Embryonic stem cell–derived hemangioblasts remain epigenetically plastic and require PRC1 to prevent neural gene expression

Tractable mouse embryonic stem (ES) cells in which differentiation pathways can be accurately modeled in vitro and in vivo has revolutionized our understanding of lineage specification¹  and provided critical information on the linear relationships between precursor cells and their progeny.^2-4  Using a well-characterized model of hematopoietic differentiation,^3-5  we have investigated stage-specific changes in chromatin at specific genes in ES cells upon mesoderm induction. Undifferentiated ES cells contain hyperdynamic chromatin⁶  where much of the genome is transcribed at a low level⁷  and many developmental regulator genes are transcriptionally primed and “bivalently marked” with histone modifications associated with both gene activation (such as trimethylation of lysine 4 of histone 3 [H3K4me3]) and polycomb-mediated repression (such as trimethylation of lysine 27 of histone 3 [H3K27me3]).^8-10  This marking is thought to reflect the lineage plasticity of ES cells because it is resolved on differentiation (eg, in neural precursors^10,11 ) and regained upon reprogramming of differentiated cells to induced pluripotent stem cells.¹²  Initial comparisons between ES, hematopoietic stem cells, and lymphocytes showed reduced chromatin accessibility and loss of bivalent marking in hematopoietic cells⁸  but did not evaluate whether such changes occurred before, coincident with, or after cells become functionally restricted. By examining the chromatin status of a panel of genes in intermediate stages of differentiation, we show here that several lineage-inappropriate (neural-specifying) genes remain bivalent and transcriptionally poised in committed hemangioblasts because the removal of polycomb repressive complex 1 (PRC1) activity results in overt gene derepression.

Bry-GFP⁵  and Ring1a^−/−/Ring1b^fl/fl/Cre-ER_T2¹³  ES cells were cultured and differentiated as described.⁵  Embryoid bodies were dissociated, stained using a phycoerythrin-conjugated anti-Flk1 antibody (BD Biosciences) and cells sorted on GFP (Bra/T) and Flk1 expression by fluorescence-activated cell sorting (FACS). Hematopoietic potential was tested using colony assays as described in supplemental Methods (available on the Blood Web site; see the Supplemental Materials link at the top of the online article).

Replication timing and gene expression were analyzed using previously published methods.¹⁴  Embryoid bodies were BrdU pulse-labeled before dissociation and sorting. Chromatin from Bra/T-GFP⁺/Flk1⁺ cells was immunoprecipitated as described in supplemental Methods.

Bry-GFP ES cells that carry a GFP reporter under the control of the mesoderm-associated Brachyury (Bra/T) gene^3,5  were differentiated into embryoid bodies and cells representing epiblast (GFP⁻/Flk1⁻, day 2.5), early mesoderm (GFP⁺/Flk1⁻, day 3.75), and hemangioblast stages (GFP⁺/Flk1⁺, day 3.75) of differentiation were isolated by FACS, as previously described⁵  (Figure 1A-B) and verified by expression analysis. Expression of pluripotency-associated factors declined (Oct4/Pou5f1) or was lost (Rex1/Zfp42, Sox2) on differentiation, whereas mesoderm-associated genes (such as Bra/T, Flk1/Kd, and Ikaros/Ikzf5) were induced, and genes important for neuroectoderm specification were detected only at low levels (Math1/Atoh1, Nkx2-2, Nkx2-9, and Sox1)^15-18  similar to undifferentiated ES cells (Figure 1B; supplemental Figure 1A). Importantly, epiblast-stage cells readily formed colonies when replated in ES media (170/1000 cells), but this capacity was not retained by cells expressing Bra/T-GFP (Figure 1C), consistent with a loss of pluripotency and commitment to mesoderm at this stage.

As accessible chromatin domains tend to replicate earlier in S-phase than heterochromatic condensed domains,¹⁹  we used replication timing analysis to assess dynamic chromatin changes accompanying differentiation.^8,20-22  Temporal replication of ES-associated (Rex1 and Sox2), mesoderm-associated (Bra/T, Flk1, Ikaros, and Myf5), and neuroectoderm-associated (Math1, Sox1, Nkx2-2, and Nkx2-9) genes was evaluated at different stages of hemangioblast differentiation (Figure 1D; summarized in supplemental Figure 1B) using established methods and controls.¹⁴  As anticipated, Rex1 and Sox2 replication was earlier in ES cells (black lines) than hemangioblasts (red lines), consistent with reduced chromatin accessibility and gene expression upon differentiation, whereas replication of constitutively accessible (α-Globin) or heterochromatic (X141) controls remained unaffected. Three mesoderm-associated genes showed slight (Bra/T, Ikaros) or significantly (Flk1) advanced replication in response to differentiation, whereas neural-specifying genes either retained early replication profiles throughout (Nkx2-2, Nkx2-9, and Sox1) or, in the case of Math1, were delayed progressively upon differentiation (Figure 1D; supplemental Figure 1B). These data showed that, although mesoderm commitment results in changed epigenetic profiles, several genes encoding neural specifiers remained early replicating in hemangioblasts in contrast to lymphocytes and hematopoietic stem cell lines (FDCPmix A4 and Pax5^−/− pro-B) characterized previously⁸  (summarized in supplemental Figure 1B).

To directly determine whether these promoters were bivalently marked in hemangioblasts, we performed chromatin immunoprecipitation (ChIP) on purified Bra/T-GFP+/Flk1⁺ samples, using antibodies specific for H3K4me3, H3K27me3, and total H3. Previous reports had shown that the promoters of Bra/T, Flk1, Ikaros, Math1, Sox1, Nkx2-2, and Nkx2-9 (but not Myf5) are bivalent in mouse ES cells.¹⁰  In hemangioblasts, we detected H3K4me3 at the promoters of expressed genes (Bra/T, Flk1, green, Figure 2A), whereas H3K4me3/H3K27me3 was codetected at Ikaros, Sox2, Math1, Sox1, Nkx2-2, and Nkx2-9. Similar bivalent marking of these genes was observed in CD41⁺-enriched hematopoietic progenitor populations (data not shown). Verification that the neural-associated genes Sox2, Math1, Sox1, Nkx2-2, and Nkx2-9 contained bivalently marked chromatin was shown by sequential ChIP⁹  in which H3K4me3-containing fragments were enriched by a second round of precipitation with antibodies to H3K27me3 (Figure 2B). Furthermore, locus-wide profiling of Nkx2-2, a gene-poor region on mouse chromosome 2 that has been extensively characterized as a prototype “bivalent domain” in ES cells,^9,23  showed that the distributions of H3K4me3 and H3K27me3 were largely preserved in Bra/T-GFP⁺/Flk1⁺ hemangioblasts (Figure 2C).

Bivalent marking of developmental regulator genes in ES cells is associated with a primed transcriptional state in which RNA polymerase II is bound, phosphorylated at Serine 5, but prevented from elongating by polycomb repressive complexes; removal of PRC1 or PRC2 results in inappropriate gene up-regulation (derepression).^8,23  Abundant PRC1 transcripts (Ring1a, Ring1b, Mel18, and Bmi1) and Ring1b protein was detected in both undifferentiated ES cells and cells at successive stages of hemangioblast differentiation (supplemental Figure 3A-B) and ChIP analysis showed Ring1b binding at Math1, Sox1, Nkx2-2, and Nkx2-9 promoters in Bra/T-GFP⁺Flk1⁺ cells (Figure 2D). To ask whether these genes were indeed functionally primed in mesoderm-committed hemangioblasts, we used Ring1a-deficient ES cells where PRC1 activity can be withdrawn on tamoxifen-induced Ring1b removal (CreER_T2 cells¹³ ; Figure 2E). Removal of Ring1b in Flk1⁺ hemangioblasts (supplemental Figure 3C-D) resulted in overt expression of Nkx2-2, Nkx2-9, and Sox1 at day 3 (Figure 2F). In contrast to ES cells, where PRC1 loss also caused an up-regulation of downstream hematopoietic regulators (such as Scl, Runx1, and Fli; supplemental Figure 3E top panel), expression of these genes by hemangioblasts was unaffected by PRC1 withdrawal (supplemental Figure 3E bottom panel). Taken together, these data show that there is extensive remodeling of lineage-appropriate genes upon mesoderm commitment, yet several lineage-inappropriate genes capable of executing an alternative fate remain bivalent and primed for expression. Ring1b-depleted Flk1⁺ hemangioblasts were unable to generate hematopoietic colonies (Figure 2G), a defect that was evident after a single day of tamoxifen treatment (supplemental Figure 3F), confirming the importance of PRC1 in the proliferation and differentiation of stem cells.²⁴ 

In conclusion, our data demonstrate that many neural-specifying genes remain bivalently marked and functionally primed for expression in ES-derived committed hemangioblasts. These include Sox1, Nkx2-2, and Nkx2-9 (Figure 2A-B), Ngn1, Ngn2, and Msx1 (data not shown), and Math1, a gene that, although bivalent, showed a delayed timing of replication and insensitivity to PRC1 withdrawal on mesoderm induction. Collectively, these data suggest that hemangioblast precursors may be much more plastic than has previously been appreciated and argue that lineage restriction is probably not the result of an abrupt stage-specific loss of potential (as has been suggested^20,25 ) but rather reflects a more gradual and progressive decline in accessibility of multiple lineage-inappropriate genes over time.

The authors thank Dr Haruhiko Koseki (RIKEN Institute, Japan) for the kind gift of Ring1a^−/−/Ring1b^fl/fl/Cre-ER_T2 ES cells, Amanda Evans (Medical Research Council NIMR, United Kingdom) for technical assistance, and the Medical Research Council CSC flow cytometry facility for FACS.

This work was supported by the Medical Research Council and Epigenome NoE Framework 6.

Contribution: L.M., H.F.J., J.S.-R., and S.P. performed the experiments; L.M., H.F.J., M.M., and A.G.F. wrote the paper; and all authors designed the study.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

The current affiliation for L.M. is Department of Experimental Oncology, European Institute of Oncology, Milan, Italy. The current affiliation for H.F.J. is Medical Research Council National Institute for Medical Research, London, United Kingdom.

Correspondence: Amanda G. Fisher, Medical Research Council Clinical Sciences Centre, Faculty of Medicine Imperial College London, Du Cane Rd, London W12 0NN, United Kingdom; e-mail: amanda.fisher@csc.mrc.ac.uk.