An adhesion code ensures robust pattern formation during tissue morphogenesis

Abstract

An outstanding question in embryo development is how spatial patterns are formed robustly. In the zebrafish spinal cord, neural progenitors form stereotypic patterns despite noisy morphogen signaling and large-scale cellular rearrangements during morphogenesis. By directly measuring adhesion forces and preferences of three types of endogenous neural progenitors, we provided a direct proof of the differential adhesion model, mediated by differential cadherin expression, assisting tissue patterning in vivo. The unique combinatorial expression of N-cadherin, Cadherin 11, and Protocadherin 19 in each cell type ensures homotypic preference ex vivo and patterning robustness in vivo. The cell fate and adhesion codes are regulated by the Sonic hedgehog gradient. We propose that robust patterning during tissue morphogenesis results from a previously unappreciated interplay between morphogen gradient and adhesion-based self-organization.

One Sentence Summary:

Pattern formation in the neural tube results from cellular self-organization based on a combinatorial adhesion code instructed by the morphogen Sonic hedgehog.

Spatial patterns of distinct cell types arise reproducibly in development. The classic French Flag model posits that a morphogen gradient forms across a naïve and static field of cells to provide positional information to specify patterned cell fates(1). However, tissue morphogenesis often requires significant cell proliferation, migration, and intercalation to change the overall tissue size and shape, leading to the disruption of existing positional information and scrambling of cell-cell neighbor relationships (2–6). Although the French Flag model has many important features, it does not explain how patterns can form reliably despite concurrent morphogenetic movements.

In the vertebrate spinal cord, neural progenitors interpret opposing gradients of Sonic hedgehog (Shh) and Bone Morphogen Protein (BMP), and commit to distinct fates to form stereotypic stripe-like patterns that are highly conserved across vertebrate species (7–9). The specification of different neural progenitor types depends on the dosage of the secreted morphogens, making the spinal cord a textbook example of the French Flag model (10–12). Through in toto imaging of zebrafish spinal cord development, we previously showed that the Shh signal is noisy, resulting in specification of neural progenitors in a mixed pattern at the onset of morphogenesis (13, 14). And the extensive cell rearrangements of convergent extension during morphogenesis are expected to further disrupt pattern. Nevertheless, the stereotypic stripe patterns still form reproducibly. This finding makes the zebrafish spinal cord an attractive system to study how robust patterning can be achieved despite imprecision in morphogen signaling and extensive cell-cell neighbor exchange during tissue morphogenesis and growth.

We first examined the generality of our previous observations that neural progenitors are specified concurrent to spinal cord morphogenesis in an imprecise fashion (13). We focused on three neural progenitor types (p3, pMN, and p0) as they can each be distinguished by the expression of a single transcription factor (nkx2.2a, olig2, and dbx1b) (Figure 1A) (15–19). At the neural tube stage, these three cell types form stereotypical stripe-like domains, as visualized by transgenic zebrafish carrying fluorescent reporters of nkx2.2a, olig2, and dbx1b, or fluorescent in situ hybridization based on Hybridization Chain Reaction (in situ HCR) (Figure 1B–D)(20–24). Through live imaging of the fluorescent reporters, we observed fate-specific marker expression of all three cell types during morphogenesis, in an imprecise pattern that mixes fate marker positive cells with fate marker negative cells (Figure S1, Movie S1–S4 and Text S1). Tracking of individual cells revealed that the mixed patterns are resolved during morphogenesis, predominantly through active cell sorting, to form cohesive stripe patterns at the neural tube stage (Figure S2).

Fig. 1. Neural progenitor cells exhibit homotypic preference.

(A) A cartoon illustration of the stripe patterns of p3, pMN, and p0 domains at the neural tube stage

(B) Cross section of the spinal cord of the triple fluorescent reporter fish, TgBAC(nkx2.2a:memGFP); TgBAC(olig2:dsRed); TgBAC(dbx1b:GFP), at 16 somite stage.

(C) Cross section of the spinal cord at 11 somite stage, stained with multiplex in situ HCR probes against nkx2.2a (green), olig2(red) and dbx1b(blue). The image is the maximal intensity projection over a 3.5 μm wide slab.

(D) Average intensity profile of nkx2.2a, olig2, and dbx1b along the ventral-to-dorsal axis of the spinal cord, visualized by either fluorescent reporters (dashed lines) or multiplex in situ HCR (solid lines). Error bar represent standard error of the mean.

(E) A cartoon illustration of the dual pipette aspiration assay. Scale bars of the inset in (E,G) are 10 μm.

(F) The adhesion force measured from 6 types of cell doublets. The boxes within the box plot represent 25^th, 50^th, and 75^th percentile of the data. * and ** represent p-values <0.05 and <0.01, respectively (t-test)

(G) A cartoon illustration of the triplet assay.

(H) Triplet assays to observe homotypic preference. Green, magenta, and blue represent p3, pMN and p0 cells. * and ** represent p-values <0.05 and <0.01, respectively (binomial test)

The cell sorting behavior suggests differences in adhesion properties among different neural progenitor types. Since the proposal of the Differential Adhesion Hypothesis, its feasibility has been demonstrated by controlled over-expression of adhesion molecules in cultured cells in vitro (25–27), while genetic evidence of errors in tissue organization following disruption of pre-existing pattern of adhesion molecules led to strong speculation that differential adhesion could operate in vivo (28, 29). However, direct biophysical evidence of endogenous cell types using differential adhesion to assist patterning in vivo is still lacking. Therefore, we set out to measure adhesion forces between different types of endogenous neural progenitors. We used the dual pipette aspiration assay to determine the separation forces of cell doublets, composed of neural progenitors isolated from the trunk region of transgenic embryos at the neural tube stage (Figure 1E and Movie S5) (30, 31). We measured the adhesion forces of six different combinations of cell doublets (Figure 1F), including three different homotypic contacts (contacts between cells of the same type) and three different heterotypic contacts (contacts between cells of different types). The average adhesion forces at the homotypic contacts between two pMN cells (7.6 ± 3.6 nN) and two p3 cells (4.0 ± 2.7 nN) are significantly greater than the pMN-p3 heterotypic contact (2.5 ± 2.2 nN). Similarly, the adhesion at the homotypic contacts between two pMN cells (7.6 ± 3.6 nN) and two p0 cells (7.2 ± 6.3 nN) are also greater than the pMN-p0 heterotypic contacts (4.4 ± 3.3 nN) (Figure 1F).

To enable more sensitive comparison of preferences between distinct contact types, we developed a triplet competition assay (Figure 1G). The triplet is composed of two cells of the same type and one cell of a different type, forming one homotypic contact and one heterotypic contact. This assay mimics the challenge faced by the cells when they are pulled by neighboring cells towards different directions in vivo. For each measurement, three isolated cells were assembled simultaneously into a triplet, allowed to form contacts for at least 3 minutes, and then pulled apart by stage-motor controlled micropipettes to observe whether the middle cell prefers the homotypic or heterotypic contact (Movie S6 and Methods). All three neural progenitor cell types showed clear preference for homotypic contacts, with pMN-pMN and p3-p3 homotypic contacts winning over pMN-p3 heterotypic contact, and pMN-pMN and p0-p0 homotypic contacts winning over pMN-p0 heterotypic contact by a ratio of approximately 2:1 (Figure 1H). Thus, each of the three neural progenitor types exhibit homotypic preference, a term we use to describe the phenomenon that cells selectively stabilize homotypic contacts over heterotypic contacts.

To identify the molecular mechanisms underlying this adhesion specificity, we obtained the transcriptomes of p3, pMN, and p0 cells, and used CRISPR-Cas9-mediated genome editing to knock out candidate adhesion molecules to look for patterning phenotypes (32). N-cadherin (cdh2), cadherin 11 (cdh11), and protocadherin 19 (pcdh19) stood out as genes with significant loss-of-function phenotypes (Figure S3 and Text S2). Therefore, we sought to determine the expression patterns of cdh2, cdh11 and pcdh19 during neural progenitor patterning. Our transcriptomic study showed that cdh2 is the most abundant cadherin in p3, pMN, and p0 cells (Table S1) (33). Using a fluorescent reporter, TgBAC(cdh2:cdh2-mCherry) (34), we found that Cdh2 exhibits a linear protein gradient along the V-D axis of the entire spinal cord that increases by 2-fold from the V-D position of 0 to the 0.8 position before decreasing towards the dorsal-most part of the neural tube (Figure 2A–C, S4A,B). Similar spatial profiles of Cdh2 are also observed by antibody staining of sectioned neural tube (Figure S4C,D) and in a different fluorescent reporter fish of cdh2, TgBAC(cdh2:cdh2-tFT) (data not shown) (35). The expression of cdh11 appears as one stripe along the entire spinal cord (Figure 2D) (36), largely overlapping with the olig2 positive domain but with a wider distribution along the V-D axis (Figure 2E,F). Pcdh19 is expressed as two stripes along the entire spinal cord (Figure 2G). Fluorescent in situ analysis showed that the ventral stripe of pcdh19 expression corresponds to the p3 domain and the medial floor plate (Figure 2H), while the dorsal stripe is positioned dorsal to the pMN domain (Figure 2I). The two pcdh19 expression domains flank the pMN domain and expression of pcdh19 and olig2 appear to be mutually exclusive (Figure 2I,J). The correlated expression between cdh11 and olig2, and between pcdh19 and nkx2.2a, and the mutual exclusivity between olig2 and pcdh19 were verified by our single cell co-expression analysis (Figure 2K–M and Methods). Together, quantitative analyses of cdh2, cdh11, and pcdh19 expression revealed a three-molecule adhesion code that is unique to each of the three cell types (Figure 2N). Importantly, the differential expression patterns of cdh2, cdh11, and pcdh19 are present at the onset of spinal cord morphogenesis, making these genes plausible regulators of cell sorting during the formation and maintenance of neural progenitor patterns (Figure S4E–H).

Fig. 2. Differential expression of cdh2, cdh11, and pcdh19 forms a unique combinatorial adhesion code for each cell type.

(A) Lateral view of a 10 somite stage embryo, showing the Cdh2 gradient in the neural tube of an transgenic embryo TgBAC(cdh2:cdh2-Cherry)

(B) Cross section of a confocal z-stack of a neural tube from a 10-somite stage transgenic embryo with TgBAC(cdh2:cdh2-Cherry); Tg(actb2:membrane-Citrine). Scale bar is 20 μm.

(C) Ventral-to-dorsal Intensity profile of cdh2-Cherry, normalized by the membrane-Citrine signal. Gray lines in the background are average intensity profiles from different embryos. The shaded regions of green, magenta, and blue in (C,F,J) represent the positions of p3, pMN, and p0 domains, and were identical to Figure 1D. Error bars are s.e.m.

(D, G) Whole mount in situ hybridization of cdh11 (D) and pcdh19 (G) at the neural tube stage (10 somite stage)

(E, H,I) Cross-section of 10-somite stage embryos with multiplex in situ HCR staining against (E) olig2 and cdh11, (H) nkx2.2a and pcdh19, and (I) olig2 and pcdh19. Images are maximal intensity projection over (E) 1.79, (H) 4.87, (I) 2.66 μm wide slices. Scale bars are 20 μm.

(F,J) Ventral-to-dorsal intensity profile of (F) olig2 and cdh11 or (J) nkx2.2a and pcdh19 based on multiplex in situ HCR staining. Error bars are s.e.m.

(K-M) Single cell co-expression pattern of olig2 -cdh11 (K), nkx2.2a – pcdh19 (L), and olig2 – pcdh19 (M) quantified from embryos stained with multiplex in situ HCR.

(N) A cartoon diagram showing the expression of cdh2, cdh11, and pcdh19 in the three neural progenitor types. The area of each dot represents the relative abundance of each adhesion molecule, quantified from the cdh2-cherry level in (C) and the in situ HCR signal for cdh11 (F) and pcdh19 (J).

We next sought to understand how the adhesion code mediates homotypic preference of the three neural progenitor types. Cdh11 is enriched in the pMN cells. Cdh11 belongs to type II cadherins, a family of cadherins that structural analyses suggested are unable to bind in trans with type I cadherins (e.g. Cdh2) (37). Cdh11-expressing cells and Cdh2-expressing cells segregate in culture, indicating that Cdh11 could mediate pMN-specific homotypic adhesion in a tissue dominated by Cdh2-based adhesion (38). Indeed, without Cdh11, pMN cells no longer exhibited homotypic preference against p3 or p0 cells (Figure 3A). Adhesion force measurements of cell doublets further showed that loss of Cdh11 specifically lowered pMN-pMN homotypic adhesion without affecting pMN-p3 or pMN-p0 heterotypic adhesion (Figure 3B).

Figure 3. The adhesion code mediates the homotypic preference ex vivo and patterning robustness in vivo.

(A) Results of the pMN-pMN-p3 or pMN-pMN-p0 triplet assay with and without cdh11.

(B) Adhesion forces between p3-pMN, pMN-pMN, p0-pMN doublets with and without cdh11.

(C) Results of the p3-p3-pMN triplet assay with and without pcdh19 or cdh2.

(D) Adhesion forces between p3-p3, p3-pMN doublets with and without pcdh19 or cdh2.

(E) Results of the p0-p0-pMN triplet assay with and without pcdh19 or cdh2.

(F) Adhesion forces between p0-p0, p0-pMN doublets with and without pcdh19 or cdh2.

(G) p3-pMN boundary at the neural tube stage, visualized by TgBAC(nkx2.2a:memGFP); TgBAC(olig2:dsRed) in a pcdh19 heterozygote or pcdh19 mutant background. Scale bars are 20 μm.

(H) Mixed cells at the p3-pMN boundary in pcdh19 mutant or morphants.

(I) pMN domain at the neural tube stage, visualized by TgBAC(olig2:GFP), in control or cdh11 morphant embryos. The cross-sections in the cdh11 morphants show the dorsal mislocalization of a pMN cell (white arrow) or a olig2:GFP negative cell mixed in the pMN domain (yellow arrow). Scale bars are 20 μm.

(J) Mixed cells in the pMN domain in cdh11 CRISPants or morphants.

(K) p0 domain at the neural tube stage, visualized by TgBAC(dbx1b:GFP) in wildtype, cdh2 mutant, or embryos injected with cdh2 morpholino in 1 cell of 8-cell stage embryos. Scalebars are 30 μm.

(L) Mixed cells in the p0 domain in cdh2 mutants or morphants (injected in 1 cell of 8-cell stage embryos).

(M) A cartoon diagram summarizing the working model for homotypic preference among p3, pMN, and p0 cells. The p3 cells utilize the Cdh2-Pcdh19 complex, the pMN cells utilize Cdh11 and Cdh2, and the p0 cells utilize Cdh2 and the Cdh2-Pcdh19 complex to achieve homotypic preference.

Next, pcdh19 is enriched in p3 cells. In vitro studies have shown that Pcdh19 can only generate very weak adhesion on its own (39). However, Pcdh19 can form a complex with Cdh2 to generate strong homophilic adhesion in trans using the extracellular domains of Pcdh19, and create a distinct adhesion mode that is mutually exclusive with Cdh2 alone (40). Since Cdh2 is expressed ubiquitously throughout the neural tube, the presence or absence of Pcdh19 in different cell types would thus generate two distinct modes of cell adhesion. Loss of either Pcdh19 or Cdh2 removed the preference for p3-p3 binding versus p3-pMN binding in our triplet assay (Figure 3C), consistent with the idea that a complex of Pcdh19 and Cdh2 is required for p3 homotypic preference. Interestingly, loss of pcdh19 does not change the p3-p3 homotypic adhesion force significantly; instead, the predominant effect of the loss of Pcdh19 is an increase of the p3-pMN heterotypic adhesion, to a level that is comparable with the p3-p3 homotypic adhesion (Figure 3D). The increase of p3-pMN adhesion without Pcdh19 is Cdh2-dependent, consistent with the role of Pcdh19 in preventing Cdh2 in p3 cells from binding in trans with Cdh2 in pMN cells, and therefore lowering p3-pMN heterotypic adhesion.

Finally, Pcdh19 and Cdh2 are enriched in p0 cells relative to pMN cells. As in the case of p3-pMN heterotypic adhesion, loss of Pcdh19 also leads to a Cdh2-dependent increase of pMN-p0 heterotypic adhesion. However, this does not entirely disrupt the homotypic preference of p0 cells (Figure 3E,F). Instead, loss of Cdh2 completely abolished the homotypic preference of p0 cells against pMN cells. We conclude that the homotypic preference of p0 cells against pMN cells is achieved primarily by higher levels of Cdh2-based adhesion, with a minor contribution by Pcdh19. To further investigate the role of the adhesion code in patterning in vivo, we examined the cohesion of pMN, p3, or p0 domains at the neural tube stage in the mutants or morphants of cdh11, pcdh19, or cdh2 (33, 39). We found a significant increase of patterning errors in each domain when the corresponding adhesion molecule is perturbed (Figure 3G–L, Movie S7, Text S3). Together, our results show that the adhesion code not only mediates the adhesion specificity underlying homotypic preference of neural progenitors ex vivo, but also mediates patterning robustness in vivo (Figure 3M).

To generate a cell-type-specific adhesion code, the expression of adhesion molecules and cell fate specification need to be closely coupled. We speculated that the adhesion molecules must share some common upstream regulators with the cell fate specification process. Shh is a key morphogen secreted from the ventral side of the spinal cord, and instructs specification of ventral neural progenitor fates (e.g., p3 and pMN cells) in a dose-dependent fashion (10, 11). We therefore perturbed Shh signaling and measured changes in expression patterns of both the cell fate markers for pMN and p3 cells (i.e., olig2 and nkx2.2a) and their corresponding adhesion molecules (i.e., cdh11 and pcdh19). We chose to down-regulate Shh signaling by treatment with cyclopamine, a chemical inhibitor of Shh activator Smoothened, and up-regulate Shh signaling by injection of shha mRNA into one-cell-stage embryos. Under cyclopamine treatment, the expression levels of olig2 and cdh11 were both significantly reduced (Fig. 4A,B, S5A,B, S6A). Conversely, when Shh is hyperactivated, both olig2 and cdh11 significantly expanded their expression domains (Fig. 4C,D, S5C,D, S6B). Interestingly, even under Shh down- and up-regulation, although the positions and levels of olig2 and cdh11 expression were both shifted, they still shared the same relative pattern with overlapping peak positions and a slightly broader domain for cdh11, and their expression remained correlated at the single cell level (Fig. S6C–F). This suggests that olig2 and cdh11 respond to Shh signaling in a similar but not identical fashion. For pcdh19 and nkx2.2a, cyclopamine treatment completely abolished nkx2.2a expression. Interestingly, pcdh19 expression switched from two stripes to one broader stripe centered near the original dorsal stripe, likely due to the disappearance of the ventral stripe (Figure 4E,F. S5E,F). When Shh is hyperactivated, both nkx2.2a and the ventral stripe of pcdh19 significantly increased the transcript level by 1.5–2 fold, and the dorsal stripe of pcdh19 expanded more dorsally than in control embryos (Figure 4G,H,S5G,H). Again, the relative patterns of nkx2.2a and pcdh19 are preserved through up- and down-regulation of Shh, suggesting shared regulatory logic between nkx2.2a expression and the ventral expression of pcdh19 by Shh (Fig. S6G–J). Finally, to further connect cell fate regulators downstream of Shh signaling with the adhesion code, we knocked down olig2, and observed the two stripes of pcdh19 expression merged into one continuous stripe with higher amplitude (Fig. 4I). This suggested the repression of pcdh19 expression in the pMN domain is olig2-dependent. Together, our data show that the combinatorial adhesion code in the ventral spinal cord is instructed by the Shh morphogen gradient (Fig. 4J).

Figure 4. The adhesion code is regulated by Shh signaling and cell fate regulator olig2.

(A-D) The ventral-to-dorsal in situ HCR profile of olig2 and cdh11 from embryos treated with (A) vehicle control or (B) 100 μM cyclopamine, or injected with 90pg of (C) H2B-BFP mRNA or (D) shha mRNA.

(E-H) The ventral-to-dorsal in situ HCR profile of nkx2.2a and pcdh19 from embryos treated with (E) vehicle control or (F) 100 μM cyclopamine, or injected with 90pg of H2B-EBFP2 mRNA (G) or shha mRNA (H).

(I) The ventral-to-dorsal in situ HCR profile of pcdh19 from embryos injected with control or olig2 morpholino.

(J) A cartoon diagram summarizing how the Shh gradient and olig2 regulates pcdh19 and cdh11

The origin of patterning robustness in tissues undergoing morphogenesis is an important open question. In the zebrafish spinal cord, we showed that robust patterning requires a previously unappreciated interplay between two classic ideas for patterning—morphogen gradients and differential adhesion. We speculate that this combination overcomes shortcomings of either model alone. By itself the morphogen model lacks the precision required to threshold sharp boundaries while the differential adhesion model may get trapped in local minima and does not specify position or orientation of pattern. In combination, a morphogen gradient can specify cells with different adhesive properties near each other in an initial rough pattern, and then differential adhesion can drive self-organization of cells to form sharp domains that stay organized throughout tissue morphogenesis.