Enhancer decommissioning by LSD1 during embryonic stem cell differentiation

Abstract

Transcription factors and chromatin modifiers are important in the programming and reprogramming of cellular states during development. Transcription factors bind to enhancer elements and recruit coactivators and chromatin-modifying enzymes to facilitate transcription initiation. During differentiation a subset of these enhancers must be silenced, but the mechanisms underlying enhancer silencing are poorly understood. Here we show that the histone demethylase lysine-specific demethylase 1 (LSD1; ref. 5), which demethylates histone H3 on Lys 4 or Lys 9 (H3K4/K9), is essential in decommissioning enhancers during the differentiation of mouse embryonic stem cells (ESCs). LSD1 occupies enhancers of active genes that are critical for control of the state of ESCs. However, LSD1 is not essential for the maintenance of ESC identity. Instead, ESCs lacking LSD1 activity fail to differentiate fully, and ESC-specific enhancers fail to undergo the histone demethylation events associated with differentiation. At active enhancers, LSD1 is a component of the NuRD (nucleosome remodelling and histone deacetylase) complex, which contains additional subunits that are necessary for ESC differentiation. We propose that the LSD1-NuRD complex decommissions enhancers of the pluripotency program during differentiation, which is essential for the complete shutdown of the ESC gene expression program and the transition to new cell states.

Figure 1. LSD1 is associated with enhancer and core promoter regions of active genes in ESCs

a, LSD1 occupies a substantial population of actively transcribed genes in murine ESCs. The pie charts depict active (green), bivalent (yellow) and silent (red) genes, and the proportion (black lines) occupied by LSD1, Pol II or the Polycomb protein Suz12 (Supplementary Table 1 and Supplementary Information). The numbers of genes bound and the total number of genes in each of the active, bivalent and silent classes are shown. LSD1 ChIP-Seq data are from combined biological replicates using an antibody specific for LSD1 as determined by knockdown experiments (Supplementary Fig. 1). The P value for each category was determined by a hypergeometric test. b, LSD1 occupies enhancers and core promoter regions of actively transcribed genes. Shown are ChIP-Seq binding profiles (reads per million) for ESC transcription factors (Oct4, Sox2, Nanog), coactivator (Med1), chromatin regulator (LSD1), the transcriptional apparatus (Pol II, TBP) and histone modifications (H3K4me1, H3K4me3, H3K79me2, H3K36me3) at the Oct4 (Pou5f1) and Lefty1 loci in ESCs, with the y-axis floor set to 1. Gene models and previously described enhancer regions^– are shown below the binding profiles. c, LSD1 occupies enhancer sites. A density map is shown of ChIP-Seq data at Oct4, Sox2, Nanog and Med1 co-occupied enhancer regions. Data are shown for an ESC transcription factor (Oct4), coactivators (Med1 and p300) and a chromatin regulator (LSD1) in ESCs. Enhancers were defined as Oct4, Sox2, Nanog and Mediator co-occupied regions. More than 96% of the 3,838 high-confidence enhancers were co-occupied by LSD1 (P < 10⁻⁹). Colour scale indicates ChIP-seq signal in reads per million. d, LSD1 occupies core promoter sites. Shown is a density map of ChIP-Seq data at transcriptional start sites (TSSs) of genes neighbouring the 3,838 previously defined enhancers (c). Data are shown for components of the transcription apparatus (Pol II and TBP) and the chromatin regulator LSD1 in ESCs. Core promoters were defined as the closest TSS from each enhancer. Colour scale indicates ChIP-Seq signal in reads per million.

Figure 2. LSD1 inhibition results in incomplete silencing of ESC genes during differentiation

a, Schematic representation of trophectoderm differentiation assay using the doxycycline-inducible Oct4 shutdown murine ESC line ZHBTc4. Treatment with doxycycline for 48 h leads to depletion of Oct4 and early trophectoderm specification. Cells were treated with dimethylsulphoxide (DMSO; control) or the LSD1 inhibitor TCP for 6 h before 2 μg ml⁻¹ doxycycline was added for a further 24 or 48 h. b, Treatment of ZHBTc4 ESCs with doxycycline leads to loss of Oct4 proteins. Oct4 and LSD1 protein levels in nuclear extracts determined by western blotting (WB) before and after treatment of ZHBTc4 ESCs with 2 μg ml⁻¹ doxycycline. Tubulin served as loading control. c, Doxcycline (Dox)-treated cells treated with TCP maintained SSEA-1 cell surface marker expression. Cells were stained for DNA (Hoechst; Hoe), Oct4 and SSEA-1. Scale bar, 100 μm. d, Expression of selected ESC and trophectodermal genes 48 h after Oct4 depletion in dimethylsulphoxide-treated and TCP-treated cells (black and grey bars, respectively). Treatment with TCP partly relieved repression of ESC genes but did not affect upregulation of trophectodermal genes. Error bars show s.d. from biological replicates. e, Genes neighbouring LSD1-occupied enhancers are less downregulated during ESC differentiation after TCP treatment. Shown is the mean fold change in expression of the 630 downregulated (at least 1.25-fold; P < 0.01) genes nearest LSD1-occupied enhancers (Fig. 1c) during differentiation of TCP-treated and untreated control cells. Alleviation of repression is significantly higher (asterisk, P < 0.005) for LSD1 enhancer-bound repressed genes than for all repressed genes.

Figure 3. LSD1 is associated with a NuRD complex at active enhancers in ESCs

a, NuRD components occupy enhancers and core promoter regions of actively transcribed genes. Shown are ChIP-Seq binding profiles (reads per million) for transcription factors (Oct4, Sox2, Nanog), coactivator (Med1) and chromatin regulators (LSD1, Mi-2β, HDAC1, HDAC2), at the Oct4 (Pou5f1) and Lefty1 loci in ESCs, with the y-axis floor set to 1. Gene models and previously described enhancer regions^– are depicted below the binding profiles. b, LSD1 is associated with NuRD components Mi-2β, HDAC1 and HDAC2, as well as with CoREST. LSD1 and HDAC1 are detected by western blotting (WB) after immunoprecipitation of crosslinked whole cell extract (WCE) with anti-LSD1, anti-HDAC1, anti-HDAC2, anti-Mi-2β or anti-CoREST antibodies. IgG is shown as a control. c, LSD1 and HDAC1 are detected by western blotting after immunoprecipitation of uncrosslinked nuclear extracts (NE) using anti-LSD1, anti-HDAC1, anti-HDAC2, anti-Mi-2β or anti-CoREST antibodies. IgG is shown as a control. d, The occupancy of enhancers by NuRD proteins (Mi-2β, HDAC1 and HDAC2) is significantly greater than the occupancy by CoREST or REST (P < 10⁻⁹). The height of the bars represents the percentage of the 3,838 enhancers co-occupied by LSD1, NuRD proteins (Mi-2β, and either HDAC1 or HDAC2), CoREST and REST.

Figure 4. LSD1 is required for H3K4me1 removal at ESC enhancers

a, H3K4me1 levels are decreased at LSD1-occupied enhancers during ESC differentiation, and this effect is partly blocked on treatment with TCP. Dox, doxycycline. b, Treatment with TCP does not affect the increase in H3K4me1 levels at trophectodermal genes during differentiation. Shown are ChIP-Seq binding profiles (reads per million) for Oct4 and LSD1 at the Lefty1 and Gata2 loci in ESCs. Below these profiles, histone H3K4me1 levels are shown for ZHBTc4 control ESCs, cells treated with doxycycline for 48 h to repress Oct4 and induce differentiation (ESCs + Dox), and ESCs treated with doxycycline and TCP (ESCs + Dox + TCP). For appropriate normalization, ChIP-Seq data for histone H3K4me1 is shown as rank normalized reads per million with the y-axis floor set to 1 (Supplementary Information). Gene models and previously described enhancer regions^, are depicted below the binding profiles. c, Sum of the normalized H3K4me1 density ±250 nucleotides surrounding LSD1-occupied enhancer regions before and during trophectoderm differentiation in the presence or absence of TCP. The associated genes were identified on the basis of their proximity to the LSD1-occupied enhancers. d, Sum of the normalized H3K4me1 density ±250 nucleotides surrounding 1,722 LSD1-occupied enhancers before and during differentiation in the presence or absence of TCP. Of the 2,755 LSD1-occupied enhancers with decreased levels of H3K4me1 on differentiation, 63% (1,722) had higher H3K4me1 levels after TCP treatment (P < 10⁻¹⁶). e, Heat map displaying the sum of the normalized H3K4me1 density ±250 nucleotides surrounding the 1,722 LSD1-occupied enhancers that retained H3K4me1 in TCP-treated ESCs compared with untreated control differentiating ESCs. Colour scale indicates ChIP-Seq signal in normalized reads per million.