Recording morphogen signals reveals origins of gastruloid symmetry breaking

Abstract

When cultured in three dimensional spheroids, mammalian stem cells can reproducibly self-organize a single anterior-posterior axis and sequentially differentiate into structures resembling the primitive streak and tailbud. Whereas the embryo's body axes are instructed by spatially patterned extra-embryonic cues, it is unknown how these stem cell gastruloids break symmetry to reproducibly define a single anterior-posterior (A-P) axis. Here, we use synthetic gene circuits to trace how early intracellular signals predict cells' future anterior-posterior position in the gastruloid. We show that Wnt signaling evolves from a homogeneous state to a polarized state, and identify a critical 6-hour time period when single-cell Wnt activity predicts future cellular position, prior to the appearance of polarized signaling patterns or morphology. Single-cell RNA sequencing and live-imaging reveal that early Wnt-high and Wnt-low cells contribute to distinct cell types and suggest that axial symmetry breaking is driven by sorting rearrangements involving differential cell adhesion. We further extend our approach to other canonical embryonic signaling pathways, revealing that even earlier heterogeneity in TGF_β signaling predicts A-P position and modulates Wnt signaling during the critical time period. Our study reveals a sequence of dynamic cellular processes that transform a uniform cell aggregate into a polarized structure and demonstrates that a morphological axis can emerge out of signaling heterogeneity and cell movements even in the absence of exogenous patterning cues.

Keywords: Gastruloid; cell signaling; self-organization; symmetry breaking.

Figure 1:. Dynamics of Wnt symmetry breaking and polarization during gastruloid morphogenesis.

(A) Illustration of gastruloid self-organization phenomenon: a transient, spatially uniform stimulus somehow triggers the formation of a polarized morphology without an exogenous pre-pattern. (B) A clonal mESC line was engineered to report Wnt activity through the expression of a destabilized iRFP downstream of Wnt-sensitive TCF/LEF enhancer sites. (C) ‘Symmetry-breaking’ protocol for gastruloid generation. mESC cultures were maintained in 2i+LIF culture media until immediately before gastruloid formation to suppress pre-existing heterogeneity in Wnt activity. Gastruloids were formed from 200 initial cells/gastruloid seeded using a cell sorter, and treated with 3 μM CHIR between $\mathrm{t}=48$ and 72 h to stimulate morphogenesis. (D) Dynamics of Wnt activity patterns during gastruloid morphogenesis. Samples were fixed at variable timepoints and imaged to measure spatial distributions of Wnt signaling. (E) Single-cell Wnt activity levels were measured by flow cytometry. Histograms of Wnt activity indicate an initially uniform response to Wnt activation with CHIR at 72 h, followed by a bimodal response. (F) Quantification of the proportion of cells which are Wnt active over time. A Wnt-inactive population is first detectable at $\mathrm{t}=90$ hours. (G) Quantification of heterogeneity and polarization in spatial patterns of Wnt activity during gastruloid morphogenesis ($\mathrm{n}=76$ gastruloids). Heterogeneity is reported as a normalized standard deviation, and polarization is reported as a normalized distance between the center of mass (COM) of Wnt activity and morphological images. Quantification indicates an onset of heterogeneity at $\mathrm{t}=90\mathrm{h}$ (consistent with flow cytometry), followed by polarization at $\mathrm{t}=108\mathrm{h}$.

Figure 2:. Recording morphogen signals with recombinase circuits.

(A) Illustration of signal recording design criteria. An ideal recorder would irreversibly label a signaling-defined population within a temporal ‘listening window’ of interest so that this population can be followed over time. (B) Schematic of recording circuit. An upstream transcription factor (rtTA) requires both signaling activity and small-molecule addition to drive recombinase expression, which in turn irreversibly changes the fluorophore in a ‘recording’ locus. (C) Characterization of a clonal Wnt recording cell line fidelity. 24 h incubation with both Wnt-activating CHIR (3 μM) and listening window-defining doxycycline (2 μg/mL) achieves complete GFP labeling (left, bottom). Treatment with either CHIR or doxycycline alone has no detectable labeling (left, top). A 1 h recording window is sufficient to achieve efficient labeling (right). (D) Characterization of switching kinetics of Wnt recorders in response to media changes. A 1 h doxycycline treatment was applied at variable lag times $\left(\mathrm{\Delta }\mathrm{t}={\mathrm{t}}_{\mathrm{d}\mathrm{o}\mathrm{x}}\text{-}{\mathrm{t}}_0\right)$ following a media change. Recorder performance approached steady-state media performance by $\mathrm{\Delta }\mathrm{t}=6\mathrm{h}$. (E) Crosstalk assessment for 3 separate clonal lines recording Wnt, Activin/Nodal, and BMP pathway activity. All recording windows utilized 100 ng/mL doxycycline and 200 ng/mL morphogen concentration. Baseline conditions report labeling with only doxycycline in basal media (N2B27). Pseudocolors indicate relative proportion of cells labeled with GFP expression. Recording windows were 6 hours for Wnt and Nodal recorders and 3 h for the BMP recorder to account for differences in sensitivity.

Figure 3:. Mapping fate information encoded in Wnt signaling histories.

(A) Schematic of experimental design. Wnt histories are recorded by varying the onset time ${\mathrm{t}}_{\mathrm{d}\mathrm{o}\mathrm{x}}$ of a doxycycline-defined listening window (90 minutes, 200 ng/mL) during gastruloid morphogenesis. (B) Representative final images $\left.{(\mathrm{t}}_{\mathrm{F}}=134\mathrm{h}\right)$ of gastruloids in which Wnt activity was recorded at different timepoints. Insets (top-right) show the corresponding pattern of instantaneous Wnt activity (Figure. 1D) during the queried listening window (scale bar = 200 um). (C) Quantification of anterior-posterior (A-P) axial patterns of Wnt recorder labeling at different timepoints throughout gastruloid morphogenesis at 6 h temporal resolution ($\mathrm{n}=119$ gastruloids measured). Shaded regions indicate standard deviation as a function of A-P position. (D) Illustration of patterns corresponding to the ‘symmetry-breaking window’ at ${\mathrm{t}}_{\text{dox}}=96\mathrm{h}$, the first timepoint at which Wnt recording predicts a clear anterior-posterior separation of cell fates. The measurement of fate information within ‘patchy’ patterns of Wnt activity suggests cellular rearrangements contribute to gastruloid polarization. (E) Kymograph of the data presented in Figure 1C to visualize different Wnt dynamics corresponding to different spatial positions. Following the emergence of an anterior domain at ${\mathrm{t}}_{\text{dox}}=96\mathrm{h}$, Wnt signaling becomes progressively more restricted to the posterior domain. (F) Integrated Wnt signaling activity between $\mathrm{t}=96$ and $\mathrm{t}=134\mathrm{h}$ shows a linear ‘temporal gradient’ associated with A-P fate.

Figure 4:. Mapping signaling histories in transcriptional space.

(A) Schematic of experimental design. Wnt signaling states are recorded in a 90 minute window initiated at ${\mathrm{t}}_{\text{dox}}$, and gastruloids are then grown to a later time ${\mathrm{t}}_{\text{seq}}$. Gastruloids are then dissociated into single cell suspensions, and separated into GFP positive and negative signaling populations via fluorescence activated cell sorting (FACS). Sorted populations are then loaded into separate lanes of a Chromium controller (10x Genomics) to prepare for single-cell RNA sequencing (see Methods). Separate gene expression libraries are then prepared with distinct library indices to disambiguate signal recording conditions in pooled sequencing. (B) Representative image of a gastruloid at ${\mathrm{t}}_{\text{seq}}=120\mathrm{h}$, labeled according to signaling domains (scale bar = 200 μm). (C) Annotating cell type clusters with signaling information. Left: cell types identified by Leiden clustering within ${\mathrm{t}}_{\mathrm{s}\mathrm{e}\mathrm{q}}=120\mathrm{h}$ gastruloids. Middle: final Wnt activity distribution across cells as measured by ${\mathrm{P}}_{\mathrm{T}\mathrm{C}\mathrm{F}/\mathrm{L}\mathrm{E}\mathrm{F}}\text{-}\mathrm{r}\mathrm{t}\mathrm{T}\mathrm{A}$ expression. Final Wnt activity is concentrated within the neuromesodermal progenitor (NMP) cluster. Right: relative composition of Leiden clusters according to Wnt activity recorded at ${\mathrm{t}}_{\mathrm{d}\mathrm{o}\mathrm{x}}=96\mathrm{h}$. Wnt activity is broadly distributed throughout most clusters, but excluded substantially from the somite, endoderm, and endothelial clusters (see also Methods). (D) Single-cell expression levels of reference genes associated with gastrulation and axial morphogenesis. (E) Illustration of reference gene organization across the anterior-posterior axis during axial elongation and somitogenesis. The organization of reference gene expression (Figure 1D) aligns with inferred histories of Wnt activity (Figure 1C), suggesting that Wnt signaling histories predict transcriptionally defined cell fates along a spatially defined anterior-posterior axis.

Figure 5:. Wnt signaling during symmetry breaking is associated with differential transcriptional and mechanical cell states.

(A) Wnt activity at $\mathrm{t}=96\mathrm{h}$ is organized into ‘patchy’ patterns of local correlated domains lacking global polarization. (B) scRNAseq analysis of gastruloids collected at ${\mathrm{t}}_{\mathrm{s}\mathrm{e}\mathrm{q}}=96\mathrm{h}$. Left: Leiden clustering identified 2 cell types. Right: Wnt activity (as measured by ${\mathrm{P}}_{\mathrm{T}\mathrm{C}\mathrm{F}/\mathrm{L}\mathrm{E}\mathrm{F}}\text{-}\mathrm{r}\mathrm{t}\mathrm{T}\mathrm{A}$ expression) was highly concentrated within cluster 2 (‘Wnt active’) and excluded from cluster 1 (‘Wnt inactive’). (C) Volcano plot reveals genes which are differentially expressed between the two Leiden clusters. (D-E) Validation of differential cadherin expression between cell clusters by immunofluorescence. (D) The Wnt-active cell type is characterized by exclusive expression of E-Cadherin/Cdh1. (E) In contrast, the Wnt-inactive cell type is enriched for Protocadherin-19/Pcdh19, an alternative homotypic adhesion marker. (F) Still images from Movie S1 illustrating dynamics of cellular rearrangements during Wnt polarization. These rearrangements sort cells from patchy expression domains into a globally polarized signaling axis (Figure 3D).

Figure 6:. Early Nodal/BMP activity predicts and controls Wnt symmetry breaking

(A) In the mouse embryo, BMP and Wnt pathway signals interact in the posterior epiblast to initiate gastrulation. Nodal activity marks the anterior-most aspect of the resultant primitive streak (i.e., the ‘node’). (B) Gastruloids show spontaneous signaling activity in both the Nodal and BMP pathways at $\mathrm{t}=48\mathrm{h}$, before Wnt activity is detectable. (C) Signal recording experimental design. Early Nodal or BMP activity was recorded immediately prior to the stimulation of Wnt activity with CHIR (3 h recording window from $\mathrm{t}=45\text{-}48\mathrm{h}$; 200 ng/mL dox). (D) Representative images of final distribution ${(\mathrm{t}}_{\mathrm{f}}=120\mathrm{h})$ of cells in which early Nodal/BMP activity was recorded, visualized through a medial optical section (scale bar = 200 μm). (E) Quantification of fate information recorded from early Nodal and BMP activity reveals that these predict a future anterior-posterior axis prior to the observation of Wnt activity ($\mathrm{n}=22$ gastruloids total). (F) Left: Illustration of a model in which differences in early Nodal activity influences differential future responses to Wnt stimulation, thereby predicting future cell fates. Right: experimental design to assess this model. (G) Single-cell Wnt activity levels at $\mathrm{t}=96\mathrm{h}$ measured by flow cytometry. Activin A pretreatment drives more cells into the iRFP-negative (i.e. ‘Wnt-inactive’) population in a dose-dependent manner. (H) Quantification of the fraction of iRFP-positive (i.e. ‘Wnt-active’) cells across $\mathrm{n}=3$ replicates for each treatment condition ($\mathrm{n}=30$ pooled gastruloids per condition per replicate). (I) Representative images of gastruloids in which Wnt activity was recorded at $\mathrm{t}=96\text{-}97.5\mathrm{h}$ and imaged at $\mathrm{t}=120\mathrm{h}$, with and without Activin A pretreatment. Scale bar = 200 μm. (H) Quantification of Wnt recording patterns demonstrates that Activin A pretreatment expands the Wnt-inactive anterior region ($\mathrm{n}=42$ gastruloids total).

Figure 7:. Proposed model of gastruloid symmetry breaking and polarization.

(A) Illustration of the phases of symmetry breaking. Initial spontaneous Nodal/BMP activity modulates the response of cells to a uniform CHIR stimulus, yielding a heterogenous Wnt pattern. The heterogeneity ultimately resolves into a single pole of Wnt activity which organizes subsequent axial elongation. (B) Approximate timeline of signaling dynamics in model. Dashed bars indicate heterogeneity between cells; width of lines indicate relative fraction of signaling-active cells within the gastruloid.