Dynamic and non-uniform expression of key transcription factors provides insights into the emergence of neural crest cells at the neural plate border
Andrew Montequin, Carole LaBonne
TL;DR
This study uses advanced imaging in frogs to show how key genes are dynamically expressed during the formation of neural crest cells, revealing new insights into their development.
Contribution
The study reveals dynamic and non-uniform gene expression patterns during neural crest formation, identifying regulatory roles of pax3 and zic1.
Findings
Neural crest genes initially show broad and overlapping expression with distinct spatial biases.
pax3 and zic1 dynamically regulate snai2 and sox8, influencing neural crest formation.
Later stages show inverse correlation between neural crest and border factors as identity emerges.
Abstract
The neural crest is a vertebrate stem cell population with broad developmental potential. While a gene regulatory network describing establishment of these cells has been generated, much remains to be learned about the dynamics of this process. Here, we use fluorescent in situ hybridization chain reaction to quantify the spatiotemporal dynamics of neural crest formation in Xenopus. We find that the initial onset of neural crest genes is broad and partially overlapping, with distinct anterior-posterior and medio-lateral biases. A shared neural crest domain emerges, but some genes retain relative expression differences that persist into migratory stages, producing stream-specific gene expression patterns. These differences correlate with dynamic expression of the neural plate border factors pax3 and zic1. Correlating relative intensities of pax3 and zic1 with the presence or absence of…
Genes, proteins, chemicals, diseases, species, mutations and cell lines named across the full text — each resolved to its canonical identifier and authoritative record.
Click any figure to enlarge with its caption.
Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9- —National Institute of General Medical Scienceshttp://dx.doi.org/10.13039/100000057
- —National Science Foundationhttp://dx.doi.org/10.13039/100000001
- —Simons Foundationhttp://dx.doi.org/10.13039/100000893
- —Northwestern Universityhttp://dx.doi.org/10.13039/100007059
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsDevelopmental Biology and Gene Regulation · Pluripotent Stem Cells Research · Genomics and Chromatin Dynamics
INTRODUCTION
During vertebrate embryogenesis, neural crest stem cells arise at the neural plate border (NPB), a transient ectodermal territory situated between neural and non-neural domains. Neural crest cells display broad multi-germ layer developmental potential that derives in part from their expression of components of the gene regulatory network (GRN) controlling pluripotency in blastula stem cells (Buitrago-Delgado et al., 2015, 2018; Schock et al., 2023, 2024; Lignell et al., 2017; Pajanoja et al., 2023; Roellig et al., 2017; York et al., 2024). Neural crest cells subsequently undergo epithelial-to-mesenchymal transition, migrate extensively, and contribute to a broad and diverse set of derivatives including craniofacial cartilage, pigment cells and peripheral neurons (Thawani and Groves, 2020).
A key early event in the genesis of neural crest cells is the establishment of the NPB. Research to date suggests that BMP, Wnt and FGF signals combine at the NPB to position the expression of pax3/7, msx1/2, zic1, tfap2a and id3 at the lateral boundary of the developing neural plate (Prasad et al., 2012; Groves and LaBonne, 2014; Thawani and Groves, 2020). These factors, and in particular pax3 and zic1, have been proposed to regulate expression of definitive neural crest genes, including snai2, foxd3 and sox8/9 (Garnett et al., 2012; Hong and Saint-Jeannet, 2007; Monsoro-Burq et al., 2003; Pla and Monsoro-Burq, 2018; Tribulo et al., 2003).
A significant body of work across multiple model organisms has now shown that, from its onset, the expression of NPB transcription factors overlaps with expression of components of the blastula pluripotency GRN (Buitrago-Delgado et al., 2015; Lignell et al., 2017; Pajanoja et al., 2023; Schock et al., 2024; York et al., 2024). Recently it was shown that Sox3, traditionally viewed as a neural plate factor, is in fact necessary for NPB formation, revealing an unexpected functional role at the neural–non-neural interface (Schock et al., 2024). Sox3 functions together with Pou5f transcription factors in promoting the expression of NPB genes; however, its expression must be downregulated for the transition from an NPB to a neural crest state to proceed. klf17, another component of the pluripotency GRN, has been shown to restrict the boundaries of the NPB, confining its domain and sharpening its definition (Rigney et al., 2025). These findings emphasize that NPB formation is not simply a passive outcome of morphogen gradients promoting expression of core NPB factors, but rather the product of active regulatory mechanisms that both promote and delimit border identity. They also suggest that the deployment of canonical NPB and neural crest factors may be more dynamic and context dependent than current GRN models depict.
Because development relies upon precise regulation of state transitions it is important to have a high-resolution understanding of the underlying gene expression dynamics. Much of our current understanding of NPB and neural crest gene expression comes from chromogenic in situ hybridization, which is largely nonquantitative and has limited spatial resolution. Recent advances in gene expression detection, such as fluorescent in situ hybridization chain reaction (HCR-FISH), allow for enhanced sensitivity, multiplexing of probe sets, and quantitative fidelity and enable the simultaneous visualization of overlapping transcripts with fine spatial resolution (Choi et al., 2018).
To gain further insights into neural crest development, we employed HCR-FISH to measure and provide relative quantification of neural crest gene expression in early Xenopus laevis embryos. We found that at the onset of their expression at late gastrula stages, neural crest genes had broad and only partially overlapping expression domains within the dorsal ectoderm. Their expression resolved to a shared pre-migratory neural crest domain during neurulation. However, within this domain the expression of snai2, sox8 and foxd3 was non-uniform, displaying anterior-posterior (A-P) differences that persisted through neural crest migration. We correlate these differences with expression levels of the NPB genes pax3 and zic1, which are highly dynamic, and show that these factors can differentially drive expression of neural crest genes. Finally, we followed co-expression of NPB and neural crest genes through neurulation and observed reduced expression of pax3 and zic1 within the pre-migratory neural crest, with differences in relative expression levels across axial levels. Taken together, these results suggest that patterns of neural crest gene expression are broadly initiated throughout the dorsal ectoderm and maintain differences in relative expression levels that correlate with overlapping patterns of NPB regulatory factors.
RESULTS
Expression of neural crest genes in the early ectoderm is non-uniform
Definitive neural crest cells have been shown to arise during late gastrulation and early neurulation (Basch et al., 2006; LaBonne and Bronner-Fraser, 1998; Mayor et al., 1995; Pajanoja et al., 2023; Spokony et al., 2002). We therefore collected wild-type X. laevis embryos at fixed time points corresponding to Nieuwkoop and Faber stages 12 to 15 (Fig. 1A) (Nieuwkoop and Faber, 1994; Zahn et al., 2022) and used HCR-FISH to probe for expression of the neural crest markers snai2, sox8, foxd3, sox9 and sox10. Expression of snai2, sox8 and foxd3 in neural crest-forming regions was first detected at stage 12.5 (Fig. 1B,C), prior to the onset of sox9 and sox10 expression (Fig. S1), although expression of foxd3 was quite low. In addition to their presumptive neural crest expression, low levels of snai2, sox8 and foxd3 were expressed in mesoderm at the dorsal midline, and sox8 and foxd3 displayed distinct expression around the blastopore during gastrulation. This is consistent with previous studies that detected foxd3 expression in mesoderm (Hanna et al., 2002; Steiner et al., 2006). We also observed sox8 expression in the anterior neural fold region at stages 12.5 and 13. At these late gastrula/early neurula stages, there appeared to be some heterogeneity in the expression of snai2, sox8 and foxd3, both medio-laterally and anterior-posteriorly. Strikingly, while expression of these genes resolved to a single, shared neural crest domain by stages 14 and 15, we observed that their relative expression differed along the A-P axis, with snai2 and sox8 expressed more strongly in anterior regions of the neural crest and foxd3 expressed more strongly in posterior regions. Additionally, expression of foxd3 was noted in the region of low snai2 expression thought to mark the boundaries of the mandibular and hyoid crest segments.
Neural crest gene expression is dynamic and non-uniform during gastrulation and neurulation. (A) Time-course collection protocol for wild-type Xenopus laevis embryos at 19°C. Images adapted from Zahn et al. (2022). Xenopus illustrations © Natalya Zahn (2022). Xenbase (www.xenbase.org RRID:SCR_003280). (B) Maximum intensity z-projections of neural crest genes snai2 and foxd3. (C) Maximum intensity z-projections of neural crest genes snai2 and sox8. All images in B,C are views of the dorsal midline, with the anterior pole facing upwards. Images are representative of a minimum of 4 samples per stage and probe combination. Scale bars: 1 mm (A); 250 μm (B,C).
Surface mapping and cell layer separation reveal key features of gene expression
In order to quantify the expression differences that we observed, we developed image analysis protocols to verify the linearity of the HCR-FISH signal and measure gene expression within specific regions of the developing ectoderm. Previous studies examining neural crest gene expression in transverse tissue sections of Xenopus embryos revealed that the strongest expression occurs in the deep ectodermal cell layer, compared to the superficial (Linker et al., 2000; Hong and Saint-Jeannet, 2007). We adapted a published image analysis package for surface detection and mapping of 3D images (Heemskerk and Streichan, 2015) to separate the superficial and deep ectodermal cell layers within our 3D confocal microscopy images (Fig. 2A). To validate our approach, we multiplexed snai2 HCR with the NPB gene pax3, which is known to have strong expression in the presumptive hatching gland in the superficial layer (Hong and Saint-Jeannet, 2007). We verified that intensity measurements in our images are suitable for relative quantification through redundant detection of pax3 expression using orthogonal HCR-FISH probe sets (Fig. S2). To create the orthogonal probe set, we generated 24 probes targeting pax3 mRNA and divided the probes between two HCR initiator sequences (B1 and B3). The result was two non-overlapping probe sets targeting pax3 that could be multiplexed and imaged in different channels. Consistent with prior observations, snai2 expression was strongest within the deep ectodermal cell layer, although it was expressed in some cells in the superficial layer (Fig. 2B). We also observed strong pax3 expression in the superficial layer in the presumptive hatching gland, which was largely non-overlapping with snai2 expression in that layer. Strikingly, when we compared the expression of these two genes in the deep layer, we observed that pax3 expression wrapped around but was largely excluded from the snai2-positive domain. Despite the appearance of colocalization in maximum-intensity z-projections (Fig. 2B, top row), surface mapping and cell layer separation revealed that snai2 and pax3 are expressed in largely non-overlapping domains by stage 15.
Gene expression patterns in the ectoderm differ between cell layers. (A) Overview of the computational surface-mapping protocol. The program detects the surface of the embryo within 3D z-stacks from confocal microscopy images, fits and shifts the surface to an appropriate depth, and creates a ‘flattened’ z-stack following the shape of the embryo surface. (B) Maximum intensity z-projections from the full, unmapped image (top row), the superficial ectodermal cell layer (middle row) and the deep ectodermal cell layer (bottom row). The stage 15 embryo was double-labeled with probes targeting pax3 and snai2. Scale bar: 250 μm (B, top row, unmapped images). Surface-mapped images are displayed without a scale bar because the mapping process induces minor length distortions that vary across the image. Images are representative of 15 samples.
Broad onset of neural crest gene expression
We next wished to carry out higher resolution imaging of neural crest gene expression starting with stage 12.5, the earliest stage that we reliably detected expression of snai2 and sox8. We used expression of the NPB factor tfap2a to delineate two A-P regions along the NPB for imaging (Fig. 3A,B). While expression of snai2 and sox8 showed some overlap particularly in the more posterior region, there were also striking spatial differences in their expression in these regions. Expression of snai2 was broader across the medial-lateral axis, whereas sox8, as noted earlier, was also expressed in the anterior NPB (Fig. 3A′,B′). Both foxd3 and sox9 also displayed distinct expression patterns at stage 13, with sox9 extending more laterally and more anteriorly than foxd3 (Fig. S3).
Non-uniform spatial onset of neural crest gene expression resolves during neurulation. (A,B) Maximum intensity z-projection of NPB marker tfap2a. Dashed lines mark the dorsal midline, annotated to show the anterior and posterior directions. Insets mark the regions shown at higher magnification in A′ and B′. (A′,B′) Maximum intensity z-projections of snai2 and sox8 from the fields of view denoted in A and B. Dashed lines denote approximate boundaries of detected expression. Scale bars: 250 μm (A-B′). Images are representative of 4 samples. (C,D) Normalized line intensity profile measurements for double-labeled snai2/sox8 images at stage 13 (n=16) (C) and at stage 15 (n=13) (D). Grayscale images on the left are representative mean-intensity projections of slices taken only from the deep cell layer after mapping, and the dashed lines denote the regions used for line profile measurements. Shaded areas represent s.d. for normalized intensity measurements at each pixel location. L, lateral; M, medial.
Despite the early spatial differences in their onset of expression, canonical neural crest genes were expressed in a shared domain by neurulation. To further investigate the expression differences we had noted along the A-P axis (Fig. 1), we used surface-mapped images and measured line intensity profiles of snai2 and sox8 at multiple developmental stages and axial levels. Line intensity profiles highlighted the higher relative expression of snai2 in the anterior of the neural crest domain at stage 13 and the higher relative expression of sox8 in the posterior region (Fig. 3C). Interestingly, these measurements also revealed that at the lateral boundaries of the neural crest domain sox8 expression was higher than snai2 in both anterior and posterior regions, although those medial-lateral differences had largely resolved by stage 15. The higher relative expression of snai2 in the more anterior neural crest had also largely resolved by stage 15; however, the higher relative sox8 expression in the posterior domain persisted (Fig. 3D). Together, these results indicate that at the onset of their expression in neural crest regions snai2 and sox8 are activated non-uniformly; these patterns resolve to a shared spatial domain during neurulation but retain relative differences in expression at different axial levels.
Differential expression of neural crest genes persists in migratory neural crest
After observing differences in relative expression levels of snai2, sox8 and foxd3 at stage 15, we investigated whether these expression differences persisted into later stages of neural crest development by imaging the deep ectodermal cell layer of embryos probed for snai2 and foxd3 (Fig. 4A, Fig. S4A), or snai2 and sox8 (Fig. 4B, Fig. S4B), by HCR-FISH at stages 15, 17 and 20. At stage 15, foxd3 expression could be clearly observed in the gap of low/no snai2 expression between the anterior and posterior segments of snai2 expression, and the reciprocal was true for regions of low foxd3 expression (Fig. 4A). In addition, there was a posterior snai2 expression domain, and a smaller anterior domain that did not express foxd3. At stage 17, the higher expression of foxd3 in the posterior versus anterior cranial neural crest was more pronounced, and there was a strong upregulation of foxd3 in the center of the neural crest domain. Expression of both genes was at this stage detected in the smaller medial neural crest domain, although here too there was notable expression heterogeneity. The expression difference between snai2 and foxd3 was most striking at stage 20. The anterior-most mandibular stream as it extended into the periphery largely lacked foxd3 expression. By contrast, the hyoid neural crest stream displayed strong expression of foxd3 but little snai2. The caudal, presumptive branchial neural crest, stream was enriched for foxd3 in its anterior cells and snai2 in its posterior cells.
Relative expression differences within the neural crest persist through neural crest migration. (A) Gene expression patterns of snai2 and sox8 within the deep ectodermal cell layer at stages 15, 17 and 20. The box in stage 15 images indicates a region of high snai2/low foxd3 in the top half of the box, and low snai2/high foxd3 in the bottom half of the box. (B) Gene expression patterns of snai2 and foxd3 within the deep ectodermal cell layer at stages 15, 17 and 20. Labels in the bottom right panels of A,B indicate the mandibular (M), hyoid (H) and branchial (B) migratory streams, as well as the direction of the anterior-posterior (A-P) axis. Images are representative of a minimum of 3 samples per stage and probe combination. Surface-mapped images are presented without a scale bar due to minor length distortions induced by the mapping.
We also observed differences when we compared snai2 and sox8 expression at the same stages and axial levels (Fig. 4B). At stage 15, both genes were expressed more highly in the anterior versus posterior cranial neural crest, but snai2 expression extended more anteriorly and the middle region of low snai2 expression was enriched for sox8. At stage 17, both genes displayed strong expression in three domains that appeared to presage the three initial migratory neural crest streams. At stage 20, there was co-expression of snai2 and sox8 in the mandibular crest stream, and, as was observed for foxd3, sox8 expression predominated in the hyoid stream and in the anterior-most half of the branchial stream. These patterns were also observed in bilateral imaging of both neural crest domains (Fig. S4).
Expression of NPB genes is dynamic and heterogenic
The quantitative differences in expression we observed for snai2, sox8 and foxd3 along the A-P axis of the neural crest domain raises the important question of how these differences are generated. Neural crest cells arise at the NPB, which has been characterized by expression of genes such as pax3/7, msx1/2, zic1, tfap2a and id3 (Prasad et al., 2012; Groves and LaBonne, 2014; Thawani and Groves, 2020). These factors, and in particular pax3 and zic1, have been proposed to regulate expression of definitive neural crest genes, such as snai2, sox8 and foxd3. Accordingly, we next sought to measure whether overlapping patterns of gene expression at the NPB correlated with the onset and maintenance of neural crest gene expression, focusing initially on pax3 and zic1 starting at stage 12, before the onset of definitive neural crest gene expression.
At stage 12, zic1 was expressed in a horseshoe-like domain comprising the anterior neural fold and the anterior portion of the NPB (Fig. 5A). By contrast, pax3 expression was excluded from the anterior neural fold and extended along the full length of the lateral NPB. Co-expression of these genes was limited to a small region at the anterior of the pax3 domain that correlates with the location of definitive neural crest gene activation. From stage 12.5 to 14, pax3 was broadly expressed in the NPB as well as in the hatching gland, clearing from the neural crest domain at stage 15. Throughout these stages, expression of zic1 was highly dynamic. Z-scored voxel intensities in dorsal ectoderm images, which represent intensity values as the number of standard deviations away from the mean intensity value, showed zic1 expression peaking at late gastrula stages (Fig. 5B). The combined expression of pax3 and zic1 is not fully consistent with a model in which overlapping expression of these genes define a pre-pattern that positions neural crest gene expression (Sato et al., 2005). We also examined expression of msx1 and tfap2a at these same stages because of their known roles in NPB specification, and they too displayed a small region of overlapping expression at stages 12 and 12.5. Interestingly, by stage 15, msx1 expression appeared to be downregulated in the definitive neural crest region whereas tfap2a expression was maintained (Fig. S5).
Expression patterns of pax3 and zic1 are dynamic during gastrulation and neurulation. (A) Maximum intensity z-projections of NPB genes pax3 and zic1 at stages 12-15. Scale bars: 250 μm. Images are representative of a minimum of 4 samples per stage. (B) Z-scored voxel intensities of pax3 and zic1 from 10× surface-mapped images selecting the deep cell layer of the dorsal ectoderm.
Transcription of snai2 and sox8 in the presumptive neural crest correlates with relative pax3 and zic1 expression levels
Given the differences in the spatial expression of both NPB and early neural crest genes, we wanted to better understand how these differences correlated. The high levels of background fluorescece in early Xenopus embryos prevented relative quantification of expression levels using probes imaged in the green channel (Alexa Fluor 488, data not shown); however, HCR-FISH probes targeting the intronic regions of RNA could reliably detect nuclear-associated puncta marking sites of active transcription using this channel (Fig. 6A,B). Using stage 12.5 embryos triple-labeled with probes targeting pax3, zic1 and intronic regions of either snai2 or sox8, we sought to measure the frequency of neural crest transcriptional loci within ranges of normalized pax3 and zic1 expression levels.
Distinct probabilities of early snai2 and sox8 transcription based on local pax3 and zic1 expression levels in the NPB. (A) Maximum intensity z-projection of DAPI and snai2 intronic probes at stage 12.5. (B) Maximum intensity z-projection of DAPI and sox8 intronic probes at stage 12.5. Insets in the left panels of A,B denote the digital zoom displayed in the following three panels. Arrows indicate examples of intronic probe detection and localization with nuclei/DAPI. Scale bars: 100 μm. (A,B) Images are representative of a minimum of 13 samples per stage and probe combination. (C) Overview of the image analysis pipeline used to estimate the levels of snai2 and sox8 transcription given local pax3 and zic1 intensities. (D) Heat map depicting t-statistics from two-sided t-tests between bootstrapped sample and null distributions estimating the percentage of neural crest-positive voxels within each bin of pax3/zic1 intensity. Data collected from n=15 images for the snai2 intronic labeling, and n=13 images for the sox8 intronic labeling.
To accomplish this, we collected images from a region of pax3 and zic1 overlap within the NPB where we expected to see transcription of neural crest genes. We downscaled each image to approximately cellular-sized voxels, and for each voxel we recorded the mean pax3 and zic1 intensities, as well as a ‘True/False’ value indicating whether the voxel contained at least one detected locus of neural crest gene transcription. We binned normalized pax3 and zic1 intensity measurements into quintiles and calculated the fraction of voxels within each bin that contained nascent neural crest transcripts. Next, we randomly resampled voxel measurements with replacement and re-calculated the fraction of neural crest-positive voxels within bins to generate a bootstrapped sample distribution of this fraction estimate. Finally, we generated a null distribution for the fraction estimate by repeatedly shuffling the pax3 and zic1 voxel intensity measurements, and we compared the sample distribution to the null distribution within each bin using a two-sided t-test (Fig. 6C, Fig. S6A,B).
The t-statistic from each of these t-tests is reported in the heat map in Fig. 6D, with positive values for a given bin indicating the probability of detecting a neural crest marker locus within that quintile range of pax3 or zic1 intensity that is greater than random chance within our NPB images. Comparing the heat maps for snai2 versus sox8 transcription revealed striking differences in the association of these two markers with pax3 and zic1 expression. We detected high frequencies of snai2 transcription only within voxels above the 40th percentile of pax3 intensity, and largely independent of zic1 intensity, suggesting that snai2 expression is Pax3 dependent. By contrast, we detected elevated frequencies of sox8 transcription even in voxels with high zic1 but low pax3 intensities, suggesting that Zic1 may be capable of activating sox8 independently of Pax3. Nevertheless, the highest frequencies of sox8 transcription correlated with high intensity of both zic1 and pax3, consistent with co-regulation.
Although our measurements of the association between pax3/zic1 expression levels and neural crest transcription were made within a zoomed-in region of the NPB, we found that global distributions of pax3 and zic1 could partially predict gene expression outside of one small region. Whole-embryo images of pax3 and zic1 expression at stage 12.5 color-coded according to the heatmaps in Fig. 6D predicted that there may be sox8 transcription in the anterior NPB (Fig. S6C-E), which is not typically associated with the neural crest, and images from this approximate region show colocalization of sox8 and zic1 expression (Fig. S6F).
Pax3 and Zic1 drive differential activation of snai2 and sox8
Given the distinct correlations of snai2 and sox8 expression with pax3 and zic1 intensities, we wanted to compare the abilities of Pax3 and Zic1 to drive expression of these neural crest genes. We measured the induction of snai2 and sox8 under a variety of conditions in pluripotent cells explanted from the animal pole of blastula-stage embryos. Previous work has shown that co-expression of Pax3 and Zic1 using mRNA constructs containing the coding sequence of these transcription factors fused to the hormone-binding domain of the glucocorticoid receptor (Pax3-GR and Zic1-GR) is capable of inducing expression of neural crest genes (Hong and Saint-Jeannet, 2007; Milet et al., 2013). We therefore expressed Pax3-GR and Zic1-GR in animal pole cells, independently and in combination, and examined snai2 and sox8 expression using HCR-FISH (Fig. 7).
Pax3 and Zic1 drive differential expression of snai2 and sox8 in animal caps. (A-D) HCR-FISH images of neural crest genes snai2 and sox8 in stage 15 animal caps for uninjected (A), Pax3-GR (B), Zic1-GR (C) and Pax3-GR+Zic1-GR (D) conditions. Images are representative of a minimum of 16 samples per condition. Scale bars: 100 μm.
Consistent with previous work, we observed robust induction of snai2 and sox8 expression in animal caps with Pax3-GR and Zic1-GR expressed in combination when compared to animal caps from control embryos (Fig. 7A,D). We found that Zic1-GR alone, but not Pax3-GR, could drive sox8 expression in animal pole cells (Fig. 7B,C) supporting our correlation analysis. By contrast, we observed low expression of snai2 in explants expressing only Pax3-GR (Fig. 5B), but not in explants expressing only Zic1-GR (Fig. 5C), consistent with our prediction that snai2 expression may be Pax3 dependent.
Within our animal cap images, we observed that neural crest gene expression was punctate, with high snai2 and sox8 intensities within a relatively small number of voxels. Because of this expression pattern, differences in median intensities between images were small and did not accurately describe the qualitative differences in expression that we observed. However, assessing the full empirical cumulative distribution functions (ECDFs) of voxel intensities from our explant images allowed us to compare the expression intensities of voxels at different percentiles (Fig. S7A,B). As an example, comparing the intensities at 0.9 on an ECDF plot between uninduced and Pax3-GR+Zic1-GR conditions allowed us to measure how much brighter the 90th percentile intensities are between the two conditions. For the conditions in which we qualitatively observed increased neural crest gene expression relative to control, we observed a rightward shift in the upper percentiles of the ECDF when plotted against the normalized marker intensity, indicative of brighter puncta of expression within the induced animal caps. We further quantified this shift by measuring the 99th percentile of the marker intensity within each explant, where we measured a statistically significant increase in snai2 expression in the Pax3-GR expressing explants, but not Zic1-GR explants, compared to control (Fig. S7C). We also observed a statistically significant increase in sox8 expression in the Zic1-GR expressing explants compared to controls (Fig. S7D). Explants expressing Pax3-GR alone displayed induced sox8 (Fig. 7B, Fig. S7), indicating that Pax3 is capable of directly driving low levels of sox8 expression.
Neurula-stage pax3 and zic1 expression inversely correlates with neural crest gene expression
While prevailing models suggest that co-expression of Pax3 and Zic1 may position neural crest gene expression to a shared domain, our HCF-FISH studies indicate that at later stages the expression of NPB and neural crest genes might be inversely correlated. While whole-mount chromogenic in situ hybridization can suggest that pax3 expression is sustained within the pre-migratory neural crest (Sato et al., 2005), our findings indicate that differential expression between superficial and deep ectodermal cell layers may confound such observations (Fig. 2B). To further probe the relationship between NPB and definitive neural crest gene expression, we used computational surface mapping and line intensity profiles to separate cell layers within the ectoderm and quantify expression levels.
When we compared snai2 and pax3 expression in the deep cell layer, we observed that pax3 levels were reduced in regions where we observe the highest snai2 expression (Fig. 8A). We quantified this reduction in relative expression levels by measuring line intensity profiles through lines drawn across the snai2 expression domain at two axial levels (Fig. 8A). We observed reduced relative expression of pax3 within both the anterior and posterior snai2 domain, but the reduction was more pronounced in the anterior neural crest where snai2 expression is higher. Notably, however, pax3 was also reduced in the region of low snai2 expression that lies between the anterior and posterior snai2 domains, suggesting that pax3 is likely downregulated by some other mechanism in this region.
Expression of pax3 and zic1 is differentially reduced in the anterior and posterior neural crest. (A) Normalized line intensity profiles for double-labeled snai2/pax3 images at stage 15 (n=12) in the deep ectodermal cell layer. Deep cell layers images from Fig. 2B (left panels) are annotated to indicate the regions analyzed. (B) Normalized line intensity profiles for double-labeled snai2/zic1 images at stage 15 (n=13) in the deep ectodermal cell layer. Grayscale images on the left of A,B are representative mean-intensity projections of slices taken only from the deep cell layer after mapping, and the dashed lines denote the regions used for line profile measurements. Shaded areas represent s.d. for normalized intensity measurements at each pixel location. L, lateral; M, medial. (C,D) Surface-mapped images of zic1/snai2 expression (C) and foxd3/snai2 expression (D; from Fig. 4A) used to indicate (dashed line) the region used to make the xy projections shown in C′ and D′, respectively. A, anterior; P, posterior. Images are representative of a minimum of 9 samples per probe combination. Scale bars: 100 μm (C,D); 100 μm (C′,D′).
When we compared snai2 expression to that of zic1 we observed similar correlations. Zic1 expression was lowest in the anterior region where snai2 was highest (Fig. 8B). In the posterior region, where snai2 expression is lower, there was a less dramatic loss of zic1. Overlaying snai2 and zic1 expression makes clear the inverse relationship between the expression levels of these factors, with snai2 expression localized to the center region of lowest zic1 (Fig. 8C,C′). That Snai2 plays a role, directly or indirectly, in controlling expression levels of pax3 and zic1 is supported by experiments in which snai2 was depleted, which led to a spatial expansion of NPB factor expression (Fig. S8). We observed similar inverse correlations between expression of snai2 and foxd3 (Fig. 8D,D′). Together, these data indicate that the anterior domain of cranial neural crest is characterized by high snai2 expression and low zic1 and foxd3 expression while the posterior domain is characterized by relatively lower levels of snai2 relatively higher levels of foxd3, zic1 and low levels of pax3, and suggest that snai2 levels at least partially control the expression of pax3 and zic1 at these stages.
DISCUSSION
While progress has been made in delineating the gene regulatory interactions underlying neural crest formation, much remains to be learned about the temporospatial dynamics of key components of the neural crest GRN and the relationships of distinct components to each other. Here, we use HCR-FISH to gain new insights into the expression of key transcription factors as cells progress from a NPB to a neural crest state, revealing previously unappreciated features of this process. We show that expression of definitive neural crest genes during gastrulation is initially broad and non-uniform before resolving to a shared pre-migratory domain during neurulation, and that subsequently these genes display spatial heterogeneity within that domain, particularly along its A-P extent. These regional differences were maintained even as neural crest cells commenced migration, and likely presage subsequent lineage segregation. For example, we observed that foxd3 expression remained highest in the posterior cranial neural crest – notably in a region corresponding to relatively low expression of snai2. Such persistence of spatial heterogeneity into migratory stages aligns with recent single-cell analyses in avian embryos that uncovered position-dependent transcriptional states among premigratory and early migratory neural crest cells (Lignell et al., 2017).
Our data also reveal that the upstream NPB genes, including pax3 and zic1, are expressed in a dynamic manner during neurulation with expression patterns that evolve and exhibit distinct spatial distributions. We found that the initial expression of definitive neural crest genes is differentially associated with specific relative levels of pax3 and zic1 within the NPB. Experiments in animal caps confirmed that Pax3 and Zic1 can drive differential activation of neural crest genes, although both are required for robust expression. We also find that neurula-stage patterns of pax3 and zic1 reflect some of the relative expression differences between neural crest genes that persist into migration. Finally, we observed that pax3 and zic1 expression become partially downregulated within the nascent neural crest domain as neurulation progresses, even as these genes remain strongly expressed in flanking NPB or neural/epidermal regions. It is tempting to speculate that the expression of Pax3 and Zic1 surrounding the neural crest domain may help restrict the boundaries of the neural crest similar to what we recently reported for Klf17 (Rigney et al., 2025).
Together, these results suggest a modified model for neural crest specification whereby overlapping patterns of NPB gene expression leads to differences in the activation and maintenance of individual neural crest genes. This transient overlap of NPB and neural crest gene expression is followed by a relative uncoupling: as expression of neural crest genes rise to high levels, the expression of NPB factors attenuates, especially in regions of high snai2 expression (Fig. 9). This is consistent with a model in which prolonged maintenance of border transcription factors can inhibit transit to a definitive neural crest state (Light et al., 2005; Nordin and LaBonne, 2014; Schock et al., 2024).
Differential onset and resolution of neural crest gene expression in the NPB. Our data suggest that neural crest gene expression is differentially induced broadly throughout the ectoderm and resolves to a non-uniform domain during neurulation. Gene name colors correspond to shaded areas overlaid on the drawings, with cross-hatched areas indicating regions of shared gene expression. NC, neural crest. Adapted from Nieuwkoop and Faber (1994) (www.xenbase.org/xenbase/anatomy/alldev.do).
Non-uniform and dynamic gene expression within the NPB
Prevailing descriptions of the NPB describe both a mixed pool of progenitor cells and a defined spatial domain within the ectoderm. Fate-mapping experiments in the chick epiblast have suggested a broad co-mingling of neural, neural crest, placodal and epidermal precursors within the border region, while tissue-grafting experiments in amphibians suggested that interactions between neural and non-neural ectoderm play a role in positioning the NPB territory (Moury and Jacobson, 1989; Garcia-Martinez et al., 1993; Ezin et al., 2009; Streit, 2025). Studies aimed at defining the NPB based on shared gene expression have found partially overlapping expression patterns in the medial-lateral and rostral-caudal directions, but none has uncovered a unique set of transcription factors that specifically mark the entire border region (Khudyakov and Bronner-Fraser, 2009; Groves and LaBonne, 2014).
An important question raised by this study is how observed spatiotemporal differences in expression of individual NPB and neural crest transcription factors is controlled. With respect to NPB factors, while the current study focuses on pax3 and zic1, other NPB factors examined also displayed qualitatively distinct spatial and temporal expression. During late gastrula stages, pax3 and zic1 expression domains exhibited gradient-like distributions, with a wide range of relative expression levels throughout different regions of the NPB. Additionally, we observed striking changes in pax3 and zic1 expression patterns over space and time, with distinct expression dynamics for pax3 in the surface ectoderm and zic1 along the rostral-caudal direction axis. GRN models describe fate specification at the NPB as a temporal hierarchy with NPB transcription factors feeding forward to inform downstream patterning of the neural crest and placodes, but our time course reveals continuously evolving expression differences in NPB gene expression between stages. Indeed, work in chick reinforced the idea that, although the NPB demarcates an anatomical region, it is not a discrete transcriptional state (Thiery et al., 2023).
Neural crest positioning within the NPB
A major question raised by the observed dynamic and non-uniform expression of NPB genes is how this informs the subsequent spatial patterning of the definitive neural crest domain. Experiments in Xenopus have emphasized the importance of pax3 and zic1 in the vertebrate NPB GRN for specifying neural crest. Sato et al. (2005) showed that co-expression of Pax3 and Zic1 in naive ectoderm is sufficient to activate a battery of neural crest specifiers, including snai2 (slug) and foxd3. Conversely, loss of function of Pax3 or Zic1 individually causes reduction of neural crest markers, and both factors are necessary, and in some contexts sufficient, for neural crest induction (Bae et al., 2014). There is also evidence that these transcription factors directly activate the expression of certain neural crest genes (Sato et al., 2005; Hong and Saint-Jeannet, 2007; Milet et al., 2013; Bae et al., 2014; Plouhinec et al., 2014). Based on this evidence, a co-activation model for neural crest induction has been proposed, whereby the NPB GRN acts solely to fine-tune expression levels of pax3 and zic1 in order to position neural crest gene expression within the ectoderm (Pla and Monsoro-Burq, 2018); however, such studies lacked quantitative measurements and relied on limited spatial resolution.
Intriguingly, homologs of pax3, zic1 and other NPB factors are co-expressed in domains analogous to the NPB in non-vertebrates, which lack neural crest. In Drosophila, the Pax homolog Paired and the Zic homolog Odd paired function in the pair-rule network to regulate the position and width of segment polarity genes including engrailed and wingless (Nusslein-Volhard and Wieschaus, 1980; DiNardo and O'Farrell, 1987; Benedyk et al., 1994; Xue et al., 2001). zic1 and pax3/7 are both expressed in planarian neoblasts, an adult pluripotent stem cell population, and are required for regeneration following injuries (Scimone et al., 2014; Vasquez-Doorman and Petersen, 2014). In tunicates, a cell lineage expressing msx, pax3/7, zicL, id and snail gives rise to cells that may be evolutionary precursors to neural crest cells (Abitua et al., 2012; Stolfi et al., 2015; Todorov et al., 2024). However, these cells in non-vertebrate chordates lack the migratory abilities and potential to form a broad set of derivatives including ectomesenchyme that is characteristic vertebrate neural crest cells. Our recent work provides evidence that such potential arose via a small number of evolutionary innovations to an ancestral neural progenitor GRN (Johnson et al., 2022; York et al., 2024).
Our current study provides the first quantification of how relative levels of NPB and neural crest gene expression vary spatially within the ectoderm in Xenopus. Notably, we did not find that a finely tuned ratio of pax3 and zic1 expression directly corresponds with the onset of expression of the neural crest genes we analyzed. Rather, we observed that individual neural crest genes are initially broadly expressed in non-uniform patterns and show differential correlations with levels of pax3 and zic1 expression. These findings suggest that other genes and signals functioning at the NPB are likely required to fully establish a neural crest state. In addition to pax3 and zic1, the NPB transcription factors msx1, hairy2 (hes4), tfap2, vent2 (ventx2.2), id3, snai1, among others, are expressed in NPB cells have been shown to activate expression of definitive neural crest markers (Luo et al., 2003; Hong and Saint-Jeannet, 2007; Nichane et al., 2008; Buitrago-Delgado et al., 2015). It also seems likely that broad, partially overlapping expression of early neural crest factors reinforce the expression of each other through cross-regulatory interactions, while feeding back to repress initial input from NPB factors. Our snai2 loss-of-function data lend support to this possibility, as we observed that loss of neural crest gene expression coincided with the spatial expansion of pax3 and zic1 expression.
Axial patterning of the presumptive neural crest
Some of the most notable differences in expression that we observed for neural crest genes occurred along the rostral-caudal axis. The temporal hierarchy in GRN models of ectodermal patterning suggests that NPB induction, neural crest specification and axial patterning of the neural crest happen in distinct steps (Simoes-Costa and Bronner, 2015; Martik and Bronner, 2021). Our results support previous findings that the domains of NPB factors show differences in the extent of their rostral-caudal expression (Streit, 2025), and we show for the first time that these differences in NPB gene expression overlap with differences in neural crest gene expression that persist through later migration stages. It will ultimately be important to confirm whether these differences are also observed at the protein level using a quantitative method such as hybridization chain reaction-immunofluorescence or single cell mass spectrometry (HCR-IF or scMS).
Interestingly, we found that the expression of NPB factors both influences and responds to neural crest specification. Whereas tfap2a becomes upregulated in the pre-migratory neural crest, pax3, zic1 and msx1 are each downregulated within specific regions of the neural crest domain. For example, expression of zic1 was downregulated in the anterior cranial neural crest, but it remained relatively unchanged in more posterior neural crest. Further experiments will be necessary to determine the mechanism by which zic1 is downregulated and whether differences in zic1 expression in anterior versus posterior cranial neural crest play a role in neural crest lineage segregation. Overall, our findings lay a foundation for re-thinking models of neural crest formation to incorporate when, where and at what amplitude gene-specific regulatory factors are expressed, in order to construct a more predictive and mechanistic neural crest GRN framework.
MATERIALS AND METHODS
Embryological methods
Wild-type X. laevis embryos were obtained using standard in vitro fertilization methods and cultured in 0.1× Marc's Modified Ringer solution (MMR). Fertilized embryos were de-jellied and stored in an incubator set to 13°C until the first cell division. Cleaved embryos were collected in 0.1× MMR and moved to a 19°C incubator to culture until the desired stages (Nieuwkoop and Faber, 1994). Embryos between stages 12 and 15 were collected at time points between 22.75 h post-fertilization (hpf) and 29.25 hpf, with a tolerance of 0.5 hpf to account for variability between embryo clutches. Clutches in which a majority of embryos did not reach stage 12 by 23.25 hpf were discarded. Embryos collected for in situ hybridization were fixed in MEMFA for up to 1 h, dehydrated in methanol overnight, and labeled using either the HCR-FISH technique described here, or a previously described chromogenic in situ hybridization protocol (LaBonne and Bronner-Fraser, 1998). Embryos injected with glucocorticoid-fused constructs (GR) were treated for 1 h in 10 μM dexamethasone diluted in 1× MMR while at stage 8. Animal cap explants were manually dissected at blastula stages and cultured in 1× MMR until collection alongside wild-type embryos from the same clutch for staging. All animal procedures were approved by Northwestern's Institutional Animal Care and Use Committee (IACUC).
HCR-FISH
The HCR-FISH methodologies were adapted from the ‘third generation’ protocol described by Choi et al. (2018). Fixed embryos dehydrated in methanol were re-hydrated with successive washes in 25% PBS+0.1% Tween 20 (PBS-T), 50% PBS-T, 75% PBS-T and 100% PBS-T. Embryos were bleached under light, washed twice in 0.1 M triethanolamine solution, and then treated with acetic anhydride diluted in 0.1 M triethanolamine. Following this treatment, embryos were washed in PBS-T and then successively incubated on ice in 50% PBS-T+50% 5× SSC-T, 100% SSC-T, and hybridization buffer. The embryos were incubated at 37°C for 30 min as a pre-hybridization step, and then incubated overnight at 37°C with 4-16 nM probes (dilution optimized for each probe set) in hybridization buffer. The following day, embryos were washed four times in 30% probe wash buffer at 37°C, and ‘pre-amplified’ at room temperature by incubating in amplification buffer for 30 min. Hairpins tagged with Alexa Fluor 488, Alexa Fluor 546 or Alexa Fluor 647 fluorophores were ‘snap cooled’ by heating to 95°C for 30 s and returned to room temperature for 30 min. Embryos were incubated in 60 nM hairpin solution diluted in amplification buffer overnight in the dark. The following day, embryos were washed in SSC-T, washed in PBS to remove the SSC-T, and incubated overnight in a refractive index matching solution (RIMS, 80% Histodenz) to optically clear the embryos for imaging.
Confocal microscopy
All fluorescent images were collected using a Nikon C2 laser-scanning confocal microscope and either a Plan Apo 4× (NA=0.2), Plan Apo λ 10× (NA=0.45) or Plan Apo λ 20× water immersion (NA=0.95) objective lens. Images at 4× magnification were collected with 1024×1024 resolution and 4× frame averaging, with 2.38 μm xy pixel size and 16.3 μm separation between z-slices. Images at 10× magnification were collected with 512×512 resolution and 2× frame averaging, with 1.89 μm xy pixel size and 2.9 μm separation between z-slices. Images at 20× magnification were collected with 512×512 resolution and 2× frame averaging, with 0.98 μm xy pixel size and 0.57-2 μm separation between z-slices. Laser power and gain settings were kept consistent between replicates sharing the same probe, hairpin and objective combinations.
Image quantification and analysis
Fiji/ImageJ was used to make maximum and mean intensity z-projections for figure displays, as well as to linearly scale the minimum and maximum color balance to uniform values for each probe within each figure. Whole-embryo images were cropped to a circle fitting the entire embryo and rotated so that the anterior poles of the embryos face upward. Lookup tables were inverted so that gray/black regions represent regions of higher measured intensities, and white regions represent regions of lower measured intensities.
Computational surface mapping was performed on selected 10× images using the Image Surface Analysis Environment (ImSAnE) described by Heemskerk and Streichan (2015). Each channel was blurred for surface detection using the ‘Planar Edge Detector’ class to detect the surface and represent the surface as a point cloud. Edge detection settings were optimized within each technical replicate. This point cloud was fit to a surface using the ‘Thin Plate Spline’ fitter and the fitted surface was pulled back to an xy stack. The same pullback settings were used for each image, with slices 2-14 used for further analysis of the ‘deep’ cell layer, and slices 19-31 used for further analysis of the ‘superficial’ cell layer. Distortion maps generated by ImSAnE were visually inspected to verify that there was minimal distortion within regions of interest in each image.
All further image analysis was performed in Python using Jupyter notebooks. To measure the technical noise due to non-specific probe binding, embryos were redundantly labeled using orthogonal probe sets targeting pax3 and imaged in the red (Alexa Fluor 546) and far red (Alexa Fluor 647) channels. The 10× images were mapped using the surface mapping procedure described above, and the deep cell layer was selected for analysis. The DAPI channel was blurred using a Gaussian filter with a kernel size of 5 pixels, and Otsu's threshold was calculated to create a masked image of the embryo. The mask was then eroded using a 9×9×9 cube in order to avoid highly distorted regions near the edge of the mapped image. Probe channels were slightly blurred using a Gaussian filter with a kernel size of 1 pixel, and the image was downscaled by taking the local mean of pixels within a 2×2×2 voxel region. Raw voxel intensities from one channel were re-scaled to have the same slope and intercept as the second channel, and the intrinsic noise was calculated within bins of 100 voxels. This intrinsic noise measurement followed the calculation described by Giri et al. (2020). Voxels within each bin were then resampled with replacement 1000 times in order to generate bootstrapped estimates of the noise due to non-specific probe binding presented in Fig. S2.
Line intensity profiles were measured in the deep cell layer of mapped images lines by manually clicking on a snai2 image to select points in the medial anterior/lateral anterior and medial posterior/lateral posterior regions of the ectoderm. Raw intensities were measured using a 51-pixel-wide line drawn between the click locations. Intensities were then normalized within each image to set the minimum intensity for each probe to 0, and to set the 99th percentile intensity to 1. Line profiles from different images were aligned by fitting a function of the shape , where x was given as the pixel location along the measured line. The x-location of the function's peak (i.e. the parameter estimated as ‘a’) was then defined as the midpoint for each image. This parameter fitting was performed separately for anterior and posterior line profiles. For each x-position along the aligned profiles, the mean and standard deviation of normalized intensities were calculated.
For the identification of transcriptional loci using intronic probes (Fig. 6), 20× images were collected within a region of the NPB using the green channel (Alexa Fluor 488) for the intronic probes. Auto-fluorescent background was subtracted from the intronic probe channel for each slice in a z-stack by applying a median filter using a disk with a kernel size of 3 pixels, and subtracting this blurred image from the raw image. The ‘peak_local_max’ function provided by the scikit-image package was used on the filtered image in order to identify transcriptional loci, and detected peaks were filtered for nuclear association by only selecting peaks colocalized with DAPI intensity above Otsu's threshold. The pax3 and zic1 channels were slightly blurred using a Gaussian filter with a kernel size of 1.5 pixels. The DAPI channel was heavily blurred with a kernel size of 15 pixels, and Otsu's threshold was applied to the blurred DAPI image in order to create a mask for regions of the image associated with the embryo. All channels were then downscaled by taking the local mean of pixels within a 16×16×8 (xyz) voxel region. Only voxels for which >95% of the original pixels were part of the embryo mask were used for further analysis. For downscaling the intronic probe images, voxels for which any pixels were associated with a detected peak were assigned a value of 1, while voxels for which no pixels were associated with a detected peak were assigned a value of 0. Voxel intensities for the pax3 and zic1 channels were scaled so that the 20th and 80th intensity percentiles ranged from 0 to 1 and subsequently binned within quintiles.
After binning these intensities, estimates of the percentage of voxels containing a detected transcriptional locus within each bin were made by bootstrapping. Voxels within each image were resampled with replacement 1000 times, and the percentage of voxels within each bin containing a detected transcriptional locus was recorded each time to generate a sample distribution. To generate a null distribution, the pax3 and zic1 channels were shuffled 1000 times and the percentage of neural crest-positive voxels within each bin was recorded. The shift in the null and sample distributions was then measured with the t-statistic from a two-sided t-test in order to assess the likelihood of more or fewer voxels containing neural crest transcription compared to random chance.
Animal cap voxel intensities were measured by blurring the DAPI channel with a median filter and applying Otsu's threshold to generate a mask for the animal cap, following the same steps described above. The channels containing HCR-FISH probe intensities were slightly blurred with a Gaussian filter and downscaled to approximately cellular-sized voxels. These voxel intensities were normalized by setting the minimum wild-type intensity to 0, and setting the 99th percentile wild-type intensity to 1.
To generate the predicted snai2 and sox8 distributions shown in (Fig. S6), representative 10× images of wild-type embryos double-labeled with pax3 and zic1 probes were surface mapped using the procedure described above. Intensities from both the deep and superficial cell layer were then downscaled and binned into quintiles. Voxels from the deep cell layer were then color coded using the same color map as the heat map in Fig. 6D.
Supplementary Material
10.1242/develop.205254_sup1Supplementary information
The reference list from the paper itself. Each links out to its DOI / PubMed record.
- 1Abitua, P. B., Wagner, E., Navarrete, I. A. and Levine, M. (2012). Identification of a rudimentary neural crest in a non-vertebrate chordate. Nature 492, 104-107. 10.1038/nature 1158923135395 PMC 4257486 · doi ↗ · pubmed ↗
- 2Bae, C.-J., Park, B.-Y., Ylee, Y.-H., Tobias, J. W., Hong, C.-S. and Saint-Jeannet, J.-P. (2014). Identification of Pax 3 and Zic 1 targets in the developing neural crest. Dev. Biol. 386, 473-483. 10.1016/j.ydbio.2013.12.01124360908 PMC 3933997 · doi ↗ · pubmed ↗
- 3Basch, M. L., Bronner-Fraser, M. and Garcia-Castro, M. I. (2006). Specification of the neural crest occurs during gastrulation and requires Pax 7. Nature 441, 218-222. 10.1038/nature 0468416688176 · doi ↗ · pubmed ↗
- 4Benedyk, M. J., Mullen, J. R. and Dinardo, S. (1994). odd-paired: a zinc finger pair-rule protein required for the timely activation of engrailed and wingless in Drosophila embryos. Genes Dev. 8, 105-117. 10.1101/gad.8.1.1058288124 · doi ↗ · pubmed ↗
- 5Buitrago-Delgado, E., Nordin, K., Rao, A., Geary, L. and La Bonne, C. (2015). Shared regulatory programs suggest retention of blastula-stage potential in neural crest cells. Science 348, 1332-1335. 10.1126/science.aaa 365525931449 PMC 4652794 · doi ↗ · pubmed ↗
- 6Buitrago-Delgado, E., Schock, E. N., Nordin, K. and La Bonne, C. (2018). A transition from Sox B 1 to Sox E transcription factors is essential for progression from pluripotent blastula cells to neural crest cells. Dev. Biol. 444, 50-61. 10.1016/j.ydbio.2018.08.00830144418 PMC 8022798 · doi ↗ · pubmed ↗
- 7Choi, H. M. T., Schwarzkopf, M., Fornace, M. E., Acharya, A., Artavanis, G., Stegmaier, J., Cunha, A. and Pierce, N. A. (2018). Third-generation in situ hybridization chain reaction: multiplexed, quantitative, sensitive, versatile, robust. Development 145, dev 165753. 10.1242/dev.16575329945988 PMC 6031405 · doi ↗ · pubmed ↗
- 8Dinardo, S. and O'farrell, P. H. (1987). Establishment and refinement of segmental pattern in the Drosophila embryo: spatial control of engrailed expression by pair-rule genes. Genes Dev. 1, 1212-1225. 10.1101/gad.1.10.12123123316 · doi ↗ · pubmed ↗