H3K9 tri-methylation at Nanog times differentiation commitment and enables the acquisition of primitive endoderm fate

ABSTRACT Mouse embryonic stem cells have an inherent propensity to explore gene regulatory states associated with either self-renewal or differentiation. This property depends on ERK, which downregulates pluripotency genes such as Nanog. Here, we aimed at identifying repressive histone modifications that would mark Nanog for inactivation in response to ERK activity. We found that the transcription factor ZFP57, which binds methylated DNA to nucleate heterochromatin, is recruited upstream of Nanog, within a region enriched for histone H3 lysine 9 tri-methylation (H3K9me3). Whereas before differentiation H3K9me3 at Nanog depends on ERK, in somatic cells it becomes independent of ERK. Moreover, the loss of H3K9me3 at Nanog, induced by deleting the region or by knocking out DNA methyltransferases or Zfp57, is associated with reduced heterogeneity of NANOG, delayed commitment into differentiation and impaired ability to acquire a primitive endoderm fate. Hence, a network axis centred on DNA methylation, ZFP57 and H3K9me3 links Nanog regulation to ERK activity for the timely establishment of new cell identities. We suggest that establishment of irreversible H3K9me3 at specific master regulators allows the acquisition of particular cell fates during differentiation.

2C. The vertical line shows the median of differentiated E14Tg2a cells. Note that the vast majority of ∆K9 cells present higher NANOG levels than E14Tg2a cells after 7 days of differentiation (KS test p<2.2e-16). Development: doi:10.1242/dev.201074: Supplementary information   Representative bright-field photomicrographs of wild-type (E14Tg2a) and ∆K9 cells during their treatment with a primitive endoderm differentiation protocol. The arrows point to clusters of primitive endoderm cells. Development: doi:10.1242/dev.201074: Supplementary information

Embryoid Bodies differentiation.
Embryoid Bodies (EBs) imaged in Fig.S5A (top) were obtained with a hanging-drop protocol where a suspension of 0.1M cells/mL was distributed in 20 µl drops (2000 cells/drop) onto inverted cover plates of several dishes filled with PBS 1X to avoid evaporation. The day after, the aggregates formed in each drop were pooled and seeded on non-cell culture treated dishes and cultured in 10%FCS-DMEM without LIF for 7 additional days. EB differentiation for outgrowths (Fig.S5A, bottom) and RNAseq analysis (Fig.S5B) was performed by seeding 0.04M cells/cm 2 in 10%FCS-DMEM/LIF on 0.1% gelatinized cell culture treated dishes. After 3 days, the aggregates were harvested (2min trypsinization), transferred to non-cell culture treated dishes for 4 days in 10%FCS-DMEM medium without LIF and then replated at very low density on 0,1% gelatinized cell culture treated dishes for 2 and 4 additionnal days in 10% FCS-DMEM medium without LIF.

Commitment assays.
Cells cultured in 2i+LIF were plated at clonal density (600 cells per well of 6-well plates) and subject to 2i_OFF differentiation, as described above. After each day, the medium was changed into 2i+LIF for 7 days, after which cells were fixed and stained for alkaline phosphatase activity as described.

Preparation of mitotic cells.
To obtain mitotic ES cells (>95% purity as assessed by DAPI staining and microscopy), we used a nocodazole shake-off approach, as described before (Festuccia et al., 2018).

Imaging analyses.
Bright field microscopy. Cell culture pictures were taken on a Nikon Eclipse Ti-S inverted microscope equipped with: CFI S Plan Fluor ELWD ×20 objective; 89 North PhotoFluor LM-75; Hamamatsu ORCA-Flash 4.0LT camera; NIS Elements 4.3 software.
cImmunofluorescence of NANOG/OCT4 in E14Tg2a cells. Cells were trypsinized, counted and fixed for 10 min in 4% formaldehyde (Sigma Cat#F8775) at room temperature and quenched immediately after with Glycine 125 mM for 5 min at room temperature. After one wash in PBS 1X, they were cytospun at 0.5M cells/200 µl/spot (4 min -300 rpm -low speed) on Superfrost+ slides. After washing the cells twice in PBS 1X, they were permeabilized with cold PBS 1X/0.5 % v/v Triton X-100 for 5 min, washed twice with cold PBS 1X, and blocked in PBS1X/1% Donkey Serum (DS) (Sigma, Cat# D9663) for 30 min on ice. Cells were incubated overnight at 4°C with primary antibodies (diluted in PBS1X/1% DS) within a humid chamber. After three washes in cold PBS1X, cells were incubated 1H at room temperature in the dark with secondary antibodies (diluted in PBS1X/1%DS), washed three times in PBS1X and nuclei counterstained with Vectashield antifade mounting medium with DAPI (Vectorlabs, Cat#H-1200). Imaging was performed with an inverted Nikon Eclipse X microscope equipped with: X20/0.45 (WD 8.2-6.9) objective; LUMENCOR excitation diodes; Hamamatsu ORCA-Flash 4.0LT camera; NIS Elements 4.3 software. Cell Profiler (Carpenter AE et al., 2006) was used for quantifications and ggplot2 (Wickham et al. 2016) for plotting in R.

c3/ Comparative immunofluorescence of NANOG (E14Tg2a vs mutant cells DK9, TKO or ZKO).
WT and each mutant cell line were trypsinized, counted and resuspended at 1M/ml in FCS free medium (DMEM-Glutamax/100 mM 2-mercaptoethanol/NEAA 1X) into sterile 1.5 mL Eppendorf tubes. Cells were then individually incubated either with 10 µM Rhodamine Red dye (InVitrogen, Cat#CMTPX C34552) or 1µM Deep Red dye (InVitrogen , Cat#C34565) for 30 min at 37°C. The labeled cells were then collected by centrifugation, washed with PBS1X, resuspended in DMEM/10%FCS+LIF medium and mixed at a 1:1 ratio for WT Rhod and mutant DeepRed cells (usually ~0.4M each). The opposite labeling (WT DeepRed & mutant Rhod ) was also performed with identical results. Mixed cells were seeded into a Poly-L-Ornithine/Laminin coated single well of a µ-slide 4 well Ph+ ibiTreat (Ibidi GmbH Ref#80446) and incubated for ~6H at 37°C and 7% CO2. Cells were then fixed directly into the well by freshly prepared PFA 4% (Fisher Scientific, Cat# 16431755) for 10 min at room temperature in the dark, washed twice in PBS1X for 10 min and used immediately for immunostaining or stored at +4°C for short time. Cells were washed in PBS-Tw0.1% (PBS1X/0.1 % Tween20 (Sigma, Cat#P9416)) and permeabilised with PBS1X/0.1% Triton X-100 (Sigma, Cat#T8787) for 10 min at room temperature. After three washes with PBS-Tw0.1%, cells were blocked with PBS-Tw0.1%/10% Donkey Serum (Sigma, Cat#D9663) for 30 min on ice in the dark and incubated overnight with the primary antibody (diluted in PBS-Tw0.1%/10% DS). Following three washes with PBS-Tw0.1%, 1H incubation with secondary antibodies at room temperature in the dark and 2 washes with PBS1X, nuclei were counterstained with 4',6-diamidino-2-phenylindole (DAPI; Sigma, Cat# D9542), washed in PBS1X and use immediately for imaging or stored at +4°C for short time. Imaging was performed with an inverted Nikon Eclipse X microscope equipped with: X20/0.45 (WD 8.2-6.9) objective; LUMENCOR excitation diodes; Hamamatsu ORCA-Flash 4.0LT camera; NIS Elements 4.3 software. Quantifications were performed using Cell Profiler (Carpenter AE et al., 2006). For each experiment, E14Tg2a and each mutant cells quantifications were attributed using the FlowJo software. For each experiment, the fluorescence intensity of each mutant cell was normalised to the median of the corresponding E14Tg2a intensities imaged on the same spot. The data was ploted using the ggplot2 package (Wickham, 2016) in R.

Immunofluorescence of NANOG and primitive endoderm markers.
Differentiated cells were processed into their respective well: one wash with PBS1X, fixation with freshly prepared PFA 4% for 10 min at room temperature, three washes of PBS1X, and used immediately for immunostaining or stored at +4°C for short time. Immunostaining was performed as in c3/ for NANOG in combination with primitive endoderm markers. Imaging was performed with an inverted Nikon Eclipse X microscope equipped with a x20 objective; LUMENCOR excitation diodes; Hamamatsu ORCA-Flash 4.0LT camera; NIS Elements 4.3 software. For each well, ~30 pictures were taken by randomly scanning the well. Quantifications were performed using Cell Profiler (Carpenter AE et al., 2006) and ploted using the ggplot2 package (Wickham et al. 2016) in R. Representative images were generated with ImageJ software using identical settings for E14Tg2a and each mutant.

Single-molecule RNA Fluorescent In Situ Hybridisation (smFISH).
Cells were grown at low density, put in no RED medium for 24H before collecting them by regular trypsinization and resuspended in DMEM/10% FCS medium. Cells were fixed with 3.7% Formaldehyde (Sigma F8775) for 10 min at RT and the reaction was stopped by addition of 0.125M glycine (SIGMA G7126) for 5 min at RT. Cells were then centrifugated at 1000 rpm at 4°C, washed once in cold PBS/0.1% BSA, resuspended in cold PBS/0.1% BSA/1 mM EDTA at 1 million cells/mL and cytospun at 500 rpm (low speed) for 4 min on SuperFrost slides (Thermo J1800AMNT) -around 500K cells/spot in order to obtain around 40-60 nuclei/field (obj.x60) and no or few clumps. Slides were air dried for 1 min and stored in 70% EtOH at 4 °C for at least one night. The slides are dehydrated for 3 min in 100% EtOH and fully air dried. Each spot was hybridized for 24H at 37°C in a denaturing chamber (2X SSC/10% formamide) with the Nanog mRNA probe cocktail (2X SSC, 10% Formamide, BSA 1 µg/mL, 1 µL of E.Coli RNAs at 1 µg/mL, 1 µL of probe at 0.25 µM). The slides were washed for 1H at 37 °C in 2X SSC/10% Formamide then 2 times in 2X SSC for 10 min and mounted in Vectashield medium with DAPI (Vector-abcys H-1200). The Nanog mRNA probe was designed using Stellaris Probe Designer version 4.2 on Biosearch Technologies website with the maximum masking level (5) and was synthetized by the same company. Image stacks (0.5 μm gap) were acquired using a Nikon Eclipse X microscope equipped with: ×63 oil immersion objective (N.A1.4); LUMENCOR excitation diodes; Hamamatsu ORCA-Flash 4.0LT camera; NIS Elements 4.3 software. The analysis was performed using ImageJ to identify smFISH spots using the Find Maxima function with adapted Noise tolerance and maximal projections of the stacks. The identified maxima were converted into grays scale images and, together with minimal projections of the DAPI staining, were imported into Cell Profiler (Carpenter et al., 2006) to count the number of smFISH spots per DAPI-segmented cell. The results were imported into R to compute the frequency of spot detection per cell and plot them with ggplot2 (Wickham, 2016).

Chromatin analyses.
Chromatin preparation. After trypsinisation, 10 7 ES cells were crosslinked for 10 min in 3 ml DMEM/10% FCS/1% formaldehyde (Sigma Cat#F8775). Crosslinking was stopped with 125 mM glycine for 5 min at room temperature then 5min in ice. Cells were pelleted and washed with ice-cold PBS1X. Cells were resuspended in 1 ml of ice-cold swelling buffer (25 mM Hepes pH 7.95, 10 mM KCl, 10 mM EDTA) freshly supplemented with 1× protease inhibitor cocktail (PIC-Roche, Cat# 04 693 116 001) and 0.5% IGEPAL (Sigma, Cat#I8896). After 20 min on ice, the suspension was passed 50 times in a dounce homogenizer. Cells were then centrifuged and resuspended in 1 ml of ice-cold D3 (0.1% SDS, 15 mM Tris pH 7.6, 1 mM EDTA) buffer, freshly supplemented with 1× PIC. Samples were sonicated using a Covaris M220 -Setpoint @6°C and 10 cycles with the following parameters for each cycle : 60 sec duration, peak Development: doi:10.1242/dev.201074: Supplementary information power of 67W, duty factor of 15% and cycles/burst of 500. A delay of 45 sec is added at the end of each cycle. Result of the average power is 10W. After centrifugation (15 min, 14000 rpm, 4 °C), the supernatant was stored at −80 °C until use. 20 µl were used to quantify the chromatin concentration and check DNA size (typically 200-600 bp).

Immunoprecipitation.
15 to 20 µg of chromatin were used for each ChIP after pre-clearing it for 1.5 hours rotating on-wheel at 4 °C in 1 ml of TSE150 (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl pH8, 150 mM NaCl) buffer containing 50 μl of protein G Sepharose beads (Active Motif, Cat#37499) 50% slurry, previously blocked with BSA (0.5 mg/ml ; Roche, Cat# 10711454001) and yeast tRNA (1 µg/ml ; Roche Cat# 10109495001). Immunoprecipitations were performed overnight rotating on-wheel at 4 °C in 500 μl of TSE150. 20 μl were set apart for input DNA extraction and precipitation. 50 μl of blocked protein G beads 50% slurry was added for 2 h rotating on-wheel at 4 °C. Beads were pelleted and washed for 5 min rotating on-wheel at room temperature with 1 ml of buffer in the following order: 2 × TSE150, 1 × TSE500 (as TSE150 but 500 mM NaCl), 1× washing buffer (10 mM Tris-HCl pH8, 0.25M LiCl, 0.5% IGEPAL, 0.5% Na-deoxycholate, 1 mM EDTA), and 2 × TE (10 mM Tris-HCl pH8, 1 mM EDTA). Elution was performed in 100 μl of elution buffer (1% SDS, 10 mM EDTA, 50 mM Tris-HCl pH 8) for 15 min at 65 °C after vigorous vortexing. Eluates were collected after centrifugation and beads rinsed in 150 μl of TE-1%SDS. After centrifugation, the supernatant was pooled with the corresponding first eluate. For both immunoprecipitated and input chromatin, the crosslinking was reversed overnight at 65 °C, followed by proteinase K treatment, phenol/chloroform extraction and ethanol precipitation. qPCR analysis. Input and IP samples were analysed by real-time quantitative PCR performed in duplicates in 384-well plates with a LightCycler 480 (Roche) using 4.6 µl of LightCycler 480 SYBR Green I Master (Roche, 04707516001), 5 µl of sample and 0.2 µl of each primer at 20 µM in a final reaction volume of 10 µl. Standard and melting curves were generated to verify the amplification efficiency (>85%) and the production of single DNA species. PCR primer sequences are listed in Table S2. The 2dCt method was used. All values were corrected to the input and ploted in R using ggplot2 (Wickham et al., 2016).
Antibodies. Antibodies used for all ChIP experiments are listed in Table S2.

Gene expression analyses.
RNA preparation, RT-qPCR and sequencing. E14Tg2a and ∆K9 cells were differentiated in parallel as embryoid bodies and cells recovered at d0, d4, d6, d8. RNA extraction and DNase treatment of 3 independent assays were made with NucleoSpin RNA Mini kit (Macherey Nagel, Cat# 740955.50) according to the manufacturer's protocol. Reverse Transcription was performed with 1 µg of total RNAs with random hexamers following manufacturer's protocol (Roche 04379012001); qPCR was performed as described above (Tbp was used as a reporter; see Table S2 for primer sequences). Stranded, poly-A selected RNA-seq libraries were prepared and sequenced (paired-end 150bp reads; around 50 millions each) by Novogene Co Ltd.

Alignments and quantification.
Stranded paired end RNA-seq reads were aligned to the mm10 genome using STAR (Dobin et al., 2013) and quantified by RSEM (Li and Dewey, 2011), with additional options " --calc-pme --calc-ci --estimaterspd --forward-prob 0.0 --paired-end". Transcripts per million (TPM) were computed after omitting outliers identified by Principal Component Analysis (30 or more standard deviations away from the mean for PC1 component). Development: doi:10.1242/dev.201074: Supplementary information Principal component analysis. PCA was run in R (prcomp function; with option center=TRUE) using log2 transformed TPM after selecting genes with at least 1 TPM in at least one sample mean (n=3). The top 1000 PC1 loadings (55% of variance) were used for gene ontology analyses. Data visualisation was made in base R.
Differentially expressed genes. For all differential expression tests RSEM estimated read counts per sample were rounded for use with DESeq2 (Love et al., 2014), which was run without independent filtering. For differentially expressed genes in undifferentiated ES cells, we selected those with an FDR < 0.05 in ∆K9.1 versus E14Tg2a and ∆K9.2 versus E14Tg2a and same fold-change direction with abs(log2FC)>0.3. For embryoid body differentiation, we considered genes with absolute log2(FC) > 1 and FDR < 0.05 at any day of the differentiation versus undifferentiated cells for either E14Tg2a or ∆K9 cells. The corresponding heatmap was made in R with ComplexHeatmaps package (Gu et al., 2016).
Clustering of differentially expressed genes and developmental annotation. K-means clustering was computed with R using the function kmeans with options k=6, nstart=50, iter.max=50. Only differentially expressed genes as identified during embryoid body differentiation were used (z scored mean TPM). The number of clusters was chosen as the minimal value identifying at least one cluster with maximal expression at each day of differentiation, including d0. Correlations to developmental gene expression were made by directly ploting the log2(FC) reported in a previous study using SCNMT-seq around gastrulation on mouse embryos (Argelaguet et al., 2019). The data was downloaded from ftp://ftp.ebi.ac.uk/pub/databases/ scnmt_gastrulation. The heatmap was made in R with ComplexHeatmaps package (Gu et al., 2016) and the boxplot with ggplot2 (Wickham, 2016). Enrichment analyses. Gene Ontology analyses were performed using Enrichr (https://maayanlab.cloud/Enrichr/) and the top association was selected. To determine enrichments of each group of differentially expressed genes in proximity to TF binding sites for Nanog and Oct4/Sox2 (Festuccia et al., 2019), we calculated Fisher tests right tail p-values for the association between differentially expressed genes of each cluster within xbp of a binding site to a background of all genes clustered within xbp of a binding site, for x in [1, 1e+9] bp. Data visualization was made in base R.  Development: doi:10.1242/dev.201074: Supplementary information