PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Cell. Author manuscript; available in PMC 2013 February 26.
Published in final edited form as:
PMCID: PMC3582033
NIHMSID: NIHMS346916

A poised chromatin platform for TGF-β access to master regulators

Abstract

Specific chromatin marks keep master regulators of differentiation silent, yet poised for activation by extracellular signals. We report that nodal TGF-β signals use the poised histone mark H3K9me3 to trigger differentiation of mammalian embryonic stem cells. Nodal receptors induce the formation of companion Smad4-Smad2/3 and TRIM33-Smad2/3 complexes. TRIM33-Smad2/3 binds the histone marks H3K9me3 and K18ac on the promoters of mesendoderm regulators Gsc and Mixl1. Binding is through the PHD-Bromo cassette of TRIM33. In the crystal structure of this cassette bound to histone H3 peptides, PHD recognizes K9me3 and Bromo an adjacent K18ac. Binding of TRIM33-Smad2/3 to H3K9me3 displaces the chromatin compacting factor HP1γ and makes nodal response elements accessible to Smad4-Smad2/3 for Pol II recruitment. In turn, Smad4 increases K18 acetylation to augment TRIM33-Smad2/3 binding. Thus, nodal cues use the H3K9me3 mark as a platform to switch master regulators of stem cell differentiation from the poised to the active state.

Introduction

The transforming growth factor β (TGF-β family plays major roles in embryonic development, tissue homeostasis, adult immunity, and wound repair (Flavell et al., 2010; Massagué, 2008; Shen, 2007; Wu and Hill, 2009). The family members nodal and bone morphogenetic proteins (BMPs) specify the differentiation of primitive embryonic cells into the three layers of the embryo and their derivative lineages. In both embryonic and differentiated cells, TGF-β family members additionally control a plethora of homeostatic functions including cell proliferation, movement, adhesion, and intercellular communication. Smad transcription factors mediate a majority of these effects. Binding of TGF-β ligands to receptor serine/threonine kinases causes the phosphorylation and activation of Smads 2 and 3 in the case of TGF-β, nodal, activin and myostatin, and of Smads 1, 5 and 8 in the case of BMPs. The activated Smads bind Smad4, accumulate in the nucleus, and target specific enhancer elements by interacting with other DNA-binding proteins. On the DNA, Smad complexes recruit histone acetyltransferases, the Mediator, and SWI/SNF chromatin remodeling complex for transcriptional stimulation of target genes, or histone deacetylases for transcriptional inhibition (Massagué et al., 2005)

Genes that control homeostasis generally are in the active state and respond to TGF-β signals with an increase or a decrease in transcriptional output. In contrast, master regulators of differentiation are transcriptionally repressed yet poised for acute activation by specific developmental signals (Young, 2011). In embryonic stem cells (ESCs) this poised state results from the action of Oct4, Sox2 and Nanog, which are core transcriptional enforcers of pluripotency. These factors stimulate polycomb group (PcG) proteins and SetDB1 to mediate histone methylation. SetDB1 catalyzes trimethylation of Lys9 on histone H3 (H3K9me3). H3K9me3 is present in the promoter region of master regulators and is also a hallmark of heterochromatin (Mikkelsen et al., 2007). It binds heterochromatin protein 1 (HP1), which mediates chromatin compaction (Ruthenburg et al., 2007). Poised genes are transcriptionally quiescent but may harbor paused RNA polymerase II (Pol II) at the start site, ready to proceed with transcription elongation in response to developmental cues (Young, 2011).

The nature of the poised state implies that activation of master regulator genes by the TGF-β/SMAD pathway has specific requirements that may not apply to the regulation of cell homeostasis genes. To investigate how developmental signals activate master regulators we focused on the role of nodal in ESC differentiation. An early effect of nodal during embryogeneis is the activation of goosecoid (Gsc) and Mix-like homeodomain protein 1(Mixl1) (Blum et al., 1992; Chen et al., 1997; Hart et al., 2002), Gsc belongs to the bicoid subfamily of paired homeobox transcription factors, and Mixl1 to the Mix/Bix subfamily. Gsc and Mixl1 are expressed in the primitive streak and function in gastrulation, axial mesendoderm morphogenesis, and endoderm formation. Activation of Gsc and Mixl1 involves a complex of Smad2/3, Smad4, and the forkhead-box family member FoxH1 (Chen et al., 1997; Labbé et al., 1998). This complex binds to sequences known as “activin response elements” (AREs) in the proximal promoters of Gsc and Mixl1 for transcription activation.

In our analysis of how Smad proteins gain access to AREs that are embedded in poised chromatin, several clues led us to focus on TRIM33, a Smad binding protein with contradictory proposed roles (Dupont et al., 2005; He et al., 2006). Trim33 deficient mouse embryos die early, lacking mesoderm and with a phenotype that is suggestive of altered nodal signaling (Morsut et al., 2010). The conditional knockout of Trim33 in premalignant pancreatic progenitors phenocopies that of Smad4 and suggests that TRIM33 and Smad4 converge on pancreatic tumor suppression (Bardeesy et al., 2006; Vincent et al., 2009). In human cells, TRIM33 is dispensable for TGF-β activation of homeostatic gene responses but implicated in TGF-β dependent erythroid differentiation (He et al., 2006). In zebrafish, TRIM33 is required for transcriptional elongation of erythroid differentiation genes (Bai et al., 2010). Notably, TRIM33 has structural features of a histone-binding protein. Here we report an essential role for TRIM33 in the activation of Gsc and Mixl1 by nodal signals, and delineate how Smads gain access to poised promoters of master regulators under the command of nodal TGF-β signals.

Results

TRIM33 is Engaged in the Nodal Smad Pathway

TRIM33 immunostaining was present in the majority of nuclei in all regions of stage E7.5 mouse embryos (Figure 1A). These regions included the node (Figure 1B) and primitive streak (Figure 1C), which under the control of nodal specify the site of gastrulation, determine bilateral symmetry, and undergo mesoderm and endoderm differentiation (Shen, 2007). To probe the involvement of TRIM33 in these processes we used mouse ESCs derived from the inner cell mass (mESCs). Nodal shares receptors with activin, including the type I receptors ALK4 (ActR-IB) and ALK7, and the type II receptors ActR-II and ActR-IIB (Reissmann et al., 2001; Yeo and Whitman, 2001). mESCs express ALK4, ALK7, ActR-II and ActR-IIB (Table S1). We used the ALK4/7 kinase inhibitor SB431542 to block stimulation by nodal-like factors in the media (Figure 1D, first lanes). For receptor activation we used activin A (hereafter, activin), as this ligand is more readily available from mammalian sources than is nodal. Activin addition to mESCs induced the formation of TRIM33-Smad2/3 and Smad4-Smad2/3 complexes, with similar kinetics (t1/2 ~30 min) (Figure 1D). TRIM33 immunoprecipitated complexes contained Smad2/3 but no detectable Smad4 (Figure S1A) and formation of the TRIM33-Smad2/3 complex did not require Smad4, as determined with Smad4-null mESCs (Figure 1E). Conversely depletion of TRIM33 by means of shRNA or knock-out did not inhibit the formation of the Smad4-Smad2/3 complex (Figures 1F, S1B, S1C). Preventing the formation of one complex enhanced the accumulation of the other (Figures 1F, S1B). Thus, nodal/activin stimulation in mESCs rapidly induces the formation of separate TRIM33-Smad2/3 and Smad4-Smad2/3 protein complexes.

Figure 1
TRIM33 is Engaged in the Nodal Smad Pathway

Trim33-dependent and -independent Nodal Gene Responses

When switched to culture conditions that are permissive for differentiation, mESCs form embryoid bodies (EBs) that can subsequently differentiate into all three germ layer fates (Murry and Keller, 2008). EBs recapitulate the effect of nodal on Gsc and Mixl1 activation and induction of mesoderm and endoderm differentiation. We shifted ESCs to differentiation permissive conditions for 2.5 days and determined the requirement of TRIM33, Smad4 and Smad2/3 for the acute induction of Gsc, Mixl1 and Smad7 in response to added activin (Figure 1G-H). Smad7 is an inhibitory Smad commonly induced by TGF-β family members for feedback regulation of the pathway. mESCs that were null for Smad4 (Chu et al., 2004), or depleted of Smad3 in a Smad2-null background (Dunn et al., 2005) (Figure S1D) lacked Gsc, Mixl1 and Smad7 responses to added activin or to endogenous (SB431542-sensitive) nodal-like factors (Figure 1G). Notably, RNAi-mediated depletion of TRIM33 strongly blunted the induction of Gsc and Mixl1 but not that of Smad7. Reconstitution of TRIM33-depleted cells with an shRNA-insensitive TRIM33 vector restored the Gsc and Mixl1 responses (Figure 1G).

Activin also failed to stimulate the expression of Gsc and Mixl1 in two independent Trim33-null ESC lines while it stimulated the expression of the common Smad2/3 target genes Smad7, Skil (encoding SnoN) and SerpinE1 (encoding plasminogen activator inhibitor-1, PAI1) in these cells (Figures 1H, S1B and S1E). Thus, induction of the master regulators Gsc and Mixl1 requires both TRIM33 and Smad4, whereas induction of the homeostasis genes requires Smad4 but not TRIM33 (Figure 1I). Lefty1, which encodes a feedback inhibitor of nodal binding to receptors, showed a complex response profile. Lefty1 induction by activin was blunted in Trim33-null mESCs (Figure S1F, and data not shown) as well as in Smad4-null mESCs that were depleted of TRIM33 (Figure S1G, H).

Genome-wide transcriptomic analysis showed that of all the activin gene responses in EBs (>3 fold change, p<0.05), a large proportion (333 of 407 responses) requires both Smad4 and TRIM33 (Figure 1J). In contrast, most of the TGF-β responses in human HaCaT keratinocytes and MDA-MB-231 breast carcinoma cells were inhibited by RNAi-mediated depletion of Smad2/3 and Smad4, but not by depletion of TRIM33 (Gomis et al., 2006) and our unpublished work). Depletion of TRIM33 did not significantly alter the TGF-β response of SMAD7, SKIL and SERPINE1 in HaCaT or MDA-MB-231 cells, as determined by qRT-PCR (Figures S1K-L). TRIM33 depletion did not diminish the response of luciferase reporter constructs driven by the TGF-β response element from SERPINE1 (PAI1) in HaCaT cells (Figure S1M) or by repeats of the Smad binding element (SBE) CAGAC in MDA-MB-231 cells (Figure S1N). TRIM33 depletion actually increased the activity of these constructs, an effect not observed with endogenous TGF-β target genes (refer to Figures S1K-L).

Mesendodermal Differentiation of ESCs Requires TRIM33

EB differentiation can be tracked by the expression of specific markers (summarized in Figure 2A; (Murry and Keller, 2008). Fgf5 marks the epiblastic primitive ectoderm, and Sox1, nestin and Pax6 mark the definitive ectoderm. Endogenous nodal drives EBs to mesendoderm differentiation through Gsc, Mixl1 and brachyury (also known as T), which mark the primitive streak stage and specify expression of Foxa2. This is followed by the expression of late mesoderm and endoderm markers (Figure 2A). In the embryo, nodal expression is first detected in the inner cell mass of the blastocyst (E4.5) and persists until the extended primitive streak stage (E7.5) (Collignon et al., 1996). In EBs, nodal was expressed up to day 6, and the expression was not affected by the absence of TRIM33 (Figure 2C). The nodal co-receptors cripto and cryptic (Yeo and Whitman, 2001) were sequentially expressed in EBs up to day 6 (Figure S2A). RNAi-mediated depletion of both co-receptors (Figure S2B) inhibited the expression of Gsc and Mixl-1 (Figure S2C and data not shown), suggesting that autocrine nodal drives the expression. Cripto/cryptic-depleted cells remained responsive to activin (Figure S2D).

Figure 2
TRIM33 is Required for Mesendodermal Differentiation of Mammalian ESCs

Absence of TRIM33 in EBs had little effect on the expression of Fgf5 but it abolished the expression of Gsc, Mixl1 and brachyury (Figure 2D) and of the late mesoderm markers Foxa2, Scl/Tal and Nkx2.5 (Figure 2E). Notably, TRIM33 was dispensable for ectoderm differentiation (Figure 2F). Similar results were obtained by RNAi-mediated depletion of TRIM33 in mESCs of a different genetic background (Figures S1C and S2E-G). Expression of the definitive endoderm marker Pdx1, which requires a high concentration of activin (Kubo et al., 2004), was also blunted in TRIM33 depleted EBs (Figure S2H). Furthermore, TRIM33 depletion in human ESCs (Figure S2I) markedly inhibited the expression of GSC, MIXL1 and T/BRACHYURY (Figure 2G) as well as that of FOXA2 and SOX17 (Figure S2J), which are endoderm markers in human ESCs (Zorn and Wells, 2009). The inhibition of mesoderm but not ectoderm differentiation in TRIM33-depleted mouse EBs was confirmed by marker protein immunostaining (Figure 2H).

TRIM33 Recognizes a Dual Histone Mark Motif

TRIM33 consists of an N-terminal region containing RING, B-box, and coiled-coil domains (tripartite motif, TRIM), a middle region that binds Smad2/3, and a C-terminal region that contains one plant-homeodomain (PHD) adjacent to a Bromodomain (Bromo) region (He et al., 2006) (Figure 3A). By reconstituting TRIM33-depleted EBs with either the wild-type TRIM33 or a TRIM33(Δphd) construct lacking the PHD finger, we observed that the PHD finger is essential for the induction of Gsc and Mixl1 by added activin (Figure 3B) or autocrine nodal (Figure 3C). TRIM33 with the double mutation C125A and C128A in the RING domain, which is designed to disrupt protein ubiquitylation (Dupont et al., 2005), also rescued the Gsc, Mixl1 and Lefty1 responses in Trim33-null EBs (Figure S3A-B).

Figure 3
TRIM33 Recognizes a Dual Histone Mark Typical of Poised Chromatin

PHD and Bromo domains can bind post-translationally modified histones (Taverna et al., 2007). Using recombinant, epitope-tagged TRIM33 PHD-Bromo cassette and histone N-terminal peptides we observed a weak basal binding of TRIM33 to unmodified histone H3, and a gain in affinity for peptides containing K9 methylation, particularly trimethylation (Figure 3D). H3K4me3 and H3K27me3 did not mediate TRIM33 binding (Figure 3E). Deletion of the PHD finger, deletion of the Bromo domain, or alanine mutation of three highly conserved residues in the PHD finger (Wang et al., 2009), abolished the H3 binding activity (Figure 3D). Isothermal titration calorimetry (ITC) binding measurements confirmed that the TRIM33 PHD prefers to bind H3 peptide containing unmodified K4 and trimethylated K9 (Figure 3F; Table S2).

We next tested the contribution of histone lysine acetylation to TRIM33 binding. A histone H3 N-terminal peptide acetylated at K18 showed enhanced binding to TRIM33 compared to the unmodified peptide whereas acetylation at K14 had little effect (Figure 3G). The combination of K9me3 and K18ac bound to TRIM33 PHD-Bromo more avidly than any of these marks alone (Figure 3G, 3H). A histone H3 N-terminal peptide containing K9me3, K14ac, K18ac and K23ac bound slightly better to TRIM33 than did the H3K9me3-K18ac peptide (Figure S3C). Collectively these results demonstrate a high affinity of the TRIM33 PHD-Bromo cassette for histone H3 containing unmodified K4, K9me3 and K18ac (summarized in Figure 3A).

Structural Basis for K9me3-K18ac Recognition by TRIM33 PHD-Bromo Cassette

Based on previous knowledge (Dhalluin et al., 1999; Lan et al., 2007; Li et al., 2006; Pena et al., 2006), we hypothesized that the TRIM33 PHD-Bromo cassette could simultaneously target unmodified K4 and K9me3 marks using its PHD finger and Kac marks on the same H3 tail using its Bromo. To address this issue, we solved crystal structures of the TRIM33 PHD-Bromo cassette both in the free state and bound to H3 peptides containing different modifications (Table S3). The 3.1 Å crystal structure of the TRIM33 PHD-Bromo cassette in the free state was compared with its TRIM24 counterpart (Figure 4A) (Tsai et al., 2010). In both structures, the PHD finger adopts the characteristic cross-braced folding topology stabilized by a pair of bound Zn ions, while the Bromo adopts the characteristic left-handed four-helical bundle topology (Taverna et al., 2007). The PHD finger and Bromo interact extensively and utilize similar interfaces in TRIM33 and TRIM24 (Figure 4A). Notably, there is a 17-amino acid insert within the Bromo domain of TRIM33 (Figure S4A), and this is reflected in an a-helical extension of helix αB in TRIM33 that abuts the binding pocket of the Bromo domain (Figure 4A).

Figure 4
Structural Basis for TRIM33 PHD-Bromo Cassette Binding to Histone H3 Marks

In the 2.7 Å crystal structure of TRIM33 PHD-Bromo cassette bound to H3(1-22)K9me3-K14acK18ac peptide (Figure 4B), a single modified H3 peptide traverses both the PHD finger and the Bromo. The cassette forms a single functional unit for potential recognition of multiple marks on a single histone tail. The H3(1-10)K9me3 segment forms an anti-parallel β sheet with a segment of the TRIM33 PHD finger (Figure 4C). The key intermolecular contacts are between the positively-charged N-terminus and the peptide backbone, multiple hydrogen bonds between the ammonium group of unmethylated K4 and acidic residues Asp884, Asp888 and the carbonyl oxygen of Glu887, and stacking of the trimethyl group of K9me3 over the indole ring of Trp889. The H3K18ac residue is inserted into the Bromo binding pocket, with sequence specificity associated with recognition of the R17-K18ac step, given that the guanidinium group of R17 forms a hydrogen bond with the side chain of Glu981 from Bromo (Figure 4D). By contrast, the K14ac side chain is disordered and appears not to be involved in intermolecular recognition in this complex.

The longer H3(1-28)K9me3-K14acK18acK23ac peptide also spans a single TRIM33 PHD-Bromo cassette in the 2.8 Å crystal structure of its complex (Figure S4B). Importantly, it is the K18ac rather than the K23ac side chain that is positioned in the Bromo binding pocket, with the side chains of both K14ac and K23ac being disordered in the complex. By contrast, the shorter H3(1-20)K9me3-K14ac peptide binds two TRIM33 PHD-Bromo cassettes in the 2.0 Å crystal structure of its complex, such that the H3(1-10) segment containing the unmodified K4 and K9me3 mark targets the PHD finger of one cassette, while the H3(11-20) segment containing the K14ac mark targets the Bromo of another (Figure S4C). These data suggest that K14ac is too close to K9me3 for recognition by the PHD and Bromo domains of the same TRIM33 molecule.

In addition to Trp stacking, a hydrogen bond between the side chain carbonyl of Gln895 and the amide proton of K9me3, as well as a stabilizing non-conventional carbon-oxygen hydrogen bond between the carbonyl oxygen of Gln894 and a methyl group of K9me3, also contribute to the K9me3 recognition (Figure S4D). In the highest resolution structure of the Bromo domain, we observe an array of water molecules that mediate Kac recognition within the Bromo binding pocket (Figure S4E, S4F). In the classical binding mode, as in the case of TRIM24, an invariant Asn in the Bromo binding pocket forms a stabilizing hydrogen bonding with the acetyllysine, whereas in the TRIM33 complex, the corresponding Asn1039 is 8.2 Å away from the oxygen of the acetyllysine head group (Figure S4G).

We used ITC to further test the cooperativity of unmodified K4, trimethylated K9 and acetylated K18 on the same H3(1-28) tail for binding to TRIM33 PHD-Bromo (Table S2). The binding of TRIM33 PHD-Bromo to H3(1-28) (KD=0.46 μM) was enhanced 2.3 fold by either trimethylation of K9 or acetylation of K18, and 3.5 fold by both marks (Figure 3H). The mutants E981A, W889A and C901W, which would disrupt binding involving H3R17, H3K9me3 and H3K4 respectively, showed reduced affinity for the H3 peptide (Figures 3G, ,4E).4E). These data establish that unmodified K4, trimethylated K9 and acetylated K18 are read in a combinatorial manner by TRIM33 PHD-Bromo, thereby resulting in the highest binding affinity.

Signal-driven Binding of Trim33 to H3K9me3 Domains of the Gsc and Mixl1 Promoters

Our evidence linked the chromatin-binding function of TRIM33 to the induction of Gsc and Mixl1 by nodal/activin signals. Therefore, we investigated the interaction of TRIM33 and Smads with the promoter regions of these two genes. Both promoters contain an ARE immediately upstream of the start site. AREs consist of a Smad binding element (SBE) and FoxH1 cognate sequences. FoxH1 binds directly to Smad2/3 in the Smad2/3-Smad4 complex, and the proteins cooperatively recognize the ARE (Chen et al., 1997; Hart et al., 2005; Labbé et al., 1998).

Chromatin immunoprecipitation (ChIP) assays in mESCs using antibodies against chromatin marks revealed a conserved H3K9me3-rich region 2.3 kb upstream of the Gsc transcription start site promoter, well separated from the ARE (Figure 5B, summarized in Figure 5A). This region also contained H3K14ac, H3K18ac, and H3K23ac, and low levels of H3K4me3 (Figure 5B, and refer to Figure S6), which was compatible with TRIM33 binding. ChIP of mononucleosome preparations showed the presence of both H3K9me3 and H3K18ac in the same nucleosome at -2,3kb in the Gsc promoter (Figures 5C, S5A). This region also contained four SBEs between -2327 and -2281. Indeed, TRIM33 ChIP revealed a strong active independent binding to this region, but not to other regions (Figure 5D). Smad2/3 ChIP showed activin-dependent binding of Smad2/3 not only to the ARE at -0.4 kb, but also to the K9me3 region at -2.3 kb (Figure 5D). Furthermore, depletion of Smad2/3 prevented binding of TRIM33 to this region (Figure 5D). Activin-dependent binding of Smad4 occurred only at the ARE (refer to Figure 6A).

Figure 5
Signal-driven Binding of TRIM33 to H3K9me3 Domains of theGsc and Mixl1 Promoters
Figure 6
Smad2/3-Smad4 Binding to AREs in Poised Promoters Required TRIM33

The region upstream of the Mixl1 gene contains three SBEs at -0.5 kb in addition to an ARE at -0.2 kb. This region also contains H3K9me3 and H3K18ac on the same nucleosome, although these marks were also present in upstream regions (Figures S5B-C, summarized in Figure 5E). Activin-dependent binding of TRIM33 occurred near the Mixl1 ARE, overlapping the region of Smad2/3 and Smad4 binding (Figure 5E,F, and refer to Figure 6B). Depletion of Smad2/3 inhibited the activin-dependent binding of TRIM33 (Figure S5D). Collectively, these results suggest that the ligand-induced TRIM33-Smad2/3 and Smad4-Smad2/3 complexes bind, respectively, to H3K9me3-K18ac and ARE sites in the Gsc and Mixl1 promoters (Figure 5G).

TRIM33 is Essential for Smad2/3-Smad4 Binding to AREs in Poised Promoters

Notably, Trim33 knockdown inhibited not only the binding of Smad2/3 to the H3K9me3 region but also the binding of Smad2/3 and Smad4 to the ARE (Figure 6A). Similar results were obtained with the Mixl1 promoter (Figure 6B). Reconstitution of TRIM33 expression in the depleted cells with wild type TRIM33 rescued the binding of Smad4 to the ARE in both the Gsc and Mixl1 promoters, whereas reconstitution with TRIM33(Δphd) did not (Figure 6C). These results suggest that the signal-driven binding of TRIM33 and Smad2/3 to the H3K9me3 region enables binding of Smad4 and Smad2/3 to the AREs (Figure 6D).

Comparison of wild-type and Smad4-null ESCs revealed that Smad4 in turn enhanced the activin-induced binding of TRIM33 to its sites (Figures 7A and S6A). As Smad2/3-Smad4 complexes recruit p300 and other histone acetyl transferases to the chromatin (Massagué et al., 2005), we asked whether activin stimulation induced acetylation of the Gsc and Mixl1 promoters. ChIP assays demonstrated a rapid, marked increase in the level of H3K18ac in the TRIM33 binding regions and throughout the promoters, in response to activin (Figure 7B). The levels of H3K14ac and H3K23ac were also increased by activin (Figure S6B). Furthermore, this increase in histone H3 K18 acetylation was dependent on Smad4 and TRIM33 (Figures 7B, and S6B-G). Activin did not affect the level of H3K9me3 throughout these regions of the Gsc and Mixl1 promoters (Figure S6D and E). Thus, while the binding of TRIM33-Smad2/3 to H3K9me3 is essential for binding of Smad4 and Smad2/3 to the AREs, Smad4 reciprocally enriches the TRIM33-binding region with H3K18ac mark and enhances the binding of TRIM33 to this region.

Figure 7
TRIM33 and SMAD Cooperation in Activation of Master Regulators

Trim33-mediated Displacement of HP1 from H3K9me3-K18ac Dual Marks

The three HP1 family members (HP1α, β, and γ) bind to H3K9me3 on pericentric and telomeric chromatin to enforce heterochromatin organization. HP1 dimerization is thought to cause strapping and condensation of H3K9me3-rich chromatin domains (Ruthenburg et al., 2007). HP1β and γ are also found in developmentally regulated, silent regions of the euchromatin in ESCs (Kwon and Workman, 2008). Therefore, we investigated the ability of TRIM33 to compete with HP1γ for binding to histone H3 peptides containing other modifications of interest (Figure 7C). A 3-fold molar excess of TRIM33 was ineffective at competing with HP1γ for binding to peptide containing K9me3 as the sole modification (Figure 7C). However, TRIM33 effectively outcompeted HP1γ in H3 peptides that additionally contained K18ac (Figure 7C). These results are consistent with the higher affinity of TRIM33 for H3K9me3-K18ac compared to H3K9me3 as the sole mark.

As a corollary to these results we investigated the effect of activin on the occupancy of TRIM33 binding sites by HP1 in ESCs. HP1γ ChIP assays revealed binding of the endogenous protein to the TRIM33 target regions of the Gsc and Mixl1 promoters and this binding was inhibited by activin addition in a TRIM33-dependent manner (Figure 7D). This was accompanied by an increase in the level of Pol II binding to the initiator region (Figure 7E), a hallmark of promoter activation (Young, 2011).

Discussion

Nodal-driven Smad Activation of Master Regulators

The Gsc and Mixl1 promoters in ESCs contain key features of the poised state, including the H3K9me3 mark of quiescent chromatin, the chromatin compacting factor HP1γ bound to this mark, and a basal level of RNA Pol II loaded on the start site. Our evidence suggests that nodal TGF-β signals activate the expression of master mesendoderm regulators through the following sequence of events (Figure 7F). Nodal signaling triggers the formation of companion TRIM33-Smad2/3 and Smad4-Smad2/3 complexes. The Smad4-Smad2/3 complex (together with FoxH1) binds to ARE sites in the Gsc and Mixl1 promoters, but for this to happen TRIM33-Smad2/3 must act first. TRIM33 serves as a histone binding protein that recognizes the dual mark H3K9me3-K18ac through the PHD-Bromo cassette. TRIM33 binds to nucleosomes containing both marks in the Gsc and Mixl1 promoters. The superior affinity of TRIM33 for H3K9me3-K18ac displaces bound HP1γ. However, this requires Smad2/3 and nodal input, suggesting that, in the complex, TRIM33 binds to H3K9me3-K18ac and Smad2/3 to an adjacent CAGAC sequence. These events may regionally remodel the chromatin to enable the access of Smad4-Smad2/3 to AREs, which are located nearby in Mixl1 or 2 kb downstream in Gsc. The cooperation between the two Smad complexes is reciprocal. The Smad4-Smad2/3 complex, which is known to recruit histone acetyltransferases, increases the regional content of H3K18ac and augments the binding of TRIM33-Smad2/3 to the promoters. As a result of these events, nodal firmly switches the master regulators Gsc and Mixl1 from poised to activated state, thereby committing primitive embryo cells to mesendodermal fates (Figure 7F).

Other nodal/activin gene responses in ESCs require Smad4 but not TRIM33. Induction of the feedback regulator Smad7 and the secreted protease inhibitor SerpinE1 are examples. Such TRIM33-independent TGF-β responses are the rule in human keratinocytes and breast cancer cells. In differentiated cells the Smad pathway primarily controls cell homeostasis through up- or down-regulation of active genes. Smad4-Smad2/3 access to such promoters may solely depend on the availability of specific DNA-binding Smad partners, as envisioned in the canonical Smad pathway (Massagué et al., 2005).

Structural Basis for TRIM33 Reading of Dual Histone Marks

The PHD-Bromo domain cassette of TRIM33 is central to its role as a nodal-driven switch of the poised chromatin state. This cassette targets TRIM33 to K9me3 and K18ac marks together with unmodified K4 on the same H3 tail. H3K9me3 stacks over a single Trp ring of the TRIM33 PHD finger, in contrast to the anticipated requirement of an aromatic cage for such cation-π recognition (Taverna et al., 2007). The structure also explains why methylation of K4 disrupts recognition by the PHD finger. The TRIM33 Bromo domain specifically recognizes the R17-K18ac motif, thereby discriminating against other Kac residues in histone H3. Furthermore, the TRIM33 Bromo forms an atypical binding pocket, with the inserted Kac side chain stabilized by a water-mediated hydrogen-bonding network. It is noteworthy that while the TRIM24 Bromo recognizes the H3K23ac mark (Tsai et al., 2010), the TRIM33 Bromo recognizes H3K18ac, as dictated by the PHD-K9me3 anchoring interaction. TRIM24 binds estrogen receptor for activation of genes associated with cell proliferation and tumorigenesis (Tsai et al., 2010), establishing an interesting parallel with the cooperation of TRIM33 with Smad2/3.

The present studies highlight the key contribution of spacer length between the methylation and acetylation marks of histone H3, as well as sequence context, to molecular recognition. A recent report describing binding of TRIM33 to two acetylated histone H3 marks (Agricola et al., 2011) missed the critical interaction of the TRIM33 PHD-Bromo cassette with K9me3, and postulated interactions of TRIM33 with the dual mark K18ac-K23ac. Binding of a TRIM33 molecule to both acetylation marks on the same histone H3 tail is not tenable based on the present crystal structures. The combination of K4, K9me3 and K18ac marks constitutes a unique recognition code, which greatly enhances both the binding affinity and specificity of TRIM33 for modified H3 tails. The combined binding specificity for modified histone and DNA mediated by TRIM33 and Smad2/3, respectively, ensures further selectivity in target gene recognition.

TRIM33 in Differentiation Control

TRIM33 is a multi-functional protein and the presently described role of TRIM33 in the activation of master regulators involves the Smad2/3 and histone binding domains. Recently TRIM33 was shown to mediate transcription elongation of erythroid differentiation genes in hematopoietic progenitors, and this effect involves TRIM33 binding to the hematopoietic transcription factor SCL and the elongation factors p-TEFb and FACT (Bai et al., 2010). These two roles of TRIM33 –providing Smad with access to poised differentiation genes, and promoting transcription elongation of differentiation genes– could represent alternative modes of transcriptional control or coupled events in a common process of gene regulation.

TRIM33 was proposed to function as a general inhibitor of TGF-β and BMP pathways because it can mediate poly-ubiquitylation and degradation of Smad4 (Dupont et al., 2005) or inhibitory Smad4 mono-ubiquitylation (Agricola et al., 2011). However, our evidence reveals an essential cooperation between TRIM33 and Smad4 in the transcriptional control of embryonic cell differentiation, and rules out TRIM33 as a general inhibitor of Smad4. Trim33 and Smad4 conditional knockout models of mouse pancreatic cancer provide further evidence for functional convergence of TRIM33 and Smad4 (Bardeesy et al., 2006; Vincent et al., 2009). TRIM33-mediated ubiquitylation could represent a built-in negative feedback activity. As precedent, the Pol II kinases CDK8 and CDK9 phosphorylate Smads1~3 for full activation, but in the process prime Smads for ubiquitylation and degradation (Alarcón et al., 2009; Aragón., 2011).

Trim33 null mouse embryos fail to form primitive streak and mesoderm, but, paradoxically, these embryos show a gain in certain nodal/Smad4 response markers (Morsut et al., 2010). Analysis of the Trim33 mutant phenotype is complicated by the involvement of TRIM33 in the regulation of Lefty-1 by nodal. Lefty-1 is negative feedback factor that prevents nodal from binding to its receptors and restricts nodal signaling (Shen, 2007). In the absence of TRIM33, a defective Lefty-1 feedback loop would enhance nodal signaling through the remaining Smad4 branch, suggesting an explanation for the paradoxical gain in Smad4 signaling markers in Trim33 null mouse embryos.

Conclusion

We propose a mechanism for the activation of master regulators in which nodal signaling drives the generation of two separate transcriptional complexes, one that targets the H3K9me3 mark of poised chromatin and the other that targets response elements in the same promoter. These insights reveal a biochemical basis for the poised state of master regulator genes. H3K9me3 provides a binding site for HP1 factors that impose gene repression but, at the same time, H3K9me3 provides an entry point for TRIM33 to displace HP1 and allow binding of signal-driven Smads for gene expression. Through this mechanism, nodal TGF-β signals turn a repressive chromatin mark into a platform for activation of master regulators of mesendodermal differentiation. Related mechanisms could operate under the control of other developmental signals that must bypass repressive chromatin marks in order to control cell fate.

Experimental Procedures

Cell lines and differentiation assays

Mouse embryonic stem cells E14Tg2a.IV were maintained in feeder-layer free LIF-supplemented medium (Keller, 1995). Stable Trim33 knockdown in mESCs cells was done using LKO1 lentiviral constructs expressing shRNA against mouse Trim33 (Sigma, Clone ID: TRCN0000039531). Trim33-null ESCs in C57Bl/6J background we maintained on mouse embryonic fibroblasts (MEFs, Global Stem) with LIF-supplemented medium. Lentiviral vector infections (Stewart et al., 2003) and plasmid transfections (Sapkota et al., 2007) were performed as previously described. mESCs were maintained on gelatin-coated plates. Embryoid body (EB) formation and differentiation were carried out as described by the supplier (ATCC). Ectoderm differentiation assays were carried out as described previously (Ying et al., 2003). Prior to total RNA extraction cells were treated with activin A (50 ng/ml, R&D Systems) or SB431542 (10 μM, TOCRIS). Human embryonic stem cells (hESCs) (WA-09; passage 35-45) were cultured on MEFs. The medium was DMEM/F12 (GIBCO), supplemented with 20% KnockOut™ Serum Replacement (GIBCO), 0.1 mM β-mercaptoethanol, and 6 ng/mL of FGF-2. Cells were passaged with 6 U/mL of dispase in hESCs media, washed, and replated at a dilution of 1:3 to 1:5. For mesodermal cell differentiation, hESCs were plated on Matrigel and cultured in RPMI-B27 medium (Invitrogen). Recombinant human activin A (100 ng/ml, Peprotech) was supplemented for 24h and 10ng/ml human recombinant BMP4 (R&D Systems) was added for 5 following days as described previously (Laflamme et al., 2007). HaCaT keratinocytes, HEK293T human embryonic kidney cells, and MDA-MB-231 cells were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% FBS.

Crystallization and structure determination

Details of crystallization conditions, data collection and refinement are provided under Supplemental Experimental Procedures. In brief, all crystals were grown at 20 °C by the hanging-drop method. Data sets for crystals of TRIM33 PHD-Bromo in the free state and its complexes were collected at synchrotron beamlines with the structures of TRIM33 PHD-Bromo cassette in both the free state and with bound peptides solved by molecular replacement using the published structure of the free form of TRIM24 (PDB: 3O33) as the search model. The initial models were rebuilt in COOT and further refined in PHENIX.

Chromatin immunoprecipitation

For chromatin immunoprecipitation, mESCs embryoid bodies at 2.5 days were incubated with human recombinant activin A (50 ng/ml; R&D Systems) for 1 h or SB431542 (10 μM; TOCRIS) for 2 h. Cells were then cross-linked with 1% formaldehyde at 37 °C for 10 min and quenched with 0.125 M glycine for 5 min at room temperature. ChIP was performed using a ChIP assay kit (Millipore) as described in the manufacturer's protocol. Immunoprecipitated DNA was analyzed by absolute qRT-PCR and the amplification product was expressed as percentage of the input for each condition. For mono-nucleosome ChIP, cell lysates were digested with titrations of micrococal nuclease (Sigma N5386) and mono-nucleosome yield was verified by agarose gel electrophoresis. PCR primer pairs used to amplify the unrelated-control or promoter regions of indicated genes are indicated in the Supplementary section.

Accession Numbers

The X-ray coordinates of TRIM33 PHD-Bromo cassette in the free states and when bound to H3(1-20)K9me3-K14ac, H3(1-22)K9me3-K14acK18ac and H3(1-28)K9me3-K14acK18acK23ac histone peptides have been deposited in the Protein Data Bank (PDB) with the accession numbers of 3U5M, 3U5N, 3U5O and 3U5P respectively. The microarray gene expression dataset for ESCs has been deposited under GEO accession number GSE32903.

Supplementary Material

01

02

Acknowledgments

We would like to thank E.J. Robertson, D. Rifkin and Y. Kang for reagents; W. He, E.P. Papapetrou, S. Acharyya and C. J. David for helpful discussions; A. Viale for valuable help with microarray analysis; and P.Y. Chen, S. Francis, and N. Fan for technical help; the staff at beamline 24ID-C of the Advanced Photon Source at the Argonne National Laboratory and beamline X29 of the National Synchrotron Light Source at the Brookhaven National Laboratory, for assistance with data collection. Support for the MSKCC Transgenic Mouse Core, Genomics Core, and Molecular Cytology Core were provided by the National Cancer Institute. LS is supported by NIH grant NS052671 as well as funds from the Starr Foundation and from NYSTEM. DJP is supported by funds from the Abby Rockefeller Mauze Trust and the Maloris Foundation, as well as funds from the Leukemia and Lymphoma Society and the Starr Foundation. JM is supported by NIH grant CA34610, and is a Howard Hughes Medical Institute Investigator.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  • Agricola E, Randall RA, Gaarenstroom T, Dupont S, Hill CS. Recruitment of TIF1gamma to Chromatin via Its PHD Finger-Bromodomain Activates Its Ubiquitin Ligase and Transcriptional Repressor Activities. Mol Cell. 2011;43:85–96. [PubMed]
  • Alarcón C, Zaromytidou AI, Xi Q, Gao S, Yu J, Fujisawa S, Barlas A, Miller AN, Manova-Todorova K, Macias MJ, et al. Nuclear CDKs drive Smad transcriptional activation and turnover in BMP and TGF-beta pathways. Cell. 2009;139:757–769. [PMC free article] [PubMed]
  • Aragón E, Goerner N, Zaromytidou AI, Xi Q, Escobedo A, Massagué J, Macias MJ. A Smad action turnover switch operated by WW domain readers of a phosphoserine code. Genes Dev. 2011;25:1275–1288. [PubMed]
  • Bai X, Kim J, Yang Z, Jurynec MJ, Akie TE, Lee J, LeBlanc J, Sessa A, Jiang H, DiBiase A, et al. TIF1gamma controls erythroid cell fate by regulating transcription elongation. Cell. 2010;142:133–143. [PMC free article] [PubMed]
  • Bardeesy N, Cheng KH, Berger JH, Chu GC, Pahler J, Olson P, Hezel AF, Horner J, Lauwers GY, Hanahan D, et al. Smad4 is dispensable for normal pancreas development yet critical in progression and tumor biology of pancreas cancer. Genes Dev. 2006;20:3130–3146. [PubMed]
  • Blum M, Gaunt SJ, Cho KW, Steinbeisser H, Blumberg B, Bittner D, De Robertis EM. Gastrulation in the mouse: the role of the homeobox gene goosecoid. Cell. 1992;69:1097–1106. [PubMed]
  • Chen X, Weisberg E, Fridmacher V, Watanabe M, Naco G, Whitman M. Smad4 and FAST-1 in the assembly of activin-responsive factor. Nature. 1997;389:85–89. [PubMed]
  • Chu GC, Dunn NR, Anderson DC, Oxburgh L, Robertson EJ. Differential requirements for Smad4 in TGFbeta-dependent patterning of the early mouse embryo. Development. 2004;131:3501–3512. [PubMed]
  • Collignon J, Varlet I, Robertson EJ. Relationship between asymmetric nodal expression and the direction of embryonic turning. Nature. 1996;381:155–158. [PubMed]
  • Dhalluin C, Carlson JE, Zeng L, He C, Aggarwal AK, Zhou MM. Structure and ligand of a histone acetyltransferase bromodomain. Nature. 1999;399:491–496. [PubMed]
  • Dunn NR, Koonce CH, Anderson DC, Islam A, Bikoff EK, Robertson EJ. Mice exclusively expressing the short isoform of Smad2 develop normally and are viable and fertile. Genes Dev. 2005;19:152–163. [PubMed]
  • Dupont S, Mamidi A, Cordenonsi M, Montagner M, Zacchigna L, Adorno M, Martello G, Stinchfield MJ, Soligo S, Morsut L, et al. FAM/USP9x, a deubiquitinating enzyme essential for TGFbeta signaling, controls Smad4 monoubiquitination. Cell. 2009;136:123–135. [PubMed]
  • Dupont S, Zacchigna L, Cordenonsi M, Soligo S, Adorno M, Rugge M, Piccolo S. Germ-layer specification and control of cell growth by Ectodermin, a Smad4 ubiquitin ligase. Cell. 2005;121:87–99. [PubMed]
  • Flavell RA, Sanjabi S, Wrzesinski SH, Licona-Limon P. The polarization of immune cells in the tumour environment by TGFbeta. Nat Rev Immunol. 2010;10:554–567. [PMC free article] [PubMed]
  • Gomis RR, Alarcón C, He W, Wang Q, Seoane J, Lash A, Massagué J. A FoxO-Smad synexpression group in human keratinocytes. Proc Natl Acad Sci U S A. 2006;103:12747–12752. [PubMed]
  • Hart AH, Hartley L, Sourris K, Stadler ES, Li R, Stanley EG, Tam PP, Elefanty AG, Robb L. Mixl1 is required for axial mesendoderm morphogenesis and patterning in the murine embryo. Development. 2002;129:3597–3608. [PubMed]
  • Hart AH, Willson TA, Wong M, Parker K, Robb L. Transcriptional regulation of the homeobox gene Mixl1 by TGF-beta and FoxH1. Biochem Biophys Res Commun. 2005;333:1361–1369. [PubMed]
  • He W, Dorn DC, Erdjument-Bromage H, Tempst P, Moore MA, Massagué J. Hematopoiesis controlled by distinct TIF1gamma and Smad4 branches of the TGFbeta pathway. Cell. 2006;125:929–941. [PubMed]
  • Keller GM. In vitro differentiation of embryonic stem cells. Curr Opin Cell Biol. 1995;7:862–869. [PubMed]
  • Kubo A, Shinozaki K, Shannon JM, Kouskoff V, Kennedy M, Woo S, Fehling HJ, Keller G. Development of definitive endoderm from embryonic stem cells in culture. Development. 2004;131:1651–1662. [PubMed]
  • Kwon SH, Workman JL. The heterochromatin protein 1 (HP1) family: put away a bias toward HP1. Mol Cells. 2008;26:217–227. [PubMed]
  • Labbé E, Silvestri C, Hoodless PA, Wrana JL, Attisano L. Smad2 and Smad3 positively and negatively regulate TGF beta-dependent transcription through the forkhead DNA-binding protein FAST2. Mol Cell. 1998;2:109–120. [PubMed]
  • Laflamme MA, Chen KY, Naumova AV, Muskheli V, Fugate JA, Dupras SK, Reinecke H, Xu C, Hassanipour M, Police S, et al. Cardiomyocytes derived from human embryonic stem cells in pro-survival factors enhance function of infarcted rat hearts. Nat Biotechnol. 2007;25:1015–1024. [PubMed]
  • Lan F, Collins RE, De Cegli R, Alpatov R, Horton JR, Shi X, Gozani O, Cheng X, Shi Y. Recognition of unmethylated histone H3 lysine 4 links BHC80 to LSD1-mediated gene repression. Nature. 2007;448:718–722. [PMC free article] [PubMed]
  • Li H, Ilin S, Wang W, Duncan EM, Wysocka J, Allis CD, Patel DJ. Molecular basis for site-specific read-out of histone H3K4me3 by the BPTF PHD finger of NURF. Nature. 2006;442:91–95. [PMC free article] [PubMed]
  • Massagué J. TGFbeta in Cancer. Cell. 2008;134:215–230. [PMC free article] [PubMed]
  • Massagué J, Seoane J, Wotton D. Smad transcription factors. Genes Dev. 2005;19:2783–2810. [PubMed]
  • Mikkelsen TS, Ku M, Jaffe DB, Issac B, Lieberman E, Giannoukos G, Alvarez P, Brockman W, Kim TK, Koche RP, et al. Genome-wide maps of chromatin state in pluripotent and lineage-committed cells. Nature. 2007;448:553–560. [PMC free article] [PubMed]
  • Morsut L, Yan KP, Enzo E, Aragona M, Soligo SM, Wendling O, Mark M, Khetchoumian K, Bressan G, Chambon P, et al. Negative control of Smad activity by ectodermin/Tif1gamma patterns the mammalian embryo. Development. 2010;137:2571–2578. [PubMed]
  • Murry CE, Keller G. Differentiation of embryonic stem cells to clinically relevant populations: lessons from embryonic development. Cell. 2008;132:661–680. [PubMed]
  • Pena PV, Davrazou F, Shi X, Walter KL, Verkhusha VV, Gozani O, Zhao R, Kutateladze TG. Molecular mechanism of histone H3K4me3 recognition by plant homeodomain of ING2. Nature. 2006;442:100–103. [PMC free article] [PubMed]
  • Reissmann E, Jornvall H, Blokzijl A, Andersson O, Chang C, Minchiotti G, Persico MG, Ibanez CF, Brivanlou AH. The orphan receptor ALK7 and the Activin receptor ALK4 mediate signaling by Nodal proteins during vertebrate development. Genes Dev. 2001;15:2010–2022. [PubMed]
  • Ruthenburg AJ, Li H, Patel DJ, Allis CD. Multivalent engagement of chromatin modifications by linked binding modules. Nat Rev Mol Cell Biol. 2007;8:983–994. [PMC free article] [PubMed]
  • Sapkota G, Alarcón C, Spagnoli FM, Brivanlou AH, Massagué J. Balancing BMP signaling through integrated inputs into the Smad1 linker. Mol Cell. 2007;25:441–454. [PubMed]
  • Shen MM. Nodal signaling: developmental roles and regulation. Development. 2007;134:1023–1034. [PubMed]
  • Stewart SA, Dykxhoorn DM, Palliser D, Mizuno H, Yu EY, An DS, Sabatini DM, Chen IS, Hahn WC, Sharp PA, et al. Lentivirus-delivered stable gene silencing by RNAi in primary cells. RNA. 2003;9:493–501. [PubMed]
  • Taverna SD, Li H, Ruthenburg AJ, Allis CD, Patel DJ. How chromatin-binding modules interpret histone modifications: lessons from professional pocket pickers. Nat Struct Mol Biol. 2007;14:1025–1040. [PMC free article] [PubMed]
  • Tsai WW, Wang Z, Yiu TT, Akdemir KC, Xia W, Winter S, Tsai CY, Shi X, Schwarzer D, Plunkett W, et al. TRIM24 links a non-canonical histone signature to breast cancer. Nature. 2010;468:927–932. [PMC free article] [PubMed]
  • Vincent DF, Yan KP, Treilleux I, Gay F, Arfi V, Kaniewski B, Marie JC, Lepinasse F, Martel S, Goddard-Leon S, et al. Inactivation of TIF1gamma cooperates with Kras to induce cystic tumors of the pancreas. PLoS Genet. 2009;5:e1000575. [PMC free article] [PubMed]
  • Wang GG, Song J, Wang Z, Dormann HL, Casadio F, Li H, Luo JL, Patel DJ, Allis CD. Haematopoietic malignancies caused by dysregulation of a chromatin-binding PHD finger. Nature. 2009;459:847–851. [PMC free article] [PubMed]
  • Wu MY, Hill CS. Tgf-beta superfamily signaling in embryonic development and homeostasis. Dev Cell. 2009;16:329–343. [PubMed]
  • Yeo C, Whitman M. Nodal signals to Smads through Cripto-dependent and Cripto-independent mechanisms. Mol Cell. 2001;7:949–957. [PubMed]
  • Ying QL, Stavridis M, Griffiths D, Li M, Smith A. Conversion of embryonic stem cells into neuroectodermal precursors in adherent monoculture. Nat Biotechnol. 2003;21:183–186. [PubMed]
  • Young RA. Control of the embryonic stem cell state. Cell. 2011;144:940–954. [PMC free article] [PubMed]
  • Zorn AM, Wells JM. Vertebrate endoderm development and organ formation. Annu Rev Cell Dev Biol. 2009;25:221–251. [PMC free article] [PubMed]