Fig 1.
Spatial patterning of TBXT, SOX2 and TBX6 in the posterior E8.5 mouse embryo.
A: Diagram representing a 4-somite pair E8.5 embryo and showing the positioning of the confocal imaging field of view (FOV) used in b and c as a dashed outline. The FOV encompasses the primitive streak (PS), and a domain comprising the neural and mesodermal bi-fated caudal lateral epiblast (CLE) plus node streak border (NSB) in blue. Anterior (A)/Posterior (P) and Left (L)/Right (R) and posterior neural tube (PNT) are indicated. The dotted red box corresponds approximately to the insert shown in (b). The node is positioned at the posterior end of the primitive streak. B: Confocal z-slices across each plane of the field of view shown in (a) with insert (red box) showing the spatially heterogeneous TBX6 expression in the SOX2+ TBXT+ CLE, and the spatially homogeneous TBX6 expression in PSM. White dotted line indicates the boundary between epiblast and the presomitic mesoderm (PSM). Unless specified, scale bars indicate 50 µm. C: 3D renders of LaminB1 signal at 4, 6, and 8 somite pair stage E8 embryos. Highlighting the increase in ‘Pringle-like’ epiblast curvature and the associated challenge posed when assessing patterning in 3D.
Fig 2.
3D Epiblast projection and alignment method: PRINGLE.
A: Single cell analysis pipeline: individual nuclei are identified from 3D confocal images and manually labeled as epiblast or non-epiblast using the PickCells software. Fluorescence intensity is quantified within each nucleus and the neighbors of each cell in 3D are identified. B: Summary of PRINGLE method to project the 3D nuclei centroids in 2D and align multiple embryos: The distance of nuclei centroids are calculated from the midline and notochord along principal curves in transverse and sagittal sections, respectively. These distances are then used as coordinates to project nuclei in 2D and create an unfolded epiblast. Unfolded maps generated from multiple embryos are then registered together by normalizing nuclei coordinates relative to tissue landmarks: the midline and edge of epiblast are used as landmarks for the left/right (L/R) axis, and the notochord origin and primitive streak posterior tip as landmarks for anterior/posterior (A/P) axis. However relative positions produce a distortion, therefore relative positions are normalized to the average epiblast width. This registration procedure results in an ‘average’ epiblast map in 2D. D: dorsal, V: ventral, A: Anterior, P: Posterior, M: Midline, L: Lateral edge of epiblast. This approach is explained in more detail in S1 Fig. C: A four-somite-pair epiblast showing SOX2, TBXT, and TBX6 expression (measured as outlined in (a)) mapped onto the average epiblast shape of multiple four-somite-pair epiblasts (established as outlined in (b)). Points represent nuclei centroids. Data for Fig 2 (C): Data file 1, https://doi.org/10.5281/zenodo.15802710.
Fig 3.
Mapping the spatial distribution of Neuro-Mesodermal Progenitors.
A: Manifold projection of nuclei centroids in an example 4-somite-pair embryo showing SOX2+ TBXT+ cells (orange) and the regions in the lateral epiblast (Lt1-2) and node-streak border (B) defined by grafting experiments [34,35] as having neural and mesodermal potency (bifated regions: black boxes). SOX2 + TBXT+ are not restricted to the known N-M-bi-fated regions and instead map onto a broader spatial region. B: Iterative kernel averaging spatially smooths TF intensity values, shown by an example 4-somite-pair embryo. C: (i) PCA dimension reduction of the smoothed TF values in order to identify cell states. PCA loadings per TF indicated by arrow direction and magnitude (n = 19 embryos). (ii) The pseudotime tool Slingshot [44] identifies a path between the SOX2 high and the TBX6 high regions to create a pseudo-space continuum (n = 19 embryos). D: The pseudo-space continuum values (corresponding to cell states as defined by expression of TBXT, SOX2, and TBX6) mapped back onto the manifold projection of the four-somite-pair epiblast. Excluded cells (gray) are determined as described in the methods section. Pseudo-space values within known N-M-bifated regions are likely to correspond to NMP identities. E: The boundary between cells excluded because they fall outside the domain of potential neural/mesoderm fate (gray) and domains containing mono-fated, neural OR mesoderm cells (blue) is determined as described in the methods section. Red nuclei correspond to cells with pseudospace values defined in (Cii, D: see also main text) as corresponding to likely NMP identities (‘NMP-like cells’). These map to a U-shaped region that extends beyond the boundaries of the known N-M-bifated regions (black boxes: based on functional grafting experiments [34,35]. Data for Fig 3 (A-E): Data file 1, https://doi.org/10.5281/zenodo.15802710.
Fig 4.
Mapping the shape and steepness of gradients of transcription factor expression in 3D.
A: Method to map the shape of gradients across 3D space (i) First scale the unit vectors between a nuclei and its neighbors to the change in smoothed TF values. (ii) Then sum the vectors to obtain a single directional vector, with magnitude representing the steepness of the gradient. B: (i) Contour plots of average binned normalized fluorescence intensity (NFI) measurements for SOX2, TBXT, and TBX6 in the normalized epiblast shape using PRINGLE, as described in Fig 2. Embryos are grouped per somite pair stage during E8.5, n numbers shown per group. White lines indicate the NMP region of interest (ROI) for each individual embryo as determined using the approach shown in Fig 3. (ii) Contour plots mapping TF gradient steepness map, calculated as shown in A. Averaged data from multiple embryos is shown, as explained in (c). Gradients of TBXT and TBX6 appear to be steepest in regions corresponding to NMP-like cells (NMP ROI) and gradients of SOX2 appear to be steepest in the node streak border. C: (i) Averages of single cell NFI measurements, as measured in Fig 2A, and measurement of gradient steepness, as measured in b (ii), per embryo and somite pair stage plotted in relation to pseudo space values, corresponding to cell identities based on integration of SOX2, TBX6, and TBXT expression as determined in Fig 3. This confirms that gradients of TBX6 and TBXT gradient are steepest within the NMP ROI and decreases into the PS, with TBX6 spatially lagging behind TBXT. Solid lines indicate the fit of a non-parametric multiple regression curve and shading indicates 0.05–0.95 confidence intervals calculated as (mean ± 1.96 * (σ/(√n)). Vertical dashed lines indicate gating from fitting TF pseudospace to the N-M-bi-fated regions in Fig 3. Data for Fig 4 (B-C): Data file 1, https://doi.org/10.5281/zenodo.15802710.
Fig 5.
Mapping the properties of cells in relation to their local neighborhood.
A: The coefficient of variation (CV) is used as a metric for the degree of local heterogeneity in fluorescence intensity values for a cell and its direct neighbors (in 3D), defined as the standard deviation divided by the mean of fluorescence intensity values for a cell and its neighbors. Neighbors are defined as in Fig 2A. B: Violin Superplots showing CV values for SOX2, TBXT, and TBX6 within NMP-like cells as defined in Fig 3. Local heterogeneity (CV) in TBX6 is greater than TBXT, which is greater than local heterogeneity in SOX2. Dots indicate the mean for each of 19 embryos. Statistical tests performed with paired t test and Benjamini–Hochberg post-hoc correction.. *** = p < 0.001. C: Maps of local variability in TF expression across the epiblast. Contour plots showing CV values mapped onto EpiMap. Data is averaged from multiple embryos as described in Fig 4). U-shaped white lines indicate the NMP ROI as measured in Fig 3. High local variability (CV) for TBXT and TBX6 is observed within the NMP ROI. D: Heightened TBXT and TBX6 CV to the NMP region and decreases into the mesodermal PS. Lines indicate the fit of a non-parametric multiple regression curve and shading indicates 0.05 and 0.95 confidence intervals calculated as (mean ± 1.96 * (σ/(√n)). E: The neighbor ratio (NR) metric can distinguish patterning scenarios where a cell is (1) surrounded by cells with an averaged expression that is higher expression than itself, (2) the same as itself, or (3) lower expression than itself. Defined as ln(Cell TF value/ Average Neighbor TF value). F: Comparing the ‘high’ cells in the TF+ populations (upper 50th percentile) for SOX2, TBXT, and TBX6 to synthetic data assuming a normal distribution for a cell’s expression and its neighbors. This shows a decrease in NR, indicating TBX6+ high cells are surrounded by cells expressing lower TBX6 levels than expected within a random pattern. Points indicate average NR per embryo, shaded areas show confidence intervals of 0.05 and 0.95 calculated as (mean ± 1.96 * (σ/(√n)). Cells were binned into even intervals and only cells within these intervals were considered for confidence intervals and statistical summary metrics. G: Statistical analysis of the difference in observed and synthetic data NR values along pseudospace bins using (i) earth mover’s distance between normalized distributions for each embryo and (ii) p-values resulting from one way ANOVA between observed and synthetic embryo means. The difference in TBX6 NR values is statistically significant in the NMP ROI with a high difference in distributions compared to TBXT or SOX2. Shaded areas show confidence intervals of 0.05 and 0.95. Dashed line in (ii) indicates p-value = 0.05. Data for Fig 5 (B–D and F–G): Data file 1, https://doi.org/10.5281/zenodo.15802710.
Fig 6.
Comparing local patterning between embryos, gastruloids and monolayer cell culture.
A: Protocol for generating hNMPs from hESC and representative confocal images of the culture fixed at 72 h and stained for SOX2, TBXT, TBX6, and LAMINB1 (further characterization S7 Fig). B: Protocol for generating gastruloids [51] with a representative confocal image of a gastruloid fixed at 120 h for SOX2, TBXT, TBX6, and LAMINB1. Red box indicates typical field of view for high resolution image acquisition and subsequent neighbor analysis. C: Density plots comparing each nucleus’s TF normalized fluorescence intensity (NFI) value to its neighbors shows strong linear correlations for TBXT and SOX2, but not for TBX6 with many ‘high’ TBX6 cells surrounded by ‘low’ TBX6 cells. Consistent in the E8 epiblast, gastruloids, and in vitro human NMP like cell monolayers. Black line indicates the median and the dashed line indicates x = y line. D: Violin superplots of TF local heterogeneity (coefficient of variation [CV]) for respective TF+ cells within NMP-like cells isolated in gastruloids (process outlined in S8 Fig) and hNMP monolayers compared to in vivo WT E8 bi-fated region. NMPLC populations within in vitro models are more heterogeneous than comparative in vivo NMP populations. All scale bars indicate 50 µm. Points represent replicate means. Error bars represent 5%–95% confidence intervals. Gastruloids n = 3 with 3 technical replicates. Data for Fig 6 (C–D): Data file 1 and Data file 2, https://doi.org/10.5281/zenodo.15802710.
Fig 7.
Mapping changes in patterning after inhibition of Notch activity.
A: Culture method where a posterior explant of an E8.5 embryo is cultured in N2B27 media with the Notch inhibitor LY at 150 nM or an equal amount of DMSO for 12 h. B: Representative 3D renders of explants for each condition, stained for LaminB1 or SOX2, TBXT, and TBX6 indicating normal tissue morphology but a reduction in TBXT and TBX6 expression. Anterior (A), posterior (P), dorsal (D), ventral (V) axes are indicated. Scale bars indicate 100 μm. C: Analysis pipeline to evaluate the effect of Notch inhibition on patterning. After using PRINGLE to create an EpiMap, the epiblasts are folded at the midline and clustered with k-means to demark comparable ROIs. Patterning metrics are compared between DMSO and LY in these regions. One way ANOVA is conducted to identify the statistical significance of the difference of patterning metrics, regions are filled with associated colors to denote the p-value. D: Contour plots of TF normalized fluorescence intensity (NFI) projected onto the EpiMap with the difference in DMSO vs. LY conditions and associated p-values per ROI. Displaying an increase of SOX2 NFI in the PS and a posterior withdrawal of the TBXT and TBX6 domains in the Notch inhibited condition. E: Contour plots of NFI gradient steepness projected onto the EpiMap with the difference in DMSO vs. LY conditions and associated p-values per ROI. Note the overall decrease in TBXT and TBX6 gradient steepness, with a new posteriorly located nonlinear TBXT gradient. F: Contour plot of TBX6 local heterogeneity (CV) on EpiMap. Note the statistically significant decrease in the CLE and NSB like regions TBX6. Data for Fig 7 (D–F): Data file 3, https://doi.org/10.5281/zenodo.15802710.
Fig 8.
Using PRINGLE to examine scaling of axial patterning between chick and mouse embryos.
A: Diagram of pre-somitic Hamburger-Hamilton stage 6 embryo, approximate field of view used in confocal imaging shown by red box (the developing head and notochord are not indicated). Confocal image stacks of HH6 embryo labeled with hybridization chain reaction (HCR) of SOX2, TBXT, and MSGN1 transcripts represented with a max projection, a single-z plane and 3D render, highlighting the difficulty in representing raw chick imaging data due to undulating nature of epithelium. B: outline of pipeline. C: Cytoplasm inference method for HCR quantification by peri-nuclear segmentation of DAPI signal. Nuclear segmentation performed with cellpose. D: Output of semi-supervised machine-learning classifier, with post-processing, to predict whether a nucleus is epiblast or non-epiblast, showing high accuracy. E: Size comparison of chick HH6 and 4-somite pair (SP) mouse epiblasts shown by PRINGLE projections to normalized A-P and L-R positions. Average SOX2/TBXT cytoplasmic signal for chick and SOX2/TBXT nuclear protein for mouse represented by colored contours (average of four embryos). White lines represent midline (vertical line) and position of node-streak border (intersection with horizontal line). F: The absolute and relative positions of peak paraxial mesoderm to the node streak border (NSB) and PS end show scaling of the focal point to ~30% of primitive streak length despite a large difference in absolute length. Peak paraxial mesoderm is defined as nuclei in the top 90th percentile for paraxial mesoderm marker expression, MSGN1 in chick and TBX6 in mouse. Error bars show standard deviation. G: Comparison of relative global gradient maps for mouse and chick epiblasts show similar “M” shape for SOX2, TBXT “T” shape, and elliptical paraxial mesoderm pattern by TBX6 in mouse and MSGN1 in chick. H: Density plot showing the pairwise expression of SOX2 and TBXT, manual thresholds shown as black dashed lines outlining SOX2+/TBXT+ population. I: Average density distribution ((no. cells per bin)/(total no. cells)) of SOX2+/TBXT+ cells forms a u-shape around the PS, similar to mouse NMP region (Fig 3). White dashed line indicates the position of the node-streak border. Data for Fig 8 (D–I): Data file 4, https://doi.org/10.5281/zenodo.15802710.
Fig 9.
Using PRINGLE to analyze patterning in Drosophila embryos.
A: Graphic outlining dorsal-ventral (D-V) patterning boundaries of neuroectoderm and presumptive ventral mesoderm in a late Stage 5 Drosophila embryo. Adapted from [53]. B: Confocal XY and XZ slices of representative embryo labeled by hybridization chain reaction (HCR) for sim, ind, and twist, with the anterior and posterior circumferential domains of salm marking the gnathal and tail regions, respectively [65]. Nuclear membrane marker Lamin labeled by immunofluorescence. White dashed line indicates the section position in the right panel. C: Output of PRINGLE projection from ventral midline after single cell quantification of HCR signal. n = 4. D: The distance from the ventral midline to the twist boundary, ind boundaries, and approximate dorsal midline along anterior-posterior (A-P) axis reveals a seemingly stable twist boundary despite changes in the embryo D-V length. ind domain boundaries appear to converge at the posterior end. E: Directly measuring ind region width shows a linear reduction in width along the A-P axis. F: The linear reduction is maintained when considering width as a relative measure from the twist dorsal boundary to the dorsal midline. G: The location of the twist boundary relative to the ventral and dorsal midlines shifts dorsally along the A-P axis and could explain some but not all the reduction in ind width. H: In terms of distance from the twist boundary, the ventral boundary of ind is maintained at ~35μm but the dorsal boundary shifts ventrally along the A-P axis. E–H: Individual embryos shown by gray lines where left-right patterns were individually calculated and the average taken. Thick line represents the average across embryos. Shaded areas refer to 0.05 and 0.95 confidence intervals (mean ± 1.96 * (σ/(√n)). I: Possible mechanisms to explain the maintenance of the ventral ind boundary with a variable dorsal boundary. Dorsal gradient and induced vnd does not change but 1. the EFG-R active domain restricts and/or 2. the Dpp activity expands to dorsally restrict ind. Data for Fig 9 (C–H): Data file 5, https://doi.org/10.5281/zenodo.15802710.