Suv39h1, Suv39h2 and Glp expression could either be due to a direct control of their transcription by pluripotency TFs or to alternative, indirect, mechanisms. To address this, we assessed the impact of the loss of individual pluripotency TFs (Oct4, Nanog and Esrrb) using dox-inducible knock-outs (23–26) (Supplementary Figure S1D). For NANOG, we used available datasets (31); for OCT4 and ESRRB they were generated for this study (Supplementary Table S1). Only one HMT, Suv39h1, was found upregulated 24h after inducing the loss of pluripotency TFs, particularly of OCT4, which depletion leads to a 2-fold increase in Suv39h1 mRNA levels (Figure 1B). Moreover, analysis of datasets from gastrulating mouse embryos (32) also showed Suv39h1 to be the only HMT to be upregulated upon differentiation of the three main germ layers (Supplementary Figure S1E). Hence, after confirming Suv39h1 expression changes by RT-qPCR (Supplementary Figure S1F), we hypothesized that OCT4 may act as a repressor of Suv39h1 expression to maintain low levels of H3K9 methylation until the onset of differentiation. Exploration of available ChIP-seq datasets (27) (Figure 1C) and direct validation by ChIP-qPCR (Supplementary Figure S1G) identified a hotspot of pluripotency TFs, including OCT4, in the vicinity of Suv39h1. However, this TF binding hotspot was found located 3′ to Suv39h1, at around 27 kb of its promoter region. Notably, we noticed that this region coincides with the promoter region of an uncharacterized gene, Gm14820 (AK010638), antisense to and largely overlapping Suv39h1 (top of Figure 1C). This antisense transcript, Suv39h1as, has been previously identified in oocytes and suggested to oppose to Suv39h1 expression at the oocyte to zygote transition (21).
Suv39h1 is downregulated by OCT4, which binds to the promoter of a Suv39h1 antisense lncRNA, Suv39h1as. (A) On the left, distribution of H3K9me2, H3K9me3, OCT4 and NANOG in undifferentiated (black; n = 4503, 3150, 3276 and 5615 cells, respectively) and differentiating (3d of LIF withdrawal – red; n = 6231, 3755, respectively) ES cell populations assessed by immunofluorescence. On the right, boxplot corresponding to different populations shown on the left: for H3K9me2/me3 distributions, the boxplots compare undifferentiated and differentiating cells (KS test P < 10–15); for OCT4 and NANOG distributions, the boxplots compare H3K9me3 between positive/negative subpopulations of OCT4 or NANOG (vertical line on the left panels; KS test P < 10–15 for both TFs). (B) Log2 fold change of the indicated gene after differentiating ES cells as in (A) or 24h after inducing the depletion of individual TFs, as indicated, using Dox-inducible knock-out cells. Each dot represents an independent replicate and the bar and error bars the corresponding means and standard errors. Suv39h1 upregulation upon differentiation or OCT4 depletion (P < 0.001) and Suv39h2 and Glp upregulation upon differentiation (P < 0.05) are statistically significant (t test). (C) Average binding profile of OCT4, SOX2, NANOG and ESRRB (reads per million) across the Suv39h1/Gm14820 locus (mm10, chrX:8 040 000–8 080 000). Suv39h1 and Gm14820/Suv39h1as are schematically represented on top. (D) RNA-seq profile across the Suv39h1/Suv39h1as locus, with forward and reverse fragment counts expressed with positive and negative values. (E) Schematic representation of Suv39h1as (red) and Suv39h1 (blue) isoforms as determined in Supplementary Figure S2. The coding probabilities calculated with CPAT and CPC2 algorithms are shown for the three isoforms of Suv39h1as. The vertical dashed lines in (C), (D) and (E) mark the position of Suv39h1as or Suv39h1 promoters.
Suv39h1as is an antisense long non-coding RNA
Stranded, total RNA-seq confirmed Suv39h1as to be expressed in ES cells, at levels comparable to Suv39h1 (Figure 1D). Using de novo transcript assembly with all the RNA-seq datasets presented in Supplementary Table S1, together with direct cDNA cloning, sequencing and RT-qPCR, we identified several isoforms expressed in ES cells (Figure 1E and Supplementary Figure S2A–C). All isoforms initiate from two distinct promoters, located in proximity to the region bound by pluripotency TFs, exhibit overlapping exons with Suv39h1 and terminate within Suv39h1 or in the vicinity of its 5′ end. Notably, Suv39h1as is annotated as a long non-coding RNA (lncRNA). Accordingly, using two different algorithms (CPAT (33) and CPC2 (34)), the nearly absent coding potential of all Suv39h1as isoforms was confirmed (Figure 1E). To further characterize Suv39h1as, we assessed the stability of its RNA products and found the half-life of its spliced and unspliced forms to be around 12h and 1h30, respectively (Figure 2A). However, Suv39h1as splicing is relatively inefficient compared to Suv39h1 or another protein coding gene, Nanog (Figure 2B), as is generally the case for lncRNA (22). Moreover, Suv39h1as was efficiently captured in poly-A selected RNA-seq, suggesting it is normally poly-adenylated (Supplementary Table S1). Next, we aimed at visualizing Suv39h1as RNA molecules in single cells. For this, we designed oligonucleotides targeting Suv39h1as exons and performed strand-specific single molecule RNA-FISH (smFISH) coupled to DNA-FISH to identify the Suv39h1as/Suv39h1 locus, using a fosmid covering the whole region (Figure 2C). We observed that Suv39h1as is mainly detected as a bright point in the nucleus, likely representing actively transcribed loci as it coincides with the DNA-FISH signal. A small number of single Suv39h1as RNA molecules could also be detected diffusing in the nucleus and, more rarely, in the cytoplasm. Quantification of the smFISH/DNA-FISH suggested a transcriptional frequency of around 50% in the population, with a median of six freely diffusing RNAs in cells presenting a transcriptionally active locus (Figure 2E). Hence, we conclude that the pluripotency TFs bind close to the two promoters of a Suv39h1 antisense lncRNA, which is mostly localised at its site of transcription, poly-adenylated and poorly spliced even though the spliced isoforms are relatively stable. Suv39h1as is, moreover, conserved in humans (Supplementary Figure S2D).
Suv39h1as is a nuclear, stable and lowly expressed lncRNA with anticorrelated expression dynamics to Suv39h1. (A) RT-qPCR analysis of the half-life of several RNA species during a transcription inhibition assay with Flavopiridol: Suv39h1 mRNA, using two trans-exonic primer pairs between exons 2 and 3 (blue – ex23) or exons 3 and 4 (purple – ex34); Suv39h1as, using two trans-exonic primer pairs between exons 1b and 2 (red – ex1b2) or exons 2 and 3 (orange – ex23) or primer pairs amplifying the unspliced RNA (magenta); Nanog mRNA (black). Each dot represents an independent replicate and the line the corresponding mean and standard error. Ribosomal RNA (28s) was used for normalization. (B) Histogram representing unspliced RNA levels relative to corresponding spliced RNAs for Nanog, Suv39h1 and Suv39h1as, as measured by RNA-seq. Each dot represents an independent replicate and the bar the corresponding mean. (C) Representative sm-FISH followed by DNA-FISH visualizing Suv39h1as RNA molecules and the Suv39h1/Suv39h1as locus, respectively, in undifferentiated WT cells. Red arrowheads indicate RNAs diffusing away from the locus, which is indicated by a yellow arrow. The proportion of actively transcribing cells is indicated (n = 358 cells). (D) Representative sm-FISH of Suv39h1as and Suv39h1 RNA molecules, followed by DNA-FISH visualising the Suv39h1/Suv39h1as locus in WT cells (n = 358). Selected loci transcribing either Suv39h1, Suv39h1as or both genes are indicated with arrow heads: blue, Suv39h1-only (30%); red, Suv39h1as-only (30%); yellow for cells transcribing both (20%). (E) Boxplots (median; 25–75% percentiles; error bars) showing the number of Suv39h1 diffusible molecules counted in cells presenting an active (TS+) or inactive (TS-) Suv39h1as gene (n = 358 WT cells). The increased of Suv39h1 diffusing molecules in Suv39h1_TS- versus Suv39h1as_TS + was assessed with KS test (P < 10–11). (F) Fold change expression of Suv39h1as (red) or Suv39h1 (blue) measured by RT-qPCR in differentiating WT cells versus undifferentiated controls (ctl). Differentiation was triggered for three days with three independent protocols: LIF withdrawal from FCS/LIF cultures (no LIF), 2i and LIF withdrawal from 2i/LIF cultures (N2B27) or EpiLC differentiation from 2i/LIF cultures (EpiLC). Values were normalized to Tbp and fold changes calculated to their respective control cultures. Each dot represents an independent replicate and the bar and error bars the corresponding means and standard errors. Gene expression differences were assessed against the respective undifferentiated controls (t test P < 0.05 for Suv39h1 and P < 0.01 for Suv39h1as for all differentiation assays). (G) Mean expression dynamics of Suv39h1 and Suv39h1as in published RNA-seq datasets of early mouse embryogenesis (left part, Zygote to 16-cell stage (35); right part, Morula, ICM and Epiblast (36)). (H) RT-qPCR analysis of Suv39h1as (red) and Suv39h1 (blue) expression upon OCT4 depletion in ZHBTC4 cells treated with Dox for the indicated time. Values were normalized to Tbp. Each dot represents an independent replicate and the line the corresponding mean with standard errors. Each time-point was compared to untreated cells with a t test (P < 0.001 for Suv39h1as and P < 0.05 for Suv39h1). (I) Analysis of publicly available (37) nuclear RNA-seq over the Suv39h1/Suv39h1as locus in untreated and OCT4-depleted ZHBTC4 cells (1 day of Dox treatment), presented as in Figure 1D.
Anticorrelated expression patterns of Suv39h1 and Suv39h1as
To investigate Suv39h1as andSuv39h1 expression patterns at the single cell level, we designed oligonucleotides across Suv39h1 exons and introns to monitor Suv39h1/Suv39h1as expression by smFISH in parallel to DNA-FISH (Figure 2D). We found around 20% of cells actively transcribing both sense/antisense genes and around 30% transcribing either one or the other. Moreover, cells actively transcribing Suv39h1as displayed significantly fewer Suv39h1 mRNA molecules (Figure 2E). Next, we differentiated ES cells using three independent protocols based on LIF withdrawal, N2B27 or EpiLC-directed differentiation. These three assays showed a strong reduction of Suv39h1as expression after 3 days of differentiation, when Suv39h1 expression increases (Figure 2F). Moreover, exploration of published RNA-seq during early embryogenesis (35,36) confirmed the anticorrelated expression patterns of Suv39h1/Suv39h1as during key events (Figure 2G): first, following fertilization (as previously suggested (21)), when Suv39h1as is highly expressed but decreases rapidly followed by Suv39h1 upregulation; second in the pluripotent ICM when Suv39h1as peaks at high levels, coinciding with a transient downregulation of Suv39h1. Hence, both ex vivo and in vivo, we observe anti-correlated expression dynamics of Suv39h1/Suv39h1as. Finally, to address whether Suv39h1as responds to OCT4 levels, we measured both RNA levels upon OCT4 depletion. We observed that the depletion of OCT4 leads to downregulation of Suv39h1as, reaching minimal levels of expression within 12 h and accompanied by a marked increase of Suv39h1 expression that reached maximal levels after 24 h (Figure 2H). Analysis of published nuclear RNA-seq (37) further confirmed this observation, underscoring a dramatic transcriptional silencing of Suv39h1as upon OCT4 depletion and a strong transcriptional induction of Suv39h1 (Figure 2I). Overall, Suv39h1 and Suv39h1as display anticorrelated transcription levels upon differentiation and during early embryogenesis. This anticorrelation stems from single cell dynamics where the transcription of the antisense is accompanied by a reduction of the transcriptional frequency of Suv39h1. Since Suv39h1as is downregulated upon differentiation and upon the loss of OCT4, our data suggests that pluripotent TFs activate Suv39h1as transcription which, in turn, downregulates Suv39h1 expression.
The Oct4-Suv39h1as-Suv39h1 circuitry
OCT4 depletion leads to Suv39h1as silencing within 12h (Figure 2H). However, to establish that this response is a primary effect mediated by OCT4 we aimed at analysing the effect of shorter OCT4 depletion. Using the same dox-inducible cells, we initially observed that Suv39h1as transcription assessed by unspliced RNA quantification starts to decrease as soon as 4 h after inducing OCT4 loss (Figure 3A, left). This prompted us to use a faster degradation system of OCT4, generated by fusion with an IAA-inducible degron (26). We observed a fast response of Suv39h1as, displaying a marked reduction of transcription as early as 2h after inducing OCT4 depletion (Figure 3A, right). In both systems, we observed that the loss of Suv39h1as transcription was followed by a nearly concomitant reduction of Suv39h1as mature RNA and an increase of Suv39h1 mRNA. Hence, these analyses establish that Suv39h1as transcription responds rapidly to the loss of OCT4. This strengthens the notion that Suv39h1as is a direct OCT4 target that downregulates Suv39h1 expression. To functionally establish the relationships between OCT4, Suv39h1as, Suv39h1 and H3K9 methylation, we designed two gRNAs to delete 5.5 kb encompassing the two promoters of Suv39h1as. Two independent KO clones, A8 and D8, were generated (Supplementary Figure S3). RT-qPCR showed a complete extinction of Suv39h1as expression (Figure 3B). We then addressed the impact of Suv39h1as depletion on Suv39h1 expression, both before and during differentiation. In the two mutant clones we observed an increase of Suv39h1 expression in undifferentiated cells, reaching the levels observed upon differentiation in wild-type (WT) cells (Figure 3B). In differentiating cells, when OCT4 binding at the Suv39h1as promoter is abrogated (Supplementary Figure S4A), and Suv39h1as naturally silenced, the deletion had no impact (Figure 3B), as expected. Analysis of several differentiation markers (Supplementary Figure S4B, left) ruled out the possibility that differences in Suv39h1 expression derive from indirect consequences linked to the differentiation status of the cells, as already suggested by the fast response of Suv39h1as transcription to OCT4 depletion (Figure 3A). Moreover, these results were independent of the differentiation protocol (Supplementary Figure S4C). Therefore, these results indicate that Suv39h1as acts as a pluripotency-associated repressor of Suv39h1 expression. To more directly address whether OCT4 represses Suv39h1 expression via Suv39h1as, we used siRNAs targeting Oct4 to test whether in the absence of Suv39h1as, the loss of OCT4 would lead to any modification of Suv39h1 expression. Whereas in WT cells the knock-down of Oct4 (above 80% efficiency, Supplementary Figure S4D) led to higher Suv39h1 expression, in mutant cells it was fully inconsequential (Figure 3C). Other OCT4-responsive genes, such as Cdx2 (24), displayed similar changes in WT and mutant cells upon Oct4 knock-down (Supplementary Figure S4B, middle panel). Thus, OCT4-dependent repression of Suv39h1 only occurs in the presence of the promoter region of Suv39h1as, suggesting that it requires Suv39h1as transcription. Next, we performed smFISH to study Suv39h1 upregulation with single cell resolution (Figure 3D). We observed a marked increase in the transcriptional frequency of Suv39h1, rising from 47.6% in WT to 76.3% and 74.1% in A8 and D8, respectively. Moreover, the number of Suv39h1 mRNAs per cell also increased substantially, with virtually no cell displaying an absence of Suv39h1 mRNAs (Figure 3E). This increase in Suv39h1 transcription in mutant cells was accompanied by higher levels of SUV39H1 protein levels, reaching those observed in WT differentiating cells (Figure 3F). Altogether, our results indicate that OCT4 directly activates Suv39h1as transcription, which in turn transcriptionally downregulates Suv39h1, leading to reduced mRNA and protein levels. This simple genetic circuitry ensures increased SUV39H1 expression during differentiation.
OCT4 represses Suv39h1 via Suv39h1as. (A) Log2 expression fold change of Suv39h1 (blue), Suv39h1as (red) and unspliced Suv39h1as (magenta) upon OCT4 depletion in ZHBTC4 (left) and OCT4-AID (right) cells. The X-axis shows the number of hours of treatment with Dox (ZHBTC4) and IAA (OCT4-AID). Two biological replicates per cell line and time-point are shown. The global effect of the treatments was assessed by comparing all time points together against untreated cells with a t test (P < 0.02 for both ZHBTC4 and OCT4-AID). (B) Expression fold change of Suv39h1as (top) and Suv39h1 (bottom) in WT and Suv39h1as-mutant cells (A8 and D8) cultured in undifferentiated or differentiating conditions (3 days without LIF). Values were normalized to Tbp. Each dot represents an independent replicate and the bar and error bars the corresponding means and standard errors. The increase of Suv39h1 in each clone versus WT cells was compared with a t test (p = 0.011 and 0.0016 for A8 and D8, respectively). (C) Expression fold change of Suv39h1as (top) and Suv39h1 (bottom) in WT and Suv39h1as-mutant cells (A8 and D8) knocked-down with either control or Oct4-targeted siRNAs. Values were normalized to Tbp. Each dot represents an independent replicate and the bar and error bars the corresponding means and standard errors. The increase of Suv39h1 expression upon OCT4 knock-down was assessed against control siRNAs with a t test (P = 0.027 in WT cells). (D) Representative sm-FISH images of Suv39h1 in A8 and D8 cells (WT cells presented in Figure 2D). Differences in the frequency of active transcription sites were assessed with a Chi2 test (P < 10–10 for both clones against WT cells). (E) Boxplots (median; 25–75% percentiles; error bars) showing the number of Suv39h1as (red) or Suv39h1 (blue) diffusible molecules counted in WT (n = 358) or mutant cells (A8, n = 289; D8, n = 270). Differences in the number of Suv39h1 molecules were assessed with a Mann-Whitney test (P < 10–15 for both clones against WT cells). (F) Representative Western-Blot of SUV39H1, LAMIN B1 and corresponding Ponceau for WT and mutant cells (A8 and D8) in undifferentiated and differentiating (3 days without LIF) conditions. On the left of each image is indicated the protein scale in kDa.
Suv39h1as modifies the chromatin of the Suv39h1as/Suv39h1 locus
We have observed that in mutant ES cells lacking Suv39h1as expression, the transcriptional frequency of Suv39h1 increases from 50 to 75% (Figure 3D). Similarly, upon OCT4 depletion Suv39h1 pre-mRNA increases substantially (Figure 2I) and rapidly (Figure 3A). Moreover, the absence of Suv39h1as is not accompanied by increased stability of Suv39h1 mRNAs (Supplementary Figure S5A). In contrast, in Suv39h1as mutant cells we observed increased chromatin accessibility of the Suv39h1 promoter region (Supplementary Figure S5B and Supplementary Table S2). Therefore, Suv39h1as is likely to act as a transcriptional repressor of Suv39h1. To explore this, and given that other antisense transcription units have been shown to modify the chromatin of their corresponding sense gene (38–40), we used a ChIP approach to establish the histone modification profile of the locus (Figure 4). First, we monitored H3K4 methylation profiles. We found H3K4me1/me2, which usually mark transcriptionally competent regions (41), to globally decorate the locus with minimal focal accumulation at promoters. Conversely, H3K4me3, a mark of activity (41), was focally enriched at the Suv39h1 promoter and displayed low levels over the antisense promoter. We then profiled the active histone acetylation marks H3K9ac and H3K27ac. Similarly to H3K4me3, we found H3K9ac to preferentially mark the Suv39h1 promoter. In contrast, both sense and antisense gene promoters where enriched for H3K27ac. In mutant cells, we observed a global decrease of H3K4me1/me2 over the region transcribed by Suv39h1as, particularly before it reaches the Suv39h1 gene body (Figure 4), indicating its transcription promotes the establishment of these marks. The lack of H3K4me1/me2 reduction within the region transcribed by both genes suggests that the increased transcription of Suv39h1 may have a compensatory role. Moreover, H3K4me2, H3K9ac and H3K27ac, all marks of gene activity, showed a slight but statistically significant increase at the Suv39h1 promoter in the absence of Suv39h1as (Figure 4). Altogether, this analysis suggests that the loss of Suv39h1as leads to increased Suv39h1 transcription at least in part mediated by increased euchromatinisation and accessibility of the Suv39h1 promoter. While undeniably small, the effects of the loss of Suv39h1as on the Suv39h1 promoter can be reproduced with different assays measuring chromatin activity.
Suv39h1as triggers complex chromatin changes across the locus. Chromatin immunoprecipitation profile of H3K4me1, H3K4me2, H3K4me3, H3K9ac and H3K27ac, as indicated, across Suv39h1/Suv39h1as locus in WT (black) and Suv39h1as-mutant cells (A8, yellow dots; D8, red dots; the red line represents the average of all data points for mutant clones). The X-axis represents genomic distances in kb with respect to the Suv39h1as transcription start site, as schematized on top. Note a break on the scale of the genomic coordinates at around X = 21kb. For the analysis of the locus-wide effects of the loss of Suv39h1as on H3K4me1 and me2, all values obtained with primer pairs located between coordinates +6 and +15 were considered (t test P < 0.001); for the effects measured at the promoter region, the position showing the highest difference for each histone mark was used (t test P < 0.05 for H3K4me2/me3, H3K9ac and H3K27ac).
Global increase of H3K9me2 and H3K9me3 in Suv39h1as mutant ES cells
The absence of Suv39h1as transcription in ES cells leads to increased SUV39H1 protein levels (Figure 3F), as confirmed by immuno-fluorescence: heterochromatic regions such as chromocenters prominently accumulate SUV39H1 in mutant cells (Figure 5A, Supplementary Figure S6A). Consequently, in both mutant clones we observed higher levels of H3K9me2 and H3K9me3 (Figure 5B, left panel), similar to those observed in WT differentiating cells (Supplementary Figure S6B), establishing a direct link between Suv39h1as and the global levels of H3K9 methylation in ES cells. According to the preferential enrichment of SUV39H1 at chromocenters, where H3K9me3-enriched peri-centromeric heterochromatin clusters (5), the increase of H3K9me3 in Suv39h1as mutant cells was higher than that of H3K9me2 (Figure 5B left panel). Hence, we aimed at more precisely characterise H3K9me3 in WT and Suv39h1as mutant cells by ChIP-seq. We identified 48,584 regions with more than 3-fold enrichment of H3K9me3 compared to the corresponding input in either WT or in at least one mutant clone (Supplementary Table S3). Comparative analysis of WT and mutant cells showed that the vast majority of these regions display higher H3K9me3 enrichment in the absence of Suv39h1as (Figure 5C), in magnitudes similar to those observed by immuno-fluorescence (Figure 5B, right panel). We also observed that more than 80% of H3K9me3-enriched regions overlap with retrotransposons (either LTRs, LINEs, or both; Figure 5D) and only minimally with cis regulatory elements (cRE in Figure 5D), as previously showed (7). Next, we quantified the effect of Suv39h1as KO at repetitive elements. We observed a strong bias towards a global and moderate increase of H3K9me3 over a multitude of families (Figure 5E, Supplementary Table S4), most notably at major satellites of peri-centromeric regions (red dot in the left panel of Figure 5E and Supplementary Figure S6C), but also at retrotransposons such as LINE L1s or LTRs (ERVKs and ERVLs; Figure 5E and Supplementary Figure S6D). We conclude that Suv39h1as mutant cells display increased efficiency to trigger heterochromatin in ES cells.
Global increase of H3K9 methylation in Suv39h1as mutant cells. (A) Representative SUV39H1 immunofluorescence of WT and Suv39h1-mutant ES cells (A8 and D8). (B) Violin and boxplots (median; 25–75% percentiles; error bars) of H3K9 methylation levels measured by immunofluorescence (left) or ChIP-seq (right). Immunofluorescence data shows relative mean intensity values of WT (black; n = 12881 for H3K9me2 and n = 12053 for H3K9me3) or mutant cells (A8, yellow, n = 3553 cells for H3K9me2 and 2081 for H3K9me3; D8, red, n = 5050 cells for H3K9me2 and 2641 for H3K9me3). KS tests were performed to compare WT and mutant cells (P < 10–8 for both marks and cell clones). ChIP-seq data shows log2 fold change of H3K9me3 over input, calculated after normalising both values to internal Drosphila spike in controls, for all the regions identified as H3K9me3-enriched in either WT or mutant clones. Differences between WT and mutant cells were evaluated with KS tests (P < 10–15 for both clones). (C) Scatter plot (X-axis: H3K9me3 levels in WT cells; Y:axis: mean H3K9me3 levels of the two mutant clones) corresponding to the data shown in (B). (D) Proportion of H3K9me3-enriched regions overlapping with both LINE and LTR retrotransposons or with only one of the two families or with cis-regulatory elements (55) (cRE) or with none of the above (Other). (E) Global analysis of repetitive elements (29), visualised as a volcano plot (X-axis: log2 fold change between mean H3K9me3 levels in mutant clones and WT cells; Y-axis: -log10(FDR); both calculated with DESeq2 (30) and considering Drosophila spike in controls. The first panel shows all repeat families, with major Satellites highlighted in red. The three following panels display selected families, as indicated, with red points highlighting elements with FDR < 0.01 and log2(FC) >0.2.
The lack of Suv39h1as leads to accelerated differentiation commitment
We finally wondered whether the increase of H3K9 methylation taking place in Suv39h1as mutant cells had any physiological impact. First, we used clonal assays to assess self-renewal and differentiation efficiency (Figure 6A, B). Either in conditions of reinforced self-renewal (2i/LIF), in traditional serum-containing culture medium (FCS/LIF) or in the absence of LIF (FCS), the number of alkaline-phosphatase colonies, a marker of pluripotent cells, was similar between WT and mutant clones. Hence, the presence of increased H3K9 methylation is largely inconsequential for self-renewal and for the loss of pluripotency. In agreement, both WT and mutant cells proliferate and differentiate normally, as evaluated morphologically (Supplementary Figure S7A) and by marker expression (Supplementary Figure S7B and Supplementary Figure S4B). However, during differentiation, the role of H3K9 methylation is to restrict cell fate and developmental competence (9,18,28,42), more than to elicit differentiation, with SUV39H1 playing a preponderant role in lineage-dependent maintenance of gene silencing in somatic cells (43). Therefore, we reasoned that the loss of Suv39h1as could modulate the timing of commitment into differentiation. To test this, we used an established assay (44) whereby WT and mutant clones were differentiated in parallel and, every day, the cells were harvested and reseeded clonally in 2i/LIF: only those cells that were not yet committed into irreversible differentiation can self-renew and form undifferentiated colonies (Figure 6C, D). As previously shown (44), we observed that commitment took place between days 2 and 3 in WT cells, with a reduction in clonogenicity of nearly 90% (Figure 6D). In mutant cells, however, the reduction in the number of undifferentiated colonies was more marked from day 2.5 onwards (Figure 6D). Therefore, the premature establishment of higher levels of H3K9me2/me3 in ES cells facilitates the irreversible commitment into differentiation, in line with the role of these repressive marks in locking cell fate changes (20).
Accelerated commitment into differentiation in the absence of Suv39h1as. (A) Representative alkaline-phosphatase staining of ES cell colonies cultured as indicated. (B) Number of WT or Suv39h1as-mutant (A8, orange points; D8, red points) colonies characterized as undifferentiated, mixed or differentiated after culturing them as indicated. Each dot represents an independent replicate and the bar and error bars the corresponding means (black for WT and red for the mean of all data points for mutant clones) and standard errors. (C) Alkaline-phosphatase staining of ES cell colonies cultured in 2i/LIF after 3 days in N2B27 for WT and Suv39h1as-mutant cells (A8). (D) Percentage of alkaline-phosphatase positive colonies cultured in 2i/LIF after differentiating them for the indicated number of days (X-axis) in N2B27. D0, undifferentiated cells were set as 100%. Each dot represents an independent replicate (WT, black; Suv39h1as-mutant clones in orange, A8, and red, D8) and the line the corresponding mean and standard error (all mutant data points were averaged to obtain the red line). Differences between WT and the mean of the two mutant clones were evaluated with Mann-Whitney tests using raw colony numbers (P = 0.0666, 0.0020 and 0.0041 for days 2.5, 3 and 4, respectively).
DISCUSSION
In this study, we have identified a genetic network linking the control of the global levels of H3K9 methylation to pluripotency. The pluripotency network, mainly through OCT4, activates Suv39h1as, an antisense lncRNA to the Suv39h1 gene; in turn, Suv39h1as represses Suv39h1 expression. Consequently, the level of H3K9 methylation is reduced. Upon differentiation, the collapse of the pluripotency network leads to the silencing of Suv39h1as, enabling increased SUV39H1 expression and H3K9 methylation, which affects the timing and efficiency of the irreversible commitment into differentiation. Given the lack of strong effects at a small number of defined regulators of differentiation in Suv39h1as mutant cells, our data suggest that their faster commitment into differentiation is achieved by globally ameliorating the efficiency of H3K9 methylation, as illustrated by higher levels of H3K9me3 at peri-centromeric major satellites and at retrotransposons. During exit from pluripotency, when proper epigenetic silencing is implemented (45), the enhancement of heterochromatinization may enable the long-term changes of gene expression required to acquire new cell identities. Moreover, it is noteworthy that the Oct4-Suv39h1as-Suv39h1 genetic axis may also act as a time-delay generator, enabling ES cells to filter out transient and short fluctuations of the pluripotency network (46): only a long decrease in activity of pluripotency TFs, such as that occurring during the exit from pluripotency, may be sufficient to elicit the increase in SUV39H1 expression that will follow the extinction of Suv39h1as.
Antisense lncRNAs are frequent in mammals, with 29% of canonical protein coding genes displaying antisense transcription (47). Given their antisense orientation and the resulting complementarity, antisense lncRNAs can theoretically regulate their cis-linked sense gene through a wide variety of mechanisms. By deleting the Suv39h1as promoter region, we found that Suv39h1as controls the transcriptional frequency of Suv39h1, whose promoter becomes more accessible and enriched in euchromatin marks. Additionally, Suv39h1as mutants display no changes of Suv39h1 mRNA stability. Moreover, OCT4 depletion leads to a fast silencing of Suv39h1as, accompanied by a transcriptional induction of Suv39h1 that takes place almost concomitantly to the decrease of mature Suv39h1as RNA. Altogether, this data supports the notion of a transcriptional control of the Suv39h1 promoter by Suv39h1as, either by the act of Suv39h1as transcription itself or through mechanisms involving the unspliced Suv39h1as RNA. Notably, this ambiguity characterises other well-known pairs of sense/antisense genes, such as Xist/Tsix (38–40). Like the loss of Tsix, that of Suv39h1as results in complex chromatin changes, which suggest that Suv39h1as triggers H3K4me1 and me2 throughout the locus and, at the same time, reduces the enrichment for active histone marks and the accessibility of the Suv39h1 promoter. Even though the reminiscence of these effects to those triggered by Tsix are striking, possibly revealing a general property of antisense transcription, we cannot exclude that they are indirect consequences of a transcriptional induction of Suv39h1 mediated by other mechanisms. For instance, a direct competition between Suv39h1/Suv39h1as and other gene promoters for shared enhancers could play an important role (48). In agreement, exploration of the topology of the extended Suv39h1/Suv39h1as locus suggests that the 3D organization is compatible with the involvement of 3D events (Supplementary Figure S8A); nevertheless, neither chromatin accessibility (Supplementary Figure S8B,C), nor expression analyses of genes previously identified (49) as sharing at least a common enhancer with Suv39h1/Suv39h1as (Supplementary Figure S8D), display any change in our mutant cells. This indicates that Suv39h1as has a very local and specific role within the locus: to control the promoter activity of Suv39h1. Nevertheless, other additional mechanisms cannot be excluded, such as direct collisions of the sense/antisense polymerases (50) or post-transcriptional mechanisms involving splicing (51).
The deletion of Suv39h1as promoter and the ensuing increase in H3K9 methylation appears to be largely tolerated by ES cells: they self-renew and differentiate efficiently. This observation is in line with others, where histone modifiers have been either invalidated or ectopically expressed in ES cells with minor consequences for self-renewal (52,53). However, despite the fact that Suv39h1as mutant cells self-renew and differentiate normally, we asked whether the timing of commitment into differentiation is altered. Our results showed a faster and more efficient commitment into differentiation, suggesting that the global levels of H3K9 methylation contribute to irreversibly lock the loss of pluripotency. This observation adds to the notion of H3K9 methylation in general, and SUV39H1 in particular, acting as an epigenetic barrier providing robustness to cell fate changes (9,18,28,42,43). Moreover, our results also underscore the dominance of pluripotency TFs over chromatin modifications (53). We had already shown that NANOG, another key pluripotency TF, controls H3K27me3 levels, particularly during early differentiation (31). Here, we complement this notion with OCT4 controlling H3K9me3 via the Suv39h1as/Suv39h1 tandem. Together, these results place the control of the global levels of heterochromatin marks under the activity of the pluripotency network, extending the concept of the genetic dominance of pluripotency. Whether our observations and conclusions apply to early mouse embryogenesis and to the acquisition and loss of pluripotency in vivo is now a question of primary importance. Notably, H3K9 methylation levels are exquisitely regulated during early embryogenesis (54). It is noteworthy that SUV39H1 is absent in oocytes and its expression starts at the 2–4 cell transition stage (35,36), when the reconfiguration of constitutive heterochromatin as chromocenters is initiated. Moreover, the overexpression of SUV39H1 during early embryogenesis leads to developmental defects (13,14): its zygotic overexpression leads to a developmental arrest during compaction. Since Suv39h1as displays an exquisite anticorrelated expression to Suv39h1 upon fertilization (as initially suggested (21) and our data reanalysis has confirmed), whether Suv39h1as holds Suv39h1 expression until the appropriate time to enable the timely establishing of the first heterochromatic structures in the embryo, represents an important question for the future.
DATA AVAILABILITY
All genome-wide datasets generated for this study are available in GEO (GSE184140): poly-A RNA-seq of EKOiE (GSM5578482 to GSM5578485) and ZHBTC4 cells (GSM5578486 to GSM5578489); total RNA-seq of E14Tg2a (GSM5578490 to GSM5578491); H3K9me3 ChIP-seq (GSM6211351 to GSM6211362); ATAC-seq (GSM6211363 to GSM6211368). Other available datasets in GEO were used: from GSE118898–GSM3350412 to GSM3350417 (44iN), GSM3350424, GSM3350426, GSM3350428 (undifferentiated cells), GSM3350429, GSM3350431, GSM3350433 (differentiating cells); from GSE87822 – GSM2341322 to GSM2341327 (nuclear RNA-seq in ZHBTC4); GSE121708 (single-cell RNA-seq of E4.5 and E7.5 embryos using processed data from: ftp://ftp.ebi.ac.uk/pub/databases/scnmt_gastrulation); GSE45719 (zygote to 16C early embryo single-cell RNA-seq)ṁ—GSM1112490 to GSM1112581, GSM1112603 to GSM1112610, GSM1112656 to GSM1112663, GSM1112694 to GSM1112705, GSM1112766 to GSM1112769. Datasets available from ArrayExpres (EMBL-EBI) were also used: E-MTAB-2958 (RNA-seq of E2.5 to E4.5 mouse embryos).
SUPPLEMENTARY DATA
Supplementary Data are available at NAR Online.
ACKNOWLEDGEMENTS
This study was conceived by P.N with inputs from L.B. and V.H. Experiments were designed and executed by L.B., with help from A.D., V.H., A.T., S.V.-P. and M.C.-T. RNA-seq analyses and assembly were done by N.G., N.O., A.C. and I.U. ChIP-seq was performed by V.F. and ATAC-seq by I.G; A.C. analysed them both. Auxin-inducible depletion experiments were done by L.E.B. and J.C.R.S. The paper was written by L.B. and P.N. P.N. acknowledges Elphège Nora for discussions on chromatin topology and all referees for constructive criticisms; in particular two referees that alerted the authors that Suv39h1as had already being identified by the Knowles lab in 2006.
FUNDING
L.B. acknowledges the Ecole Normale Supérieure and Sorbonne Université for funding; P.N. and M.C.-T. acknowledge the Labex Revive [Investissement d’Avenir; ANR-10-LABX-73]; Institut Pasteur; CNRS. Funding for open access charge: Pasteur core funding.
Conflict of interest statement. None declared.
Notes
Present address: Noa Gil. Friedrich Miescher Institute for Biomedical Research, Basel, Switzerland.
Present address: Nick Owens, Institute of Biomedical and Clinical Science, University of Exeter Medical School, Exeter, UK.
Comments