Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2017 Nov 7.
Published in final edited form as: Dev Cell. 2016 Oct 13;39(3):302–315. doi: 10.1016/j.devcel.2016.09.016

A balance between secreted inhibitors and edge-sensing controls gastruloid self-organization

Fred Etoc 1,2, Jakob Metzger 1,2, Albert Ruzo 2, Christoph Kirst 1, Anna Yoney 1,2, M Zeeshan Ozair 2, Ali H Brivanlou 2,*, Eric D Siggia 1,*
PMCID: PMC5113147  NIHMSID: NIHMS818336  PMID: 27746044

Summary

The earliest aspects of human embryogenesis remain mysterious. To model patterning events in the human embryo we used colonies of human embryonic stem cells (hESCs) grown on micropatterned substrate and differentiated with BMP4. These gastruloids recapitulate the embryonic arrangement of the mammalian germ layers and provide an assay to assess the structural and signaling mechanisms patterning the human gastrula. Structurally, high-density hESCs lateralize their TGF-β receptors to their lateral side in the center of the colony, while maintaining apical localization of receptors at the edge. This relocalization insulates cells at the center from apically applied ligands while maintaining response to basally presented ones. Additionally, BMP4 directly induces the expression of its own inhibitor, Noggin, generating a reaction-diffusion mechanism that underlies patterning. We develop a quantitative model that integrates edge sensing and inhibitors, to predict human fate positioning in gastruloids, and potentially the human embryo.

Graphical abstract

graphic file with name nihms818336f8.jpg

eTOC

In the embryo, cell-fates are specified by a combination of chemical and physical factors. Using an in vitro model for human gastrulation, Etoc et al. show that a complex developmental transition can be reduced to two independent inhibitory mechanisms linked to differential cell polarization and the diffusion of NOGGIN.

Introduction

The gastrulating embryo is a remarkable example of self-organizing system: from a seemingly homogeneous epiblast layer, cells are allocated into the three germ layers as the body plan unfolds. There is a complex interplay between geometry and signaling. At the structural level, early mammalian embryos share common features despite differences in shapes. The epiblast is always directly juxtaposed to the visceral endoderm. The apical surface of the epiblast surrounds the amniotic cavity while the apical surface of the visceral endoderm borders the yolk sack. At the signaling level, studies in the mouse have established that a BMP4 signal from the extraembryonic-embryonic boundary initiates a positive feedback loop through Wnt and Nodal (Arnold and Robertson, 2009; Nowotschin and Hadjantonakis, 2010; Stephenson et al., 2012). Both inductive and inhibitory signals are required for patterning, and they reside in specific tissues. There are three major questions that need to be addressed in this regard: how signaling is coupled to embryo geometry, how signals move through and between tissue types, and how they are selectively targeted.

At the level of fate determination, the molecular basis of patterning embryonic tissues with sharp boundaries has been the subject of intense scrutiny in model systems for decades. Morphogens play a dominant role and are able to induce different fates based on their concentration, and dynamics of presentation (Warmflash et al., 2012). It is still unclear how morphogens and inhibitors interact with geometrical factors to create spatially organized differentiation domains.

These questions are impossible to study in humans and technically difficult to address in the mouse, since gastrulation occurs just after implantation. Therefore, there is a need for the development of in vitro assays reconstituting the spatial arrangement of human embryonic germ layers. Mouse embryonic stem cells aggregates were shown to spatially segregate germ layer populations under specific differentiation conditions (Poh et al., 2014; van den Brink et al., 2014). However, these structures show tremendous variability in size and shape, which prevented a precise dissection of the mechanism underlying their self-organization. We recently devised an assay for differentiating human embryonic stem cells (hESC) into gastruloids: micropattern colonies that recapitulate the spatial arrangement of germ layers (Warmflash et al., 2014). Our technique provides a simple entry point to study pattern formation and is amenable to mechanistic investigations since micropatterns can be easily imaged and analyzed at the single cell level. We can directly access quantitative measurements of signaling and fate-specification dynamic that cannot be performed in embryos. In this study, we unveil the mechanism of germ layer positioning in vitro and address to what extent observations made in model systems are relevant to fate determination in hESCs and possibly in the human embryo.

In our approach, cells are confined to disks of 500µm diameter and differentiated with BMP4 for 42 hours. From edge to center, trophectoderm, endoderm, mesoderm, and ectodermal fates are specified in a radially symmetric pattern (Warmflash et al., 2014). The mesendodermal fates arise, as they do in the primitive streak, by an epithelial-to-mesenchymal transition (EMT). Surprisingly, gradual reduction of the colony diameter selectively eliminated the “center” fates. This demonstrated that hESCs establish their fate by measuring their distance from the edge. How human cells measure their distance from the edge with such a dramatic consequence for fate remains completely unknown.

We demonstrate that the morphology of pluripotent hESCs in micropatterned colonies varies systematically with density and radial position. The colonies are polarized epithelia with the apical side facing the media. At high densities, TGF-β receptors relocalize from the apical to the lateral side of the cells, which become insensitive to TGF-β signaling. Cells at the edge of the colony do not undergo subcellular receptor lateralization and are thus sensitive to TGF-β morphogens. This establishes how geometry imposes a pre-pattern and how colony boundaries are distinguished. In addition, we found that BMP4 directly induces the expression of its own inhibitor NOGGIN, thereby defining a classical reaction-diffusion system (Meinhardt, 1982). This is supported by the observation that NOGGIN is both a necessary and sufficient chemical signal to explain aspects of the radial pattern of BMP4 signaling and subsequent fate determination due to cell-cell communication.

An important consequence of our work is the ability to predict fate patterning in a wide range of conditions. This is accomplished through a quantitative model that decomposes the spatio-temporal signaling dynamics into two independent modules: (i) a pre-pattern in the pluripotent state, and (ii) the spreading of inhibitor by diffusion constrained by geometry. These two modules are parameterized by population-averaged measurements at different times and can be joined with only one free parameter. Our framework decomposes a complex space-time gene network into several invariant components that can be measured and quantified in descriptive terms and then merged together. We believe that approaches such as ours contribute to the basic understanding of human development as well as providing the necessary framework to realize the potential of hESC for regenerative medicine.

Results

The differentiation of 500µm diameter disk-shaped hESC micropatterns, with a starting density around 2500 cells/mm2, leads to radial organization of the germ layers after 42 hours of 50ng/ml BMP4 exposure (Figure 1A). This defines our set of standard conditions for differentiation of hESCs into gastruloids.

Figure 1. Spatial organization of the response to TGF-Beta ligands at short time scales.

Figure 1

Open in a new tab

(A) Gastruloid are hESC micropattern colonies differentiated with a high level of BMP4. They yield rings representative of extraembryonic fates and the three germ layers, see (Warmflash, 2014) for additional markers. Top: data, Bottom: scheme (B) Four colonies of different densities were stimulated with 5ng/ml BMP4 for one hour and subsequently stained for pSMAD1. Bottom line: the pSMAD1 response was sufficiently binary that individual nuclei were classified as positive (red dots) or negative (black dots). (C) Top: equation used to fit the fraction of pSMAD1 positive nuclei as a function of radius, for various densities and BMP4 concentrations (Hill coefficient of 1.4). Sensitivity to BMP4 is parameterized as Kmp. Bottom: value of Kmp as a function of colony density and radial position. 1250 colonies. All scale bars: 100µm.

Cell Density Restricts Spatial Response to TGF-β Ligands

First we examined the immediate response of single pluripotent hESCs to different concentrations of BMP4. The levels of nuclear pSMAD1, which transduces signals on behalf of BMP4, were quantified after one hour of BMP4 treatment. We find that the response was sigmoidal with an inflection point at 0.1ng/ml, establishing the sensitivity of hESCs to BMP4 in unstructured colonies (Figures S1A–B).

We then asked whether hESCs would maintain a similar sensitivity to BMP4 when cultured at different densities on micropatterned colonies. We found that, for a fixed BMP4 concentration, pSMAD1 levels were uniform at low densities, but surprisingly became progressively restricted to the edge of the colony as the density increased (Figure 1B). The same effect was also visible in unstructured colonies and was insensitive to Cycloheximide, establishing that new protein synthesis was not required (Figures S1C–D). Moreover, 1-hour responses were homogenized when performed in calcium free medium, which leads to disruption of tight-junctions, suggesting that epithelial integrity is essential for the establishment of the pattern (Figure S1E). In every BMP4 stimulated colony, nuclear pSMAD1 distribution was bimodal and the area occupied by responsive cells was a decreasing function of the cell density (Figure S1F–S2A). We then quantified sensitivity towards BMP4 as a function of cell density and radial position in micropatterns (Figure 1C). Sensitivity was defined as Kmp: the BMP4 concentration required to activate half of the cells at a given radius, for colonies of a given density. Cells at the edge of the colony always maintained a high sensitivity to BMP4 regardless of the density, as indicated by the low Kmp (<0.5ng/ml). However, cells at the center, located further than 50µm away from the edges, gradually lost sensitivity towards BMP4 in a density-dependent manner (Figure 1C). This establishes that both cell density and radial position influence BMP4 response. Similar effects were observed with the SMAD2/3 branch of the TGF-β pathway, as probed by Activin stimulation (Figure S2B–C–D–E–F). We conclude that a density dependent-inhibition of both branches of the TGF-β pathway is occurring at the micropattern center.

Differential Receptor Relocalization Unveils a Pre-pattern in Pluripotent hESC Colonies

hESCs are of epithelial character (Krtolica et al., 2007). Recent studies about TGF-β signaling in epithelial cells have shown density-dependent TGF-β inhibition due to crosstalk with the HIPPO pathway (Varelas et al., 2010) or receptor localization to the baso-lateral surface of epithelia (Nallet-Staub et al., 2015).

Analysis of the localization of the hippo effector YAP/TAZ in pluripotent colonies demonstrated a loss of nuclear localization as a function of increasing density. This was largely unaffected with 1 hour of BMP4 treatment, and resulted in the nuclear localization of YAP/TAZ in only a few peripheral cells at high densities (Figure S2G–H). The YAP/TAZ profiles didn’t correlate with the early pSMAD1 responses as shown in our Figure 1C, as high density colonies always present a ~ 40 µm ring of pSMAD1 activity when presented with BMP4. Therefore Hippo is unlikely to control early pSMAD1 responses.

In the Hippo independent mechanism, TGF-β inhibition was attributed to receptor localization to the baso-lateral surface of epithelia. When ligands are delivered apically, they cannot contact the receptors that are sealed off by the tight junctions. We therefore asked if a similar mechanism was operating in hESC colonies. We first used RNAseq to identify the TGF-β receptors expressed in pluripotent hESC grown on micropatterns (Figure S3A). The most highly expressed type I and II receptors, BMPRIA, BMPRII, ActRIB, ActRIIB, were epitope-tagged and expressed under Doxycycline (Dox) control (Figure S3B). Immunofluorescence was used to detect the receptors, the tight junction marker ZO-1, an apical surface marker (WGA), and the nucleus (Figure 2A). Surprisingly, only cells on the colony edge had receptors localized to the portion of membrane facing the extracellular medium, while cells in the center showed a subcellular relocalization of their receptors to the lateral side, below the tight junctions (Figures 2A–B). This unexpected finding establishes the existence of a pre-pattern in pluripotent hESC grown on micropatterns.

Figure 2. Microcolonies are polarized epithelia with laterally positioned receptors.

Figure 2

Open in a new tab

(A) The Activin and BMP receptors expressed in high density colonies were epitope tagged and transiently induced for 6 hours with Doxycycline (Methods). They show apical localization at colony edges but lateral expression elsewhere. Samples were stained for tight junctions (ZO-1, red), apical membranes (wheat germ agglutinin, green), DAPI in blue and the indicated receptors in white. (B) Violin plots for the distance histograms of receptors and nuclei relative to the apical surface defined by ZO-1 show lateral localization. (C) Cells expressing a membrane localized fluorescent protein (red) as well as a YFPH2B fusion protein (green) were diluted 1:10 with parental RUES2 cells. In high-density colonies, we observe a radial elongation of the cells at the periphery, while the center cells are randomly oriented. Scale bar: 100µm. (D) Different morphology of cells at the periphery and center of a high-density colony. Membrane is red, nuclei green and centrioles white. The x–y plots are maximum z-projections. The heavy dashed line is the colony border, and the light dashed line defines the vertical slice shown below with z-up.

In order to precisely examine cell polarity as a function of position within the micropattern, we created a cell line expressing nuclear, membrane, and centriole markers (RUES2-NMC, Figure S3C). All cells in high-density colonies had apically localized centrioles (Figure S3D). Moreover, cells at the edge displayed a radially elongated shape with their basal side at the colony periphery and their apical side facing inward (Figure 2C–D-S3E). Cells away from the periphery had fewer extensions and positioned their centriole above the nucleus (Figure 2D). Low-density colonies retained ZO-1 staining but receptors distributed uniformly over all surfaces (Figure S3F–G–H). The length-scale of apically available receptors at the edge of high-density colonies was around 40–50 µm (Figure 2A), and was comparable to the extension of the region at the colony edge in which cells keep high sensitivity to BMP4 (Figure 1C). Therefore the inaccessibility of receptors in the colony center could explain the differential radial response to TGF-β stimulation shown in Figure 2.

Lateralization of TGF-β Receptors Leads to Polarity-Selective Signaling Response

To establish the functional consequences of baso-lateral receptor localization in our hESC, we grew cells on permeable filters that allowed ligand stimulation from both the basal and apical side. Trans-epithelial resistance across the hESC layer increased with density and reached values as high as 515Ω/cm2 (Figure S3I) demonstrating that hESC are indeed a complete epithelium sealed by tight junctions (Claude, 1978). BMP4 or Activin were presented selectively from either the apical or basal side. Low-density cultures on filters responded equally to both apical and basal ligand presentation. In contrast, in high-density cultures, cells were responsive to ligands only when presented from the basal side (Figure 3A–B). Moreover, homogeneous activation was obtained even though the YAP/TAZ was mostly nuclear-excluded, further excluding the contribution of the HIPPO pathway in the early TGF-β responses (Figure S2I–J). This establishes the functional consequences of TGF-β receptors lateral localization for signal transduction. In response to apically applied BMP4, cells on the filters mimicked those in the center of the micropatterns as a function of density (Figure 3C–D). Cells at the colony edge are always responsive due to the presence of available receptor at their surface facing the medium, for all densities.

Figure 3. Cells grown on filters show different sensitivites to apical vs basal ligands.

Figure 3

Open in a new tab

(A) Cells grown in transwell plates to high density do not respond to apically applied BMP4. A high dose of BMP4 (50ng/ml) is presented for one hour to the top or bottom compartment and the response is probed by staining against pSMAD1. (B) The same asymmetry in the response of transwell colonies is seen with 100ng/ml Activin stimulation. Low density is 1200 cells/mm2, and high density 7000 cells/mm2. (C) The apical response to BMP4 stimulation for one hour diminishes with increasing density. (D) Three colonies of increasing density were stimulated as in panel C and show the same patchy response in colony center for comparable densities. (E) A calcium depletion shock is applied to micropatterns, followed by our differentiation protocol in presence or absence of Rock Inhibitor (RI, 10µM). In presence of RI, the patterning collapses due to loss of tight-junction integrity. All scale bars are 100µm.

We then interfered with the tight-junctions integrity during our differentiation protocol by a brief calcium depletion followed by incubation in Rock-inhibitor, conditions known to prevent building of tight junctions (Walsh et al., 2001). This led to a loss of the patterning (Figure 3E). Conversely, neither of the two treatments alone destroyed the radial organization of the fates. Taken together, our data indicates that immediate response to TGF-β ligands is determined by receptor accessibility, which is critical for fate patterning.

Additional Negative Feedbacks Mechanisms are Necessary to Explain Long-Term SMAD1 Dynamics

We have previously described the gradual restriction of BMP4 signaling at the micropattern edge over time, during the 42-hour differentiation window (Warmflash et al., 2014). We explored whether receptor lateralization alone was sufficient to explain the pSMAD1 dynamics over longer time scales: as cells proliferate, density increase would lead to progressive receptor relocalization at the baso-lateral side and therefore gradual exclusion of pSMAD1 from the center as quantified in Figure 1C. We compared micropatterns stimulated for 1 or 24 hours with BMP4, at similar final densities in the range of 4000 to 7000 cells/mm2, and compared the resulting pSMAD1 profiles (Figure 4A–B). In this density regime, the center of the colony retains a sufficient density of apically-available receptors (Figure 1C) that a 1 hour 50ng/ml BMP4 stimulation results in a homogeneous pSMAD1 profile. In contrast, we observed that, at 24 hours, the mean pSMAD1 coverage was reduced compared to the 1-hour time point (Figure 4A–B). This suggests the induction of a negative feedback in the BMP4 signaling network, which takes place over longer time scales (> 1hour), and reduces BMP signaling in the center.

Figure 4. BMP4-induced negative feedback by NOGGIN is necessary for fate patterning.

Figure 4

Open in a new tab

(A) pSMAD1 stainings of colonies stimulated for 1 or 24 hours with 50ng/ml BMP4. Only micropatterns with final density falling in the 4000 to 7000 cells/mm2 were considered. (B) pSMAD1 radial profiles for colonies shown in panel A. n> 20 for each condition. (C) RNA-Seq profiling of colonies stimulated for 0, 4, 12 and 24 hours with 50ng/ml BMP4, showing the production dynamics for the principle BMP secreted inhibitors. (D) NOGGIN induction was probed by qPCR in unstructured small colonies after 4 hours stimulation, and reported as fold change compared to un-stimulated samples. The conditions collectively suggest NOGGIN as a direct BMP target; CHX cyclohexamide, SB and IWP2 respectively inhibit Activin/Nodal and Wnt signaling. (E) 4 hours stimulation of mouse epiblast converted stem cells with 50ng/ml BMP4 leads to strong up-regulation of ID1 but not NOGGIN, while hESCs show induction of both genes. (F) pSMAD1 profiles in colonies after 24 hours stimulation with 50ng/ml BMP4 and with densities falling in the 6000 to 7000 cells/mm2 range. The NOGGIN knockout line (bottom) shows complete penetration of the signal into the colony. (G) Quantification of the fraction of pSMAD1 positive cells in the experiments shown in panel F (n=7 for each condition). More colonies are shown in Figure S5C. (H) Fate pattern after 42 hours of BMP4 stimulation, 50ng/ml, showing elimination of the center, Sox2 fate, when NOGGIN is knocked out. In the NOGGIN −/− condition, the center fate is then Cdx2+, while WT colonies show a central Sox2+ domain. Bottom: Quantification of the fate acquisition into Cdx2, Bra and Sox2 positive cells, at the single cell level for the experiment presented in G. Yellow curves represents wild type colonies (n=7), red the NOGGIN−/− colonies (n=16). All scale bars are 100µm.

NOGGIN is a Direct BMP4 Target in hESCs

Since this type of graded and dynamic signaling has been extensively linked to interaction between morphogens and inhibitors in model organisms, we investigated the expression of the BMP4 inhibitors in our system. We performed RNA-seq analysis on colonies stimulated for 0, 4, 12 and 24 hours with 50ng/ml BMP4 and profiled the expression of BMP4 secreted inhibitors (Figure 4C). Among the different inhibitors, we found that NOGGIN is the earliest and most strongly up-regulated at 4 hours (Figure S4A). qPCR experiments validated the results obtained by RNAseq, confirming that NOGGIN is indeed a BMP4 target in hESCs (Figure 4D). Addition of the protein synthesis inhibitor Cycloheximide (CHX) did not inhibit induction, suggesting that NOGGIN is a direct immediate early target of BMP4. Consistently, the SMAD2/3 pathway inhibitor SB-431542, or the Wnt inhibitor IWP2 on top of BMP4, did not have any effect on NOGGIN induction. Interestingly, mouse epiblast cells, considered as the developmental equivalent of hESC, did not up-regulate NOGGIN following the BMP4 stimulation, while the BMP4 target ID1 was strongly induced (Figure 4E and S4B–C). We conclude that NOGGIN is directly induced by BMP4 specifically in hESCs and not in mouse EpiSCs.

NOGGIN is Necessary for Spatial Pattern Formation in standard differentiation conditions

To address the function of NOGGIN in spatial patterning of the micropatterns, we created two homozygous NOGGIN knockout lines (RUES2-NOGGIN−/− c1 and c2, Figure S4D). These lines showed normal pluripotent behavior, demonstrated intact tight junctions, and presented the same polarized response to BMP4 when grown on transwells (Figure S4F–H). We analyzed colonies with densities below 7000 cells/mm2 so that cells in the center remained sensitive to BMP4 at the high concentrations employed. After 24 hours in BMP4, while wild-type colonies showed a restriction of BMP4 signaling to the colony edge, we found that deletion of NOGGIN led to spatially uniform pSMAD1 profiles (Figure 4F–G, S4E–G). This demonstrates that NOGGIN is necessary for the proper establishment of signaling asymmetries in our system.

We next examined the emergence of discrete fates in RUES2-NOGGIN−/−. Using our standard protocol for gastruloid differentiation, we found that the central SOX2+ ectodermal domain in the RUES2-NOGGIN−/− lines was lost and replaced by a mixture of CDX2+ cells and patches of BRA+ cells (Figure 4H, Figure S4I). Thus, in our standard set of differentiation conditions, NOGGIN is necessary for proper patterning of cell fates.

NOGGIN Expression in Microcolonies Follows pSMAD1 Profiles

Our observation that BMP4 directly induces NOGGIN is reminiscent of the reaction-diffusion schemes postulated to be operating in model systems during early embryonic development to establish patterns through Turing mechanisms (Meinhardt, 1982). In order to characterize NOGGIN activity, we first established its expression profiles on micropatterns by performing in situ hybridization (Figure S5A–B). We compared early pSMAD1 responses and NOGGIN induction profiles for different concentrations of BMP4 and at similar density ranges. pSMAD1 profiles were restricted at the edge at low concentrations and spatially homogeneous for high concentrations (Figure 5A–D), as quantified in Figure 1C. Consistently with NOGGIN being a direct transcriptional target of BMP4, the NOGGIN in situ profiles matched the pSMAD1 profiles for both low and high BMP4 concentrations and were edge-dominated at low concentrations and spatially homogeneous at high concentrations (Figure 5B–C). Therefore the spatial profiles confirm the direct link between BMP4 and NOGGIN production established in Figure 4D.

Figure 5. NOGGIN induction in micropatterns is sufficient to restrict pSMAD1 signaling at the edges.

Figure 5

Open in a new tab

(A) The response to one hour BMP4, 2.5ng/ml or 50ng/ml. The pSMAD1 response is restricted at the edge for low concentrations and spatially homogeneous for high concentrations. (B) In situ hybridization for Noggin mRNA in response to 6 hours BMP4 stimulation at 2.5 and 50ng/ml. Each detected mRNA molecule is represented as a red dot. NOGGIN mRNA is localized at the edge for low concentrations and becomes spatially homogeneous for high concentrations (C) Number of NOGGIN mRNA detected as a function of the radial distance for colonies stimulated for 6 hours with 0, 2.5 and 50 ng/ml (n=8 per condition) as shown in B. (D) Quantification of the fraction of pSMAD1 positive cells as a function of the radial distance for 1 hour stimulation with 2.5 or 50 ng/ml as shown in A (n>34 for each condition). In panels A to D, colonies are within the 4000–7000 cells/mm2 range. Standard deviations are plotted as shaded error bars. (E) RUES2-ePB-Dox-NOGGIN cells expressing Noggin under the control of Dox are induced for 6 hours (bottom) or left in control medium (top) and subsequently stimulated with BMP4 for one hour. Noggin expression restricts the pSMAD1 response to the boundary. (F) Quantification of the fraction of pSMAD1+ cells in the experiment presented in panel E. Colonies are selected within a similar density range of 4500 to 5000 cells/mm2 (n=8). (G) The data in panel E were repeated using different ratio of secreting cells diluted with wild-type RUES2 (100%, 33%, 10%, 3%, 1% and 0.5%) in colonies of similar densities to E–F. Data are in black and the fit using the model presented in H–I is in red. (H) Noggin profile in panels result from a fixed production term p, diffusion in two dimensions with N=0 on the colony boundary and an homogeneous degradation term λ. Left: residuals of the fit of data in panel J for different values of D and λ. Right: Noggin concentration profile in the microcolony geometry after 6 hours Dox induction with D=10µm2/s and λ=0.001s−1. (I) Qualitative explanation of the data: pSMAD1 gets activated when the free BMP4 concentration (red) is higher than the sensitivity curve Kmp (blue) from Figure 1C. Therefore there is pSMAD1 exclusion in the center only in the presence of a NOGGIN profile (green) peaking at the center and lower at the edge.

NOGGIN Expression in hESC Microcolonies Restricts BMP Signaling to the Edges

To best quantify the effects of Noggin on the microcolonies we wanted to control its expression directly. To this end, we generated a transgenic RUES2 line that expresses NOGGIN under the control of a Doxycyclin inducible promoter (RUES2-ePB-Dox-NOGGIN). This decouples NOGGIN activity from BMP4 induction and makes it spatially uniform. RUES2-ePB-Dox-NOGGIN cells were cultured on micropatterns, NOGGIN was induced for 6 hours and followed by one hour BMP4 treatment to examine the status of BMP4 signaling. We observed a tight restriction of pSMAD1 to the colony edges, while non-stimulated colonies of similar low densities showed homogeneous pSMAD1 levels (Figure 5E–F). In order to assess NOGGIN activity at lower expression levels, we mixed varying proportions of a RUES2-ePB-Dox-NOGGIN with RUES2. As the number of NOGGIN producing cells decreased, the pSMAD1 domain at the edge moved progressively inwards (Figure 5G, S5C). Therefore NOGGIN by itself is also sufficient to restrict pSMAD1 signaling to the colony periphery.

Quantitative Modeling Demonstrates a Dome-Shaped Inhibitory Profile for NOGGIN

It is striking that ubiquitous expression of NOGGIN by RUES2-ePB-Dox-NOGGIN results in a spatial asymmetry of BMP4 signaling. This implies a stronger inhibition of BMP4 by NOGGIN at the center of the micropattern (Figure 5E–F–G). We explored computationally the potential mechanisms of NOGGIN transport that could lead to the restriction of BMP signaling at the edge. We assumed NOGGIN production began when Doxycycline was added and continued at a fixed rate. NOGGIN transport was modeled as 2D diffusion on the colony surface with a diffusion coefficient D (Figure 5H). Loss of NOGGIN can come from degradation, endocytosis or loss in the medium above the colony surface. This loss is represented by a homogeneous degradation rate λ. We imposed the NOGGIN concentration outside the micropattern to be zero since molecules leaving the colony at the edge are likely to be lost in the 3D medium. In order to translate the NOGGIN profile into a spatially modulated BMP4 signaling activity, we modeled the reduction in free BMP4 levels by Noggin by pairwise binding (Zimmerman et al., 1996), which is incorporated into the single free parameter of the model (SI). Finally, the density-dependent effects of receptor lateralization on BMP4 signaling were taken into account and modulated sensitivity towards BMP4 as a function of radius as presented in Figure 1C. We can now evaluate the model predictions by fitting the pSMAD1 profiles obtained experimentally with different ratios of NOGGIN producing cells (Figure 5G).

We quantified the performance of our model for varying values of D and λ over multiple orders of magnitude. Higher diffusion coefficients (>10µm2/s) and lower λ best fit our experimental observations (Figure 5G–H). This led to a dome-shaped NOGGIN concentration profile (Figure 5H, right). Our model recapitulated quantitatively the low dependence in the NOGGIN production strength: Noggin levels were varied by 200 folds, which resulted only in a small ingression of the pSMAD1 domain (Figure 5G). In conclusion, a model of NOGGIN with fast diffusion through the colony recapitulated the observed inhibition patterns of BMP4 at the colony level. In the fast-diffusion regime, the actual values of λ only slightly changed the fit quality (SI Figure M1).

There is a subtle interplay between Noggin diffusion and the radial profile of cell sensitivity sets by the receptors. At the micropattern density in Figure 5H–I, cells are very sensitive to BMP4 at the edge (Kmp<0.5ng/ml) and less sensitive at the center (5ng/ml<Kmp<10ng/ml, Figure 1C). A 1-hour stimulation using 50ng/ml BMP4 is thus sufficient to activate BMP signaling in every cell (Figure 5I, top). As for Docycycline induction, transport of NOGGIN results in a concentration profile peaking at the center and low at the edge. NOGGIN locally inhibits BMP4 and reduces its effective concentration. The free BMP4 is now elevated at the edge where it is high enough for pSMAD1 activation, while lower than Kmp at the center, resulting in pSMAD1 exclusion from the center (Figure 5I, bottom).

Receptor Lateralization and Noggin Induction are Sufficient to Recapitulate Signaling Dynamics Within 24 Hours Differentiation

We next asked if the combination of receptor lateralization and the intrinsic NOGGIN induction by BMP4 is sufficient to recapitulate the pSMAD1 spatio-temporal dynamics within the first 24 hours of gastruloid differentiation. We observed that receptor lateralization at the colony center was maintained after BMP4 presentation (Figure S6A–B–C) when compared to unstimulated colonies at higher densities. We thus hypothesized that the impact of receptor localization on BMP4 signaling, as characterized in pluripotency conditions (Figure 1C), was unchanged during the first 24 hours of differentiation. The other critical inputs to the model that were all measured are (i) the averaged NOGGIN production from our RNA-seq data (Figure S6D), (ii) proliferation rate (Figure S6E–F), which changes cell density by about 3-folds over 24 hrs and modulates the receptor influence on the cell sensitivity profile. We use the same model for NOGGIN dynamics as developed in Figure 5J–K–L, setting D to 10µm2/s and λ to 0.001 s−1, with the assumption that NOGGIN is produced locally and proportionally to pSmad1 activity. For our standard density the model matched the data for the 24 hour time course, Figure 6A.

Figure 6. Modeling pSMAD1 spatio-temporal dynamics at long time scales by superposing Noggin induction and density dependent receptor relocalization.

Figure 6

Open in a new tab

(A) (top) pSMAD1 spatial profiles at 1, 6, 12 and 24 hours after BMP4 presentation, starting from a density in the 2200–2700 cells per mm2 range at t=0, and terminating with a density in the 5900–7200 cells per mm2 range at t=24hrs. The model is used with D=10µm2/s and λ=0.001s−1. (B) Decomposition of the model: for each time point, we show the sensitivity curve, Kmp (blue), the NOGGIN profile (green) and the reduced BMP4 concentration (free BMP, yellow). pSMAD1 is activated only where the free BMP4 concentration is above the Kmp curve. (C) Left: data for pSMAD1 profiles vs radius at different times (1, 6, 12 and 24 hours) with increasing starting densities.

We can rationalize how NOGGIN induction and progressive receptor relocalization are combined at the colony level and result in the observed pSMAD1 spatio-temporal dynamic during gastruloid formation (Figure 6B). At early times, BMP4 concentration is high enough to overcome edge-to-center differences in sensitivity so that every cell gets activated, which results in homogeneous pSMAD1 profiles (Figure 6B, 1 hour time points) and therefore homogeneous NOGGIN production. Diffusion then establishes the dome profile exactly as reported previously in Figure 5H. At longer times, NOGGIN induction reduces the effective BMP4 concentration on the micropattern (“free BMP” curves). NOGGIN inhibition is stronger at the center, where cell sensitivity is also the lowest. Therefore, the effective BMP4 concentration can locally drop below the activation threshold of individual cells, giving rise to pSMAD1 negative domains at the center (Figure 6B, 12 and 24 hours time points). NOGGIN is now expressed from the edge but its concentration is homogenized in the center due to fast diffusion.

To further validate the model, we fitted with no additional parameters the pSMAD1 spatio-temporal dynamic after BMP4 presentation in micropatterns with a 10× range of starting densities. Increasing density or increasing time have comparable effects of limiting pSMAD1 to the boundary, but the former variable operates through receptor localization, while increasing induction time implies a level of Noggin production. The two effects are coupled since time implies proliferation. Their combined activity is not simply additive and requires a quantitative model, which explains the pSMAD1 profiles over a wide range of induction times and densities (Figure 6C and SI).

Receptor Lateralization and Noggin Induction are Sufficient to Predict Fate Positioning in a Wide Range of Conditions

We finally addressed to what extent receptor lateralization and NOGGIN induction determine cell fate determination and positioning. We started by modulating the initial density of cells differentiated on transwells. When applied basally, BMP4 led only to the induction of CDX2+ trophectodermal cells only, regardless of the starting density (Figure 7A). However, when BMP4 was applied apically, we observed a spatially mottled transition from CDX2+ to BRA+ to SOX2+ cells (Figure 7A). At the highest densities, cells retained pluripotent fates as shown by NANOG and SOX2 expression (Figure S7A). We then examined fates as a function of density in micropatterns. In order to link starting density with fate positioning, we generated a cell line constitutively expressing a nuclear marker (RUES2-H2B-citrine). Colonies were imaged just before BMP4 presentation to establish their starting cell density by counting nuclei. After 42 hours of BMP4 presentation, we measure the positions of the CDX2+ and BRA+ domains within micropatterns. At the lowest densities, all cells were CDX2+. As cell density increased, a BRA+ domain emerged at the center. At higher densities the BRA+ domain was pushed outward. Finally, at the highest densities tested, the BRA+ and CDX2+ domains overlapped at the edge (Figure 7B, S7B). Therefore, density is also a key regulator of fate patterning.

Figure 7. Fate patterning as a result of density-dependent receptor relocalization and Noggin production.

Figure 7

Open in a new tab

(A) Fate acquisition after 42 hours of BMP4 treatment (50ng/ml) in transwell plates as a function of density at the time of stimulation and the side of ligand presentation. The transition in fate with density follows the trend observed for pSMAD1 in Figure 3C. (B) Colonies of cells expressing a nuclear marker (Citrine-H2B) were imaged at the time of stimulation to assess the starting density. Cells were then returned to the incubator and left to differentiate for 42 hours in 50ng/ml BMP4. Colonies were then fixed and stained for Cdx2 and Bra and sorted by their initial density. At low density, colonies are uniform Cdx2+, there are distinct Bra+ and Cdx2+ territories at intermediate density, and complete overlap at high densities. (C) Fate patterning obtained as a function of the starting density and the relative Noggin levels in the system: compared to the wild type situation, Noggin can be either eliminated (Noggin−/− cell line) or boosted by over-expression (RUES2-ePB-Dox-NOGGIN). Three regimes of densities were obtained by plating 200 000 cells (low), 550 000 cells (medium) or 850 000 cells (high) on individual chips, followed by stimulation 24 hours later. (D) Left: pSMAD1 and SMAD2 profiles after 24 hours BMP4 50ng/ml were simply thresholded to define fates for four different starting densities as noted for the corresponding colony after 42 hours differentiation with the same color code (right). The domains are defined as: green (S1>0.6 and S2>0.45), red (S1<0.6 and S2>0.45), blue (S1<0.6 and S2<0.45). (E) Prediction for pSMAD1 and SMAD2 using the model presented in figure 5G and Methods. The nine conditions parallel those shown in C. Overlayed on the SMAD profiles, we define blue, red and green domains based on the same criteria as in panel D showing good correspondence with the experiments under the matrix of conditions. All scale bars are 100µm.

To further probe the effects of NOGGIN on fate patterns, we used RUES2, RUES2-NOGGIN−/−, and RUES2-ePB-Dox-NOGGIN, plated at three different densities. We examined the expression patterns of CDX2, BRA, and SOX2 in cells differentiated with high BMP4 concentrations (Figure 7C and S7C). For fixed plating density increasing NOGGIN levels led to an expansion of the SOX2+ and shrinkage to the edges for CDX2+ and BRA+ domains. While NOGGIN was necessary for proper patterning in the standard conditions (RUES2, middle density), NOGGIN knock-out could be compensated by a stronger influence of the receptor relocalization effect (RUES2-NOGGIN−/−, high-density). Conversely, in the low-density regime, where receptors are available apically, boosting the NOGGIN levels restored spatial patterning (RUES2-ePB-Dox-NOGGIN, low-density). The matrix of conditions presented in Figure 7 therefore demonstrates how different patterning outcomes are obtained by independently tuning two key parameters, NOGGIN levels and receptor relocalization.

We next asked if our model could account for the relative positioning of fates obtained in the diverse conditions of Figure 7. In this regard, SMAD2 signaling is necessary for mesendoderm fates and must be considered with pSMAD1. We observed a strong correlation between the SMAD1/2 profiles measured at 24 hours and the fate-territories at 42 hours (Figure 7D). It was therefore natural to define fate territories by imposing a threshold on the SMAD1/2 levels to make them binary and to define regions (from inside out) as; (SMAD1 off, SMAD2 off) blue, (SMAD1 off, SMAD2 on) red, (both on) green. There are of course complex gene regulatory networks that intervene between the SMAD signaling and the fate markers, as well as Wnt signaling among others. However, we investigated if a minimal definition of the fate domains based on SMAD1/2 levels would be sufficient to predict patterning outcomes.

To model the activation of SMAD2, we assumed Nodal was produced locally by pSMAD1, diffuses in the layer, decays, and is lost from the boundaries (see SI). This simple implementation was able to reproduce the SMAD2 profiles at 24 hours (Figure S7D). Moreover, when we generated signaling profiles at 24 hours for the nine conditions of Figure 7C, we were able to produce signaling activities and fate domains that qualitatively matched the spatial patterns observed experimentally (Figure 7E). Our model is also consistent with the previous observation (Warmflash et al., 2014) that gradual reduction of the colony diameter selectively eliminates the “center” fates (Figure S7E) and can predict fate positioning in micropattern of different radii. Taken together, our model captured the main features of the signaling profiles and the resulting fate choices over a wide range of conditions.

Discussion

Self-organization of differentiation patterns within multicellular systems is the hallmark of early development, as observed in vivo or recapitulated in vitro. However, the rules that govern the emergence of complex multicellular structures from homogeneous populations are largely unknown (Sasai, 2013). Complex sets of interactions between cells are coupled to intracellular regulatory networks and make a detailed understanding of the process difficult. Instead, we need to reduce system complexity by identifying its core regulatory modules. Decomposing pattern formation into a limited number of connected modules allows understanding normal and pathological development, as well as identifying the key nodes that direct the spatial arrangement of differentiation patterns into precisely controlled structures.

Here, we have studied self-organization of differentiation patterns in an in vitro model for human gastrulation. The first step was to understand the spatio-temporal dynamics of BMP4 signaling. There are two distinct mechanisms that control BMP4 signaling in the micropatterns: density-dependent receptor relocalization already present in the pluripotent state, and secreted Noggin inhibitor, induced by BMP4. Together they shape pSMAD1 signaling profiles as a function of time and cell density. We showed that with a single adjustable parameter one could recapitulate their combinatorial inhibition on BMP4 signaling and explain the pSMAD1 spatio-temporal dynamics in a wide range of conditions. There is a substantial radial dependence in each of the two effects and their combined outcome is nonlinear and impossible to predict without mathematical modeling.

We then demonstrated that the complex process of positioning cell fates can be reduced into: (i) a first BMP4 signaling module comprising of two inhibitory nodes, receptor relocalization and NOGGIN induction, (ii) a Nodal signaling module that is directly linked to the outcome of the first BMP4 module. A quantitative modeling of this network led to the recapitulation of SMAD1 and SMAD2 spatio-temporal signaling dynamics. As a consequence, ordering of the fate domains was qualitatively predicted, demonstrating successful decomposition of the patterning mechanism into a minimal set of functional modules.

Our study has revealed an unexpected level of organization within hESC colonies: cell polarity is differentially established at the colony level as a function of density and proximity to the boundary. This led to spatial modulation of signaling responses to TGF-β ligands in the pluripotent state due to receptor lateralization. The cytoplasmic signals that specify the lateral targeting for TGF-β receptors were previously defined (Murphy et al., 2007). At the colony edge, receptors cannot be lateralized due to the absence of neighbors and therefore are exposed to the medium.

Density and radial structure also influences the behavior of unconfined colonies growing on dishes. It has previously been shown, for example, that hESC differentiation occurs more efficiently at the colony edges (Rosowski et al., 2015). No mechanisms have been proposed to explain variable morphogen sensitivity within colonies. In this study, we attribute the lower differentiation potential of interior cells to absence of accessible receptors.

Our observation that both BMP4 and Activin/Nodal receptors are not accessible to apically applied ligands has important implications for morphogen transport within the embryo. In mouse, the visceral endoderm shares a basal membrane with the epiblast prior to gastrulation and the cells at its distal tip, expressing the Wnt and Nodal inhibitors, move laterally to define the future head (Arnold and Robertson, 2009). Thus inhibitors applied basally are positioned appropriately to block laterally expressed receptors. In addition, the absence of receptors on the apical side makes signals passing through the amniotic cavity irrelevant. Interestingly, gastrulating human embryos have a cell density in the epiblast that matches those in our high-density colonies. From the Carnegie collection (embryo #7801) in early gastrulation, stage 6b, we measured a linear density of about 200 cells/mm versus 170 cells/mm in the top right colony in Figure 3A. It is therefore likely that the same principles about lateral receptor localization and signal propagation between layers apply in the human embryo. In conclusion, our study unveils the communication channels used for cell-cell communication in the embryo, and demonstrates that signal reception is as relevant to patterning as signal production.

Our knowledge about morphogen/inhibitor induction is very limited in mammals. We presented strong evidence in hESCs that NOGGIN is a direct target of BMP4. This result contrasts with our knowledge about dorso-ventral patterning in Xenopus (Harland and Gerhart, 1997). Surprisingly, we did not observe any NOGGIN induction in mouse ESC converted to epiblast, (Figure 4D and S4B–C). This suggests important differences in the inductive rules between vertebrates, and will require further investigations with more direct murine equivalents of hESCs like epiblast cells directly derived from embryos, rather than epiblast converted cells. The chick embryo is closer to human in geometry than mouse and there is evidence for a rapidly diffusing inhibitor that insures only one streak forms per embryo (Bertocchini et al., 2004). In vertebrates there is evidence for direct induction of Noggin by BMP4 in osteoblasts and related cells lines (Gazzerro et al., 1998; Nifuji and Noda, 1999). Thus the embryonic induction of NOGGIN by BMP4 may have been derived in human from an ancestral role in the adult.

Here, we suggest a novel function for NOGGIN in early human gastrulation. In mammals, the earliest documented function of NOGGIN is to control neural induction in the ectoderm, downstream of its secretion from the node (McMahon et al., 1998) and a double knock-out of the BMP4 inhibitors NOGGIN and CHORDIN in the mouse does not show any phenotype at gastrulation stage (Bachiller et al., 2000). Surprisingly, we found in our system that NOGGIN was critical in positioning fate domains over a wide range of density in our human gastruloids, which strongly suggests a role for NOGGIN in positioning the primitive streak and mesendodermal populations during early human gastrulation. This could constitute an important mechanistic difference between human and mouse in the mechanisms positioning of germ layers.

Our micropattern assay also provides an ideal platform for quantifying paracrine signaling in apical-basal polarized epithelium. Noggin inhibits signaling by binding to BMP4 with pico-molar affinity (Zimmerman et al., 1996). How activators and inhibitors spread between epithelial cells is obscure (Alexandre et al., 2014; Reilly and Melton, 1996; Stanganello et al., 2015). Regardless of the precise mechanism of transport, a simple model for the Noggin activity profile based on intra-layer diffusion was able to quantitatively fit the experimental data. We explored computationally how violations to our idealized model affect our conclusions and found that qualitatively, our results were compatible with a broad range of parameters (Methods). Thus a model for NOGGIN with two-dimensional diffusion within the colony quantitatively recapitulates our data.

Our predictions of the fate-domain positioning were close to the measured differentiation patterns, and the qualitative ordering of the domains was correctly recapitulated. Ultimately, cell movements and cell-to-cell adhesive interactions will refine our predictions that are based on signal integration. This contrasts with more conventional models considering multiple genes in each signaling pathway, and a gene regulatory network for their targets. In some instances, mathematics beyond the usual gene by gene differential equations is required to tie together qualitative facts that are intrinsic to development (Corson and Siggia, 2012). An extensive description of all these links is impossible at the level of millimeter-sized colonies on the time scale of days. Thus if one hopes to explain morphogenesis in engineering terms that can be used in a predictive way to guide regenerative medicine, some astute phenomenological descriptions of sub-processes are necessary. Exploiting these facts in our system unveiled a novel mechanism that ties together multiple qualitative facts and led to a spatio-temporal description of gastruloid development.

Experimental Procedures

Cell culture

Experiments were performed with the RUES2 hESC line. All hESC lines were grown in HUESM medium that was conditioned by mouse embryonic fibroblasts (MEF-CM) and supplemented with 20 ng/ml bFGF. Cells were tested for mycoplasma infection before beginning experiments and then again at 2-month intervals. Cells were grown on tissue culture dishes coated with Matrigel (BD Biosciences, 1:40 dilution). Dishes were coated in Matrigel overnight at 4 °C and then incubated at 37 °C for 1 h immediately before the cells were seeded on the surface.

Micropatterned cell culture

We used micropatterned glass coverslips from CYTOO using the protocol described in Warmflash et al., 2013. Compared to this previous study, we used colonies of a fixed size: disks of 500um diameter coated with Laminin 521. As shown in Figure 1a, this experimental setting allowed recapitulating the same fate ordering as presented in our last publication.

Transwell experiments

We used Costar Transwells made of clear polyester membrane inserts, in a 24 well design. Membranes were coated by 2 hours incubation with 250µl of a pre-warmed solution containing 10 µg/ml of laminin 521 Biolamina diluted in PBS +/+. The membrane was then rinsed four times with PBS+/+ and finally with culture medium. A suspension containing a defined number of single cell in growth medium containing Rock-inhibitor Y27632, 10 µM was then applied to the menbrane immediately following the last wash. Rock-Inhibitor was removed the next day. Cells were stimulated 24 hour after Rock Inhibitor removal. For imaging of the cells on the membrane, the transwell was removed from the multi-well plate and, on top of a coverslip and imaged with a 10× objective.

Quantification of receptor localization

We used the ZO-1 stain to define an apical surface in our colonies. First, ZO-1 confocal stacks were background-substracted by directly substracting the median-filtered image stack. Images were then binarized using Otsu thresholding. For each position in the (x,y) plane, we measured the mean z localization of the positive pixels. We used this information to interpolate a continuous surface defined as the apical surface. We then binarized the other channels using Otsu threshold and measured the z-localization of positive pixels with respect to the z-coordinate of the previously defined apical surface. Results were presented as violin plots in Figure 3B.

Supplementary Material

Highlights.

  • hESC polarization controls response to TGF-Beta ligands.

  • Secreted inhibitor NOGGIN spatially restricts BMP activity at the colony edge.

  • Signaling dynamic results from gradual cell polarization and NOGGIN diffusion.

  • Differentiation patterns within gastruloids can be quantitatively predicted.

Acknowledgments

Our research was supported by an NSF PHY-1502151 to EDS a NIH 5R01HD080699-02 to AHB, a Deutsche Forschungsgemeinschaft fellowship to JM, a CHDI fellowship to AR, Starr Grant 2014-001 fellowship to ZO, a NSF grant #DGE-132526 to AY and a Simons foundation grant supported CK. FE thanks the Bettencourt-Schueller foundation for a young researcher award. We thank Alessia Deglincerti for constructive comments and Shu Li for technical assistance. We thank Sophie Morgani and Kat Hadjantonakis for helpful discussions and help with mouse cells.

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.

Contribution:

Conceptualization and writing, F.E., A.B. and E.S.; Investigation, F.E., A.R. and A.Y.; Modeling, F.E., J.M. and E.S.; Software, C.K.; Resources, M.O.; All authors reviewed the MS.

References

  1. Alexandre C, Baena-Lopez A, Vincent JP. Patterning and growth control by membrane-tethered Wingless. Nature. 2014;505:180–185. doi: 10.1038/nature12879. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Arnold SJ, Robertson EJ. Making a commitment: cell lineage allocation and axis patterning in the early mouse embryo. Nature reviews Molecular cell biology. 2009;10:91–103. doi: 10.1038/nrm2618. [DOI] [PubMed] [Google Scholar]
  3. Bachiller D, Klingensmith J, Kemp C, Belo JA, Anderson RM, May SR, McMahon JA, McMahon AP, Harland RM, Rossant J, et al. The organizer factors Chordin and Noggin are required for mouse forebrain development. Nature. 2000;403:658–661. doi: 10.1038/35001072. [DOI] [PubMed] [Google Scholar]
  4. Ben-Haim N, Lu C, Guzman-Ayala M, Pescatore L, Mesnard D, Bischofberger M, Naef F, Robertson EJ, Constam DB. The nodal precursor acting via activin receptors induces mesoderm by maintaining a source of its convertases and BMP4. Developmental Cell. 2006;11:313–323. doi: 10.1016/j.devcel.2006.07.005. [DOI] [PubMed] [Google Scholar]
  5. Bertocchini F, Skromne I, Wolpert L, Stern CD. Determination of embryonic polarity in a regulative system: evidence for endogenous inhibitors acting sequentially during primitive streak formation in the chick embryo. Development. 2004;131:3381–3390. doi: 10.1242/dev.01178. [DOI] [PubMed] [Google Scholar]
  6. Claude P. Morphological factors influencing transepithelial permeability: a model for the resistance of the zonula occludens. The Journal of membrane biology. 1978;39:219–232. doi: 10.1007/BF01870332. [DOI] [PubMed] [Google Scholar]
  7. Corson F, Siggia ED. Geometry, epistasis, and developmental patterning. Proc Natl Acad Sci U S A. 2012;109:5568–5575. doi: 10.1073/pnas.1201505109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Gazzerro E, Gangji V, Canalis E. Bone morphogenetic proteins induce the expression of noggin, which limits their activity in cultured rat osteoblasts. The Journal of clinical investigation. 1998;102:2106–2114. doi: 10.1172/JCI3459. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Harland RM, Gerhart J. Formation and function of Spemann organizer. Annual review of cell and developmental biology. 1997;13:611–667. doi: 10.1146/annurev.cellbio.13.1.611. [DOI] [PubMed] [Google Scholar]
  10. James D, Noggle SA, Swigut T, Brivanlou AH. Contribution of human embryonic stem cells to mouse blastocysts. Dev Biol. 2006;295:90–102. doi: 10.1016/j.ydbio.2006.03.026. [DOI] [PubMed] [Google Scholar]
  11. Jullien J. Morphogen gradient interpretation by a regulated trafficking step during ligand-receptor transduction. Genes &amp; Development. 2005;19:2682–2694. doi: 10.1101/gad.341605. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Krtolica A, Genbacev O, Escobedo C, Zdravkovic T, Nordstrom A, Vabuena D, Nath A, Simon C, Mostov K, Fisher SJ. Disruption of apical-basal polarity of human embryonic stem cells enhances hematoendothelial differentiation. Stem cells. 2007;25:2215–2223. doi: 10.1634/stemcells.2007-0230. [DOI] [PubMed] [Google Scholar]
  13. Meinhardt H. Models of Biological Pattern Formation. 1982:1–211. [Google Scholar]
  14. Muller P, Rogers KW, Yu SR, Brand M, Schier AF. Morphogen transport. Development. 2013;140:1621–1638. doi: 10.1242/dev.083519. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Murphy SJ, Shapira KE, Henis YI, Leof EB. A unique element in the cytoplasmic tail of the type II transforming growth factor-beta receptor controls basolateral delivery. Mol Biol Cell. 2007;18:3788–3799. doi: 10.1091/mbc.E06-10-0930. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Nallet-Staub F, Yin X, Gilbert C, Marsaud V, Ben Mimoun S, Javelaud D, Leof EB, Mauviel A. Cell Density Sensing Alters TGF-β Signaling in a Cell-Type-Specific Manner, Independent from Hippo Pathway Activation. Developmental Cell. 2015;32:640–651. doi: 10.1016/j.devcel.2015.01.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Nifuji A, Noda M. Coordinated expression of noggin and bone morphogenetic proteins (BMPs) during early skeletogenesis and induction of noggin expression by BMP-7. Journal of bone and mineral research : the official journal of the American Society for Bone and Mineral Research. 1999;14:2057–2066. doi: 10.1359/jbmr.1999.14.12.2057. [DOI] [PubMed] [Google Scholar]
  18. Nowotschin S, Hadjantonakis A-K. Cellular dynamics in the early mouse embryo: from axis formation to gastrulation. Current Opinion in Genetics &amp; Development. 2010;20:420–427. doi: 10.1016/j.gde.2010.05.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Paine-Saunders S, Viviano BL, Economides AN, Saunders S. Heparan sulfate proteoglycans retain Noggin at the cell surface: a potential mechanism for shaping bone morphogenetic protein gradients. J Biol Chem. 2002;277:2089–2096. doi: 10.1074/jbc.M109151200. [DOI] [PubMed] [Google Scholar]
  20. Poh Y-C, Chen J, Hong Y, Yi H, Zhang S, Chen J, Wu DC, Wang L, Jia Q, Singh R, et al. Generation of organized germ layers from a single mouse embryonic stem cell. Nature communications. 2014;5:4000. doi: 10.1038/ncomms5000. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Reilly KM, Melton DA. Short-range signaling by candidate morphogens of the TGF beta family and evidence for a relay mechanism of induction. Cell. 1996;86:743–754. doi: 10.1016/s0092-8674(00)80149-x. [DOI] [PubMed] [Google Scholar]
  22. Rosowski KA, Mertz AF, Norcross S, Dufresne ER, Horsley V. Edges of human embryonic stem cell colonies display distinct mechanical properties and differentiation potential. Scientific reports. 2015;5:14218. doi: 10.1038/srep14218. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Sasai Y. Cytosystems dynamics in self-organization of tissue architecture. Nature. 2013a;493:318–326. doi: 10.1038/nature11859. [DOI] [PubMed] [Google Scholar]
  24. Sasai Y. Scaling of dorsal-ventral patterning by embryo size-dependent degradation of Spemann&apos;s organizer signals. Cell. 2013b;153:1296–1311. doi: 10.1016/j.cell.2013.05.004. [DOI] [PubMed] [Google Scholar]
  25. Sela-Donenfeld D, Kalcheim C. Localized BMP4-noggin interactions generate the dynamic patterning of noggin expression in somites. Dev Biol. 2002;246:311–328. doi: 10.1006/dbio.2002.0672. [DOI] [PubMed] [Google Scholar]
  26. Stanganello E, Hagemann AI, Mattes B, Sinner C, Meyen D, Weber S, Schug A, Raz E, Scholpp S. Filopodia-based Wnt transport during vertebrate tissue patterning. Nat Commun. 2015;6:5846. doi: 10.1038/ncomms6846. [DOI] [PubMed] [Google Scholar]
  27. Stephenson RO, Rossant J, Tam PPL. Intercellular interactions, position, and polarity in establishing blastocyst cell lineages and embryonic axes. Cold Spring Harbor Perspectives in Biology. 2012;4 doi: 10.1101/cshperspect.a008235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. van den Brink SC, Baillie-Johnson P, Balayo T, Hadjantonakis AK, Nowotschin S, Turner DA, Martinez Arias A. Symmetry breaking, germ layer specification and axial organisation in aggregates of mouse embryonic stem cells. Development. 2014;141:4231–4242. doi: 10.1242/dev.113001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Varelas X, Samavarchi-Tehrani P, Narimatsu M, Weiss A, Cockburn K, Larsen BG, Rossant J, Wrana JL. The Crumbs complex couples cell density sensing to Hippo-dependent control of the TGF-beta-SMAD pathway. Dev Cell. 2010;19:831–844. doi: 10.1016/j.devcel.2010.11.012. [DOI] [PubMed] [Google Scholar]
  30. Walsh SV, Hopkins AM, Chen J, Narumiya S, Parkos CA, Nusrat A. Rho kinase regulates tight junction function and is necessary for tight junction assembly in polarized intestinal epithelia. Gastroenterology. 2001;121:566–579. doi: 10.1053/gast.2001.27060. [DOI] [PubMed] [Google Scholar]
  31. Warmflash A, Arduini BL, Brivanlou AH. The molecular circuitry underlying pluripotency in embryonic stem cells. Wiley Interdiscip Rev Syst Biol Med. 2012;4:443–456. doi: 10.1002/wsbm.1182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Warmflash A, Sorre B, Etoc F, Siggia ED, Hemmati-Brivanlou AH. A method to recapitulate early embryonic spatial patterning in human embryonic stem cells. Nature methods. 2014;11:847–854. doi: 10.1038/nmeth.3016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Zimmerman LB, De Jesus-Escobar JM, Harland RM. The Spemann organizer signal noggin binds and inactivates bone morphogenetic protein 4. Cell. 1996;86:599–606. doi: 10.1016/s0092-8674(00)80133-6. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

NIHMS818336-supplement.pdf (4.2MB, pdf)

RESOURCES