ERK1/2 signalling dynamics promote neural differentiation by regulating the polycomb repressive complex

Fibroblast Growth Factor (FGF) is a neural inducer in many vertebrate embryos, but how it regulates chromatin organization to coordinate the activation of neural genes is unclear. Moreover, for differentiation to progress FGF signalling has to decline. Why this signalling dynamic is required has not been determined. Here we show that dephosphorylation of the FGF effector kinase ERK1/2 rapidly increases chromatin accessibility at neural genes in mouse embryos and, using ATAC-seq in human embryonic stem cell derived spinal cord precursors, we demonstrate that this occurs across hundreds of neural genes. Importantly, while Erk1/2 inhibition induces precocious neural gene transcription, this step involves dissociation of the polycomb repressive complex from gene loci and takes places independently of subsequent loss of the repressive histone mark H3K27me3 and transcriptional onset. We find that loss of ERK1/2 activity but not its occupancy at neural genes is critical for this mechanism. Moreover, transient ERK1/2 inhibition is sufficient for polycomb protein dissociation and this is not reversed on resumption of ERK1/2 signalling. These data indicate that ERK1/2 signalling maintains polycomb repressive complexes at neural genes, that its decline coordinates their increased accessibility and that this is a directional molecular mechanism, which initiates the process of neural commitment. Furthermore, as the polycomb repressive complexes repress but also ready genes for transcription, these findings suggest that ERK1/2 promotion of these complexes is a rite of passage for subsequent differentiation.


Introduction 44
The identities of signals that induce particular cell fates are now well-established, but how such signalling 45 regulates chromatin to coordinate the transcription of differentiation genes and so orchestrate 46 engagement of a differentiation programme is not well understood. Fibroblast growth factor (FGF) 47 signalling has been implicated in the acquisition of neural cell fate in many vertebrate embryos [1][2][3][4][5][6][7]; 48 reviewed in [8] (although the timing of involvement varies between species and may reflect differences 49 in induction of anterior and posterior regions [6,9,10]). Intriguingly, while FGF is required for neural 50 induction in most of these contexts, its decline is also necessary for differentiation progression. 51 The requirement for transient FGF signalling to promote neural differentiation is particularly 52 evident in the elongating embryonic body axis in which the spinal cord is generated progressively, as here 53 there is a clear spatial separation of the temporal events of differentiation. FGF acts, along with Wnt 54 signalling, in the caudal lateral epiblast (CLE)/node streak border (which later form the tailbud) to 55 maintain a multipotent cell population known as neuromesodermal progenitors (NMP) which 56 progressively gives rise to the spinal cord and paraxial mesoderm ( Figure 1A) [11][12][13][14][15]

reviewed in 57
Henrique et al., 2015). Blocking FGF signalling in this cell population accelerates the onset of neural 58 differentiation genes and ectopic maintenance of FGF inhibits this step [16][17][18]. During normal 59 development differentiation onset is promoted by rising retinoid signalling, provided by adjacent paraxial 60 mesoderm, which represses Fgf8, restricting it to the tail end ( Figure 1A) [17][18][19][20] reviewed in [21]. These 61 findings indicate that decline in FGF signalling promotes neural differentiation, however, little is known 62 about the mechanism(s) by which such signalling dynamics mediate the coordinated activation of neural 63 differentiation genes. 64 Importantly, as NMP progeny leave the CLE and embark on neural differentiation they experience 65 loss of the FGF effector kinase ERK1/2/MAPK (hereafter referred to as ERK1/2) activity in both chick and 66 mouse embryos [22,23]. This suggests that ERK1/2 signalling dynamics may regulate onset of neural 67 differentiation. The in vitro manipulation of embryonic stem (ES) cells has provided some insight into the 68 To determine whether the PRCs regulate chromatin compaction around the Pax6 locus in vivo, we carried 143 out chromatin immunoprecipitations (ChIPs) for the PRC-mediated histone modification H3K27me3. 144 Mouse embryo microdissection was performed to enrich for the caudal region at E8.5 for ChIP-qPCR 145 ( Figure 1E   To test this, we developed an in vitro system to study neural differentiation during the generation of the 159 spinal cord. We reasoned that the use of human ESCs, with the slower progression to neural progenitor 160 cell identity, would provide a better temporal resolution of mechanisms mediating neural differentiation. 161 To this end, we directed the differentiation of human ESCs towards spinal cord [13,64] ( While these data suggested that PRC loss is required for the onset of neural gene transcription, it 199 remained unclear whether this regulation of chromatin accessibility is specific for PAX6 or reflects a 200 general mechanism that mediates onset of neural gene transcription. To address this we performed a 201 global analysis of the chromatin accessibility landscape in cells as they differentiated into NPs from the 202 NMPL cell state, using ATAC-seq [65]. This involved differentiation of NMPL (D3) cells (as in Figure 2A) 203 and sampling cell populations at intervals (D5, D6 and D8). Analysis of the chromatin configuration across 204 the PAX6 locus revealed increased accessibility at multiple sites, some as early as D5 and most prominent 205 by D6 ( Figure 3A), including one region described as an intragenic PAX6 enhancer region [66] (red box 206 Figure 3A). These early increases in chromatin accessibility correlated with decline in PRC protein 207 occupancy at the PAX6 locus detected by ChIP-qPCR during differentiation from D5 to D7 ( Figures 3B, B'), 208 which becomes significant by D8 ( Figures 2D, D'). In contrast with this apparently progressive decrease, 209 particularly in Jarid2 occupancy, H3K27me3 levels at the PAX6 TSS and along the gene body remain 210 unchanged between D5 and D7 ( Figure 3B') and drop dramatically only at D8 ( Figure 2D''). These data 211 align decreasing PRC occupancy with increasing chromatin accessibility, while acute loss of H3K27me3 212 appears a later step that coincides with robust PAX6 transcription. 213 cells revealed that although global accessibility levels remained relatively constant over time, a clear peak 215 in accessibility appeared at D6 of this differentiation ( Figure 3C). Detailed analysis comparing D3 and D8 216 identified 7877 regions with increased, and 11603 regions with decreased, accessibility (analysis 217 performed with diffReps with thresholds FDR ≤ 0.01 and log2FC ≥1). These regions of increased 218 accessibility correlate well with known active enhancer marks such as H3K4me1 and H3K27ac detected 219 in human embryonic spinal cord and brain (data from the ENCODE regulatory element database [67](see 220 Figure S1) and we termed these regions "neural sites". Analysis of these 7877 neural sites revealed that 221 they are associated with 4001 genes (Table S1) and that accessibility increased across the TSS and body 222 of these genes over time, with accessibility peaking at D6 ( Figure 3C') -this included additional chromatin 223 opening across genes already accessible at D3 as well as new genes with increased accessibility by D8 224 ( Figure 3D). 225 GO term analysis of the genes associated with regions of increased accessibility on days 5, 6 and 226 8 revealed terms related to neural development and showed a progression with more genes associated 227 with these terms over time ( Figure 3E, Table S1). Focussing on the GO term Neurogenesis we selected 2 228 key neural progenitor genes in addition to PAX6 and examined the ATAC-seq signal in detail around the 229 gene body. Chromatin accessibility around RARB and GLI3 showed an increase similar to that associated 230 with PAX6 as early as D5 ( Figures 3F, F'). Overall, these findings indicate that coordinated regulation of 231 chromatin accessibility is a general mechanism that mediates the onset of neural gene transcription, 232 while analysis of PRC occupancy dynamics at the exemplar PAX6 locus identifies this repression complex 233 as a target of such regulation. 234 In chick and mouse embryos, FGFR inhibition elicits precocious Pax6 transcription [16, 33] and FGFR and 253 ERK1/2 dephosphorylation (this study) decompact this gene locus. We therefore next tested whether 254 ERK1/2 signalling regulates polycomb protein occupancy and if such signalling globally directs chromatin 255 accessibility at our defined neural sites. To address this, we first tested whether inhibition of ERK1/2 256 signalling advances human neural differentiation in vitro; cells were cultured as in Figure 4A and exposed 257 to MEKi or DMSO cells during differentiation ( Figure 4B) and PAX6 transcript levels analysed by  This revealed precocious PAX6 transcription beginning at D5 and peaking at D6 in the presence of MEKi 259 ( Figure 4C). This correlated with changes in polycomb protein occupancy: MEKi exposure on D5 and D6 260 resulted in decreased Ring1B at the PAX6 and control locus HOXD11, and Jarid2 levels also showed a 261 declining trend (Figures 4D, D' and supplementary Figure S2). Furthermore, reduced H3K27me3 levels at 262 the PAX6 locus on D6 ( Figure 4D'') correlated with maximum precocious PAX6 transcription elicited by 263 inhibition of ERK1/2 signalling ( Figure 4C). 264 To determine whether ERK1/2 activity acts globally to regulate chromatin accessibility around 265 neural differentiation genes we performed ATAC-seq on cells at D5 and D6 exposed to MEKi or DMSO To interpret the genome-wide chromatin accessibility changes induced by MEKi, we analysed the 277 GO biological processes associated with these gene cohorts. Strikingly, MEKi treatment on D5 278 corresponded to an increased representation of genes associated with key neural development terms 279 ( Figure 4F). Moreover, at D6 of MEKi treatment, genes associated with later neural development, such 280 as synapse organisation, were specifically increased ( Figure 4F'). We therefore analysed the GO terms for 281 genes associated with more accessible regions that were unique to the MEKi data sets on D5 and D6 and 282 found a strong bias towards later neuronal differentiation (Table S3). This suggests that MEKi treatment 283 leads to increased global accessibility at neural differentiation genes and the accelerated progression to 284 To investigate the mechanism by which ERK1/2 signalling maintains PRC occupancy at differentiation 309 genes we asked whether ERK1/2 acts directly to mediate recruitment of polycomb machinery, as 310 reported in mES cells [56]. ChIP-qPCR was first used to determine ERK1/2 occupancy across PAX6 and 311 control genes during in vitro differentiation. In NMP-L cells, ERK1/2 was found at the PAX6 TSS and gene 312 body and the HOXD11 TSS and was also detected at the GAPDH TSS. Although ERK1/2 localisation at both 313 PAX6 and HOXD11 decreased in a manner consistent with increased accessibility of these loci during 314 differentiation, this also occurred at the locus of the constitutively transcribed gene GAPDH ( Figure 5A). 315 We next analysed whether ERK1/2 occupancy is dependent on its phosphorylation status by treating cells 316 with MEKi during differentiation (as in Figures 4A). No statistical difference was found between MEKi 317 treated cells and control (untreated or DMSO treated) on either D5 or D6 ( Figures 5B, B'). These data 318 reveal no consistent relationship between ERK1/2 chromatin association and gene transcription (ERK1/2 319 is found at the constitutively expressed GAPDH locus as well as PAX6 and other loci) during 320 differentiation. Moreover, the pattern of ERK1/2 occupancy is unaltered upon its de-phosphorylation 321 indicating that change in ERK1/2 activity, but not its chromatin association, regulates PRC occupancy. 322 These data raise the possibility that ERK1/2 activity regulates PRCs by supporting Jarid2 occupancy and 350 suggest that Jarid2 dissociation is an initial step in the disassembly of this complex, distinct from later 351 mechanism(s) regulating transcriptional onset of target genes. 352 353 Transient ERK1/2 dephosphorylation in NMP-L cells induces dissociation of both Jarid2 and Ring1b and 354 chromatin decompaction at the PAX6 locus, but does alter H3K27Me3 nor elicit transcription 355 To assess the consequences of loss of ERK1/2 activity at a later time point in NMP-L (D3) cells, these were 356 exposed to MEKi for 12 hours before conducting the same panel of ChIP-qPCR experiments ( Figure 6A). 357 At the end of this period ERK1/2 phosphorylation levels had now returned to control levels ( Figures

383
These data indicate that a transient reduction in ERK1/2 signalling is sufficient to trigger loss of 384 both PRC2 and PRC1 proteins. To determine the significance of such loss in this context, we further 385 assessed whether 12h MEKi exposure also altered chromatin accessibility around the PAX6 locus using 386 FISH. Significant decompaction of this region was found in MEKi treated cells compared with both DMSO 387 and untreated controls ( Figures 6E, E'). These findings correlate PRC loss with a distinct increase in 388 chromatin accessibility across this key neural progenitor gene and in showing that this step is not 389 reinstated on resumption of ERK1/2 activity, suggest that this is a directional molecular mechanism. 390 Moreover, these data demonstrate that removal of H3K27me3 is not required for such chromatin re-391 organisation and indicate that regulation of this histone modification is molecularly distinct from the 392 initial effects of ERK1/2 dephosphorylation, indeed this correlated well with later transcriptional onset 393

398
Histone modifications H3K4me3 and H3K27me3 are both detected (bivalent configuration) in the presence of FGF 399 and following dissociation of PRCs, and the PRC-mediated gene silencing mark H3K27me3 is lost only later 400 coincident with transcriptional onset: this indicates that this is a distinct regulatory step, which may depend on 401 retinoic acid (RA) signalling which is known to be required for PAX6 transcription. As transient loss of ERK1/2 activity 402 is sufficient to remove PRC2 and PRC1 and these complexes are not re-imposed on resumption of ERK1/2 signalling, 403 regulation of ERK1/2 activity and so PRC occupancy constitutes a directional molecular mechanism that 404 synchronises neural gene accessibility and promotes engagement of the neural differentiation programme.
In this study, we uncover the molecular mechanism by which the neural inducing signal FGF regulates 408 higher order chromatin organisation to orchestrate engagement of the neural differentiation 409 programme. We demonstrate that loss of the activity of FGF-effector kinase ERK1/2 leads to rapid 410 chromatin decompaction at the neural gene PAX6 in both the caudal lateral epiblast of mouse embryos 411 and in analogous hESC-derived NMP-L cells. Using ATAC-seq we show that this reflects a global action 412 involving increased chromatin accessibility across hundreds of neural genes. Focussing on PAX6 as an 413 exemplar gene, we find that ERK1/2 dephosphorylation specifically results in removal of polycomb 414 proteins, while ERK1/2 association with chromatin at this locus is unaltered. This demonstrates that 415 ERK1/2 activity and not occupancy is critical for this mechanism. We further show that while PAX6 is a 416 bivalent gene that is poised to be expressed in NMP-L cells, PRC protein loss following ERK1/2 inhibition 417 is distinct from later actions that mediate removal of the gene silencing mark H3K27me3 and 418 transcriptional onset. Importantly, transient ERK1/2 inhibition was sufficient to trigger PRC loss and 419 increased chromatin accessibility, indeed polycomb protein occupancy was not reinstated on resumption 420 of ERK1/2 signalling, indicating that this is a directional molecular mechanism that advances 421 differentiation. These findings suggest a model in which FGF/ERK1/2 signalling promotes or maintains 422 PRCs at neural gene loci while its subsequent decline during development initiates the neural programme 423 by synchronising removal of this repressive complex and so the increased accessibility of neural genes 424 Pax6 has recently been identified as a PRC target in this context too [72]. The rapid decompaction of 432 chromatin (within just one hour) following ERK1/2 inhibition in the mouse embryo shown here further 433 suggested that ERK1/2 activity may directly maintain PRC occupancy. Exposure of differentiating human 434 NMP-L cells to MEKi lead to loss of PRC1 component Ring1B, while PRC2 protein Jarid2 was not 435 significantly changed. This raises the possibility that PRC1 dissociates first and is the primary target of 436 ERK1/2 activity and PRC2 is lost secondarily. However, Jarid2 occupancy does show a marked downward 437 trend following ERK1/2 inhibition and the lack of significant difference may reflect low amounts of Jarid2 438 protein at chromatin or differences in antibody effectiveness. In support of the initial loss of PRC2, we 439 found that inhibition of ERK1/2 in NMP-L cells (which were maintained in FGF and Wnt agonist) for 3h 440 elicited a declining trend in Jarid2 occupancy at the PAX6 locus and that only after 12h were both Jarid2 increasing chromatin accessibility at many early neural genes. The extent to which such genes are direct 457 PRC targets or are regulated by other de-repressed neural genes remains to be determined, however, 458 their increased accessibility on D5, prior to PAX6 transcription, supports the possibility that they are also 459 PRC targets. 460 There are a number of ways in which ERK1/2 activity may influence PRC occupancy at neural 461 genes. Work in mES cells has shown that ERK1/2 occupies overlapping targets with polycomb proteins 462 and in particular Jarid2 and that abrogating ERK1/2 signalling reduces Jarid2 and H3K27me3 levels raising 463 the possibility that ERK1/2 protein is directly involved in PRC recruitment and/or maintenance [56]. 464 Although we detected ERK1/2 at the PAX6 locus and observed a decline in occupancy during 465 differentiation, we also found this protein at the house-keeping gene GAPDH, indicating that in this 466 context its association with DNA is not PRC target gene specific. Furthermore, inhibition of ERK1/2 activity 467 did not alter its association with chromatin at the PAX6 locus. This indicates that decrease in ERK1/2 468 activity and not ERK1/2 occupancy correlates with reduced PRC present at the PAX6 locus in The mouse Pax6 fosmid pair (WIBR-1 Mouse Fosmid Library, Whitehead Institute/MIT Center for 555 Genomic Research) and the human Pax6 BAC clone pair (WIBR-2 Human Library, see Table S3) were 556 prepared using a standard Mini-prep protocol. Using Nick transcription the fosmids and BAC clones were 557 labelled with Digoxigenin-11-dUTP and Biotin-16-dUTP. Unincorporated nucleotides were removed with 558 Quick Spin G50 Sephadex columns (Roche) and labelled probes quantified by dot blotting. 559 Embryos were exposed for short term to MEKi or vehicle control DMSO by 1h hanging drop culture, then 560 fixed (4% PFA, overnight), washed and dehydrated through a methanol series before being cleared in 561 xylene and embedded in paraffin for sectioning (7 μm). 562 containing neural tube or caudal lateral epiblast tissue were heated to 65°C for 30 min, then xylene 564 washed (4 x 10 min each) and re-hydrated through an ethanol series to dH20. The coverslips were 565 microwaved for 20 min in 0.1 M citrate buffer, pH 6.0, then left to cool and washed in dH20. For probe hybridisation, the same probe mix as for the FISH on mouse material was used (except probes 583 specific for the human Pax6 locus). The probes were denature on a heat block at 75°C for 3 min before 584 hybridisation (overnight, 37°C, humid chamber). The coverslips were then washed and probes were 585 detected using the same antibody reactions as for the FISH on mouse samples; nuclei were 586 counterstained with DAPI, coverslips were mounted with Slowfade Gold and imaged using a widefield 587 Deltavision Microscope. 588 Using OMERO insight regions of interest (ROIs) were selected over at least 3 z-sections. The 3D inter-589 probe distances in ROIs were measured using a custom script called OMERO mtools) [94], by segmenting 590 the objects from the background and calculating the distance between the centroids as d in μm. For ease 591 of comparison the ratio of number of base pairs per nm was calculated using the inter-probe distance 592 known in bp and measured in nm. 593 Chromatin immunoprecipitation (ChIP) 594 For each IP and control IP 25 µL of dynabeads coated with protein A (Invitrogen) were used. Beads were 595 washed 3 times in blocking buffer (0.1% BSA in PBS) and resuspended in blocking buffer containing the 596 antibody for immunoprecipitation or an equal amount of unspecific IgG (see Table S4) and incubated for 597 2 to 3h at 4°C. Beads were then washed twice in blocking buffer, resuspended in blocking buffer and 598 stored at 4°C until immunoprecipitation. 599 For ChIP on human ES cells, the cells were grown and differentiated in 10 cm culture dishes at 37°C and 600 5% CO2. Cells were fixed with formaldehyde (Sigma-Aldrich®, 1%, 10 min, RT) and quenched with glycine 601 Agilent Technologies). Each sample was analysed in 3 biological replicates and each biological replicate 636 was analysed in technical triplicates using primer pairs in Table S5. were first blocked in 3% BSA in TBST for at least 2h at RT, before incubation with primary antibody (see 659   Table S7) overnight. After washing in TBST (3 x 10 min, RT) the membrane was incubated with secondary 660 antibodies (see TableS7, 90 min, RT). Following TBST washes signals were visualised by scanning on LI-COR 661 Odyssey® system. Band intensity measurements were performed for quantification. The loading controls 662 were used to normalise for minor differences in loading before the phosphorylated ERK1/2 or PKB levels 663 were compared to the pan ERK1/2 and PKB levels of the same samples.            Semprich et al. Figure 6 relative transcription levels